This PDF file contains the front matter associated with SPIE Proceedings Volume 12596, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

The economic losses caused by the invasion of alien species from ballast water to China have reached 100 billion RMB per year. It is extremely important to equip ships with a fast and effective ballast water treatment system. Taking ship ballast water as the object of study, this paper introduces the common methods of ship ballast water treatment device and the existing problems. To solve these problems or shortcomings, a ship ballast water treatment device is designed by using software AutoCAD2017 from the design requirements of ship ballast water treatment device, and its working principle and main components are introduced. An automatic monitoring system with PLC as the lower computer and LabVIEW as the upper computer is designed. The results show that the device is high practicability and feasibility.

Reciprocating compressor is a kind of power machinery widely used in the industrial field, in order to achieve the condition monitoring of reciprocating compressor and fault diagnosis of key components, the reciprocating compressor vibration signal acquisition experimental platform is designed, and the mechanical performance of the compressor is monitored by means of vibration detection and signal analysis. In addition, a fault feature extraction method integrating time domain, frequency domain and entropy value is proposed, the fault feature extraction of the processed vibration signal is carried out, the extracted fault feature is used as input, and the compressor fault diagnosis is carried out by using the limit learning machine algorithm, and the results show that the method can better diagnose and identify the faults of different parts of the reciprocating compressor.

This paper is an essay about based on machine learning methods for patients with cancer probability prediction research, based on the existing machine learning just write and some of the cancer related information, dedicated to the study that can according to patients' basic living conditions, living habits, body external factors such as the age to the patient's cancer probability prediction, The aim is to allow patients to enter their own data at home to predict the incidence of cancer, so as to reduce the number of patients who come to hospital with advanced cancer. The machine learning methods adopted in this paper are mainly logistic regression and multiple linear regression, and the confounding matrix is used to verify the results, and finally the cancer-related information is obtained.

In view of the poor recognition effect of power equipment fault features in China, a method for building power equipment fault feature model based on unified semantic expression is proposed. The power equipment fault information is identified by combining the unified semantic expression principle. And the phase space reconstruction algorithm is constructed according to the feature semantics of the identified fault information. The power equipment fault feature model is optimized based on the reconstruction results. Finally, it is verified by experiments, the power equipment fault feature model based on unified semantic expression can quickly identify the semantic features of fault information in the process of practical application, and effectively improve the recognition effect.

The main problems faced by the repair of large motors are the difficulty of trouble finding, the unknown technical performance parameters of the repair materials, and the repair equipment and tools on-site cannot guarantee the implementation of the repair technology. Aimed the repair technology For the short circuit to ground fault of a large DC motor, a method of connecting bulb to find the fault of armature winding with equalizing voltage winding is introduced in the paper. Epoxy AB glue is used to varnish the motor in whole at room temperature, and ensuring the mechanical strengthen of coil conductor in groove of iron core, and commutating poles is circumvoluted by mica tape that is pressed by a special fixture to ensure the insulation performance of the poles. The test of applying 10 times of rated voltage to the main pole winding and measuring the voltage drop of each main pole winding is proved that repair method is feasible and can used to rehabilitate large DC motor on site.

Firstly, this paper introduces the importance of welding specialty in colleges and universities and the application direction of welding technology, then analyzes the problems existing in welding training courses in colleges and universities, and finally puts forward the application of virtual reality technology to change the current situation of welding training. Based on the in-depth study of virtual reality technology, the author decided to design and develop a welding training simulation platform based on Web3D. The overall design framework of the platform is B/S combined with MVC design pattern, and the realization of each functional module is based on ASP.NET technology. The virtual reality part uses Unity 3D and 3DMAX software to complete modeling and animation interaction. The construction of this platform aims at improving the practical teaching effect of welding specialty and cultivating students' welding technical ability on the basis of safety, environmental protection and cost saving.

In order to meet the automatic injection and cleaning functions of digital PCR detector, a double-layer injection needle and its liquid path device were designed. Firstly, the finite element analysis of the injection needle was carried out based on ANSYS Workbench. Combined with the theoretical calculation, it is found that the stability is poor and the bending stiffness is insufficient. Then, to solve this problem, a spiral spring was added between the inner needle and the outer tube to optimize the injection needle, and the influence of spring position, length and pitch on the deformation of the injection needle was explored. Finally, considering the strength, deformation and cleaning effect, it is determined that the best scheme is that the spring is 5mm from the outlet of the outer tube, the length is 5mm and the pitch is 1mm. At this time, the maximum deformation caused by 0.1N lateral force is reduced from 0.527mm to 0.158mm, the maximum stress is reduced from 151.5MPa to 86.25MPa and the critical buckling pressure is increased from 4.276N to 28.468N. After optimization, the strength, stiffness and stability of the injection needle are guaranteed. The research results are of great significance to the design and optimization of the injection needle.

Based on characteristics of plastic head cover, the paper carried out 5 design concepts to improve its NVH property. It’s useful to improve NVH properties of product if follow the concept when design or optimize a head cover. In this paper, finite element and multi-body dynamics analysis methods were used to analyse a plastic head cover firstly. At second the head cover was optimized with the concept in the process. Finally, the plastic head cover met the development target. Test results proved the reliability of design concepts as well.

Since it is difficult to accurately identify the minor fault of the early unbalanced exciting force fault of the linear vibrating screen, a fault diagnosis method based on the operating attitude of the screen box is proposed. Firstly, the dynamic analysis of the vibration system of the double axis linear shale shaker is carried out. Based on ADAMS environment, the dynamic simulation model of the double axis linear shale shaker is established, and six kinds of dynamic models of exciter failures are simulated to study the motion law of the screen box under the unbalanced excitation force failure. Further, the simulation analysis and field experiment of each fault dynamic model are carried out, and different fault data are trained and analyzed through the ELM neural network diagnosis algorithm. A set of attitude data acquisition system for the screen box of double axis linear vibrating screen is designed. The results show that the fault of the phase angle of the exciter can be diagnosed by detecting the screen box attitude.

At present, the common profile processing machine tools have a single function, and the use cost of the machining center series is high. In order to save costs and realize the multi-functional processing of profiles, a profile sawing and milling machine is designed. Mainly study the dynamic characteristics of the machine tool, suppress the vibration of the machine tool, and improve the machining accuracy. Through 3D modeling, ADAMS simulation, changing the thickness of the worktable, adding damping force and other methods, the natural frequency of the machine tool, the milling cutter mechanism and the amplitude of the worktable are obtained, and the structure of the machine tool is optimized. The results show that the resonance frequency of the self-excited vibration of the machine tool is mainly concentrated in the low frequency range of 0.1-100 Hz. The maximum frequency response of the machine tool is 18.19Hz with an amplitude of 39.84mm. The thickness of the worktable is increased by 20%, the maximum frequency response of the machine tool is shifted to the left by 0.41Hz, and the maximum amplitude of the worktable is reduced by 6.15mm. Adding the damping force, the maximum amplitude of the y+ and z- milling cutters is reduced by 3.96mm and 7.33mm respectively. It can be seen that the two optimal designs can effectively suppress the self-excited vibration of the machine tool, improve the machining accuracy, and make the machine tool design structure more completed and feasible.

In order to realize the lightweight design of the wing structure, the internal structure of the wing was taken as the research object. Topology optimization took structural compliance as the objective function, and size optimization took mass as the objective function. The variable density method based on the SIMP stiffness interpolation model was used to optimize the structural layout of the wing. On this basis, the finite element model (FEM)of shell-rod structure was established, and the feasible direction method (MFD) was used to optimize the wing structure size. The mass of structure is reduced by 28.7% compared with the initial structure, the optimal design scheme of the wing is obtained.

Aiming at the problem that the accuracy of conventional algorithms is low in the case of few samples for bearing vibration signal fault diagnosis, this paper proposes a bearing fault diagnosis method based on prototypical network in few-shot and zero-shot scenarios. The method first uses the original vibration signals or spectrogram features as input; then uses the neural network model to extract the distinguishable features, and prototype center of each category is learned through prototypical network; finally, the classification of each sample is completed by the distance measurement method. The experimental results show that prototypical network method with scaled CQT features as input and convolutional neural network as encoder has excellent performance in few-shot and zero-shot bearing fault diagnosis.

Power equipment failure prediction method has the problem of high cumulative deterioration, and a power equipment failure prediction method based on dynamic ontology modeling technology is designed to solve the above problem. It evaluates the health status of power equipment, clarifies the performance degradation range of equipment according to the characteristics reflected in different stages, constructs a residual life judgment model by combining the mechanism of reliability function, clarifies the performance degradation conditions and failure threshold of power equipment, and optimizes the fault prediction process by using dynamic ontology modeling technology. The test results showed that the mean values of cumulative degradation of the power equipment failure prediction method in the paper and three other power equipment failure prediction methods are 1.612, 3.263, 3.207, and 3.234, respectively, indicating that the power equipment failure prediction method designed after incorporating dynamic ontology modeling technique has higher use value.

Aiming at the problems of insufficient utilization of cooling capacity and low ice making efficiency in the ice making process of small household ice maker, a new type of water tank was designed by calculating and analyzing the cooling capacity in the ice making process of ice maker. Simulate the ice making process with software AnsysFluent, compare the ice production of ice maker with new and original water tank in the same time, and carry out experimental verification. The simulation result shows that 75.68% of the cooling capacity is wasted in the original water tank, and only 28.85% in the new water tank when ice making time is 9 mins. At the same time, the cold energy consumption in the new water tank is 43.34% lower than that of the original water tank, and the ice production is increased by 65.22%. The experiment result shows that the ice production of ice maker with the new water tank is increased by more than 75% when ice making time is between 8 and 10 mins; the ice production of ice maker with the new water tank is increased by more than 35% when ice making time is between 11 and 12 mins, and the optimization effect is obvious.

Wire and arc additive manufacturing is a technology to fabricate solid metal components by layer-by-layer surfacing using arc as energy carrier beam. Wire and Arc Additive Manufacturing has the advantages of high material utilization, high molding efficiency, low equipment cost and unlimited size of molded parts, etc., which has received wide attention from scholars in various countries. However, poor forming accuracy limits the development of this technology. In this paper, a quadratic regression model between weld pass size and welding voltage, welding current and welding speed was established by using the quadratic universal rotary assembly experiment method in single-layer single-pass Wire arc additive manufacturing test. The model could effectively predict weld pass size.

In the market, the thermoelectric device are mostly flat plate, and the solder overflow in the process of Bi₂Te₃ and copper welding is called wall-hanging phenomenon (WHP). This phenomenon mostly occurs in the production line and is inevitable, but it is seldom studied by scholars. In this paper, COMSOL software is used to simulate and analyze the wallhanging phenomenon, and the structure is optimized for this phenomenon. The simulation results show that when the load resistance is small, the wall-hanging phenomenon can improve the output power and conversion efficiency, which is beneficial, when the load resistance is large, the wall-hanging phenomenon will reduce the conversion efficiency, which is harmful. The smaller the size factor is, the less the impact of wall-hanging on the conversion efficiency of the thermoelectric parts. Therefore, when the load resistance value is large, the harm of wall-hanging can be reduced by reducing the size factor.

In the process of pumping and shore blowing of trailing suction dredger, the mud and sand movement mechanism of mud tank and sludge discharge pipeline is complex and strongly coupled. It is difficult to obtain the relationship between mud transportation concentration and pumping hatch, mud pump, high-pressure flushing, submarine diversion valve and pipeline through mechanism analysis. Aiming at this problem, this paper proposes a prediction method of instantaneous output of trailing suction dredger pumping and bank blowing based on BP neural network. Through the training of historical construction data, PSO and GA algorithms are used to optimize respectively, and the prediction model of instantaneous output of trailing suction dredger pumping and bank blowing is established. The simulation results show that this method can effectively predict the production of mud from the suction dredger.

The general purpose FPSO may encounter captain misjudgment, mooring system failure, direct alignment with the direction of the coming wave, and lead to a variety of wave directions, such as oblique wave, cross wave, and aft wave. The fluid analysis software FLUENT was used to simulate the general FPSO model, and seven wave directions of 0°, 30°, 60°, 90°, 120°, 150°, and 180° were simulated. The results show that the ship’s bow can withstand the extreme waves with the 100-year recurrence period in the Brazilian sea area successfully, but all the other conditions have different degrees of upsurge phenomenon. In addition, the bow board can resist not only the bow board up-wave phenomenon but also the bow board and cross wave, which makes the bow board up-wave degree lower. In the case of wave direction of 120°, 90°, and 180°, the threat degree of upsurge is very high, and it is necessary to further evaluate the capsizing risk. This conclusion has certain guiding significance for the design of ship-type FPSO in the following extreme sea conditions.

Gas disaster has always been a major safety problem in the coal mine field. The prediction of gas concentration in fully mechanized mining face is of great significance to ensure the safety of mine production and the safety of underground personnel. A Long short-term Memory (LSTM) neural network model based on time series is proposed for the prediction of gas concentration. Since there are many factors affecting the gas emission and there is a complex nonlinear relationship between them, a method of data preprocessing is proposed to weaken the data volatility, combined with the powerful GPU function of the computer, to build an LSTM neural network in the Tensorflow environment Gas Emission Prediction Model, using root mean square error (RMSE) and running time, for evaluating forecast performance. The prediction results are compared with the SVR network, and the results show that the LSTM model has higher prediction accuracy and prediction stability.

A multi-objective optimization method for electromagnetic actuator with permanent magnet is presented in this paper. The OLS-RBF neural network is improved by introducing gradient descent operator, and the coupling relationship between optimization objective and optimization factor of electromagnetic actuator with permanent magnet is fitted. NSGA-II algorithm is used to solve the multi-objective optimization of the approximate model obtained by fitting, and its effectiveness and feasibility are verified by simulation.

The geometric dimensions and manoeuvrability of a forklift in narrow aisles are important in industrial enterprises and warehouses. In this paper, the influence of the geometric dimensions of the AGV forklift and the transported load on the dimensions of the aisle width is studied. The difference between aisle widths and turning radii required for the movement of the forklifts is shown, which affects to the efficiency of industrial sectors and warehouses. At the same time, the turning radii of three and four-wheeled AGV forklifts were analyzed and represented mathematical expressions.

In view of the existence of a large number of conventional state monitoring devices in power transmission and transformation equipment, the use of state monitoring communication gateway is an effective method to solve the problem that conventional state monitoring devices are connected to the power transmission and transformation property management platform. Based on the analysis of Modbus and IEC61850 models, the model mapping method of Modbus and IEC61850 is proposed, including the mapping between IEC61850 object reference and MODBUS address, the mapping between public data type and MODBUS parameter type, and the mapping between abstract communication service and Modbus protocol data unit. The data model of Modbus is built by object-oriented technology. According to the principle of hierarchical semantic equivalence, the mapping relationship of information and service models among Modbus, IEC61850 and MMS is established. Based on the established mapping model, the design of Modbus and IEC 61850-8-1 conversion module and the reading and writing service of status monitoring data are completed, which verifies the correctness and feasibility of the above design method.

According to the characteristics of cabbage harvesting machinery, SOLIDWORKS and simulation software are used to carry out structural design and simulation analysis of its cutting device. The physical characteristics and planting status of Chinese cabbage and its rhizome were studied, and the cutting device meeting the actual harvest requirements was designed. Theoretical analysis was used to analyze the arrangement and force of the cutting knives, and a reasonable layout of the double disc knives was designed to ensure that the cabbage rhizomes could be effectively clamped and cut off, and the cutting qualification rate was improved. The virtual cutting motion simulation test of the disc rhizome cutter was carried out with ADAMS software. The movement of the single point on the knife is analyzed, and the movement law diagram of the displacement and trajectory of the single point of the knife is obtained. It is determined that the rotation of the blade can well cover the travel area to prevent missed cuts and reduce power loss. Carry out orthogonal test on the cutting device, determine the influence degree of each factor on the cutting force, and obtain the optimal parameter combination.

A novel closed Brayton cycle is proposed that integrates a supercritical CO₂ or CO₂/xenon mixture as working fluid with the ORC in this paper.. The parametric study is performed to evaluate the impacts of some key parameters on the performance of the combined system, including the pressure ratio in sCO₂, inlet temperature of turbine1 and evaporation temperature Results indicate that CO₂/xenon mixture can improve the thermodynamic performance of the combined system, which brings the highest exergy efficiency of 58.80%. At the same time, CO₂/xenon mixture presents lower efficiency and power.

Based on the stable operation of the intelligent production line for molten salt electrolysis of rare earth oxides, Flink technology is used to monitor the information of equipment, materials and relevant operators in the electrolysis process online, centralize processing and analysis, alert abnormal information in time, use big data analysis technology to make the production elements controllable and in an optimal state, solve the problems of unstable product quality, unstable power consumption and unstable material ratio in the refining of rare earth metals (alloys) that exist in the manual operation of the whole industry, and realize the centralization, digitalization, remoteness and intelligence of rare earth metal smelting.

In this paper, a Fractional-Order Global Sliding Mode Control (FOGSMC) scheme based on a neural network with approximation property (NNO) is mainly focused on study the Thermal-Structural Test (TST) system. Since the nonlinear dynamic system of the thermal-structure test with quartz lamp is susceptible to external interference and parameter variation, a novel FOGSMC system is designed based on improved fractional order global terminal sliding surface to acquire the desired trajectory, and real time estimation of system disturbance using neural network observer with Gaussian Function, meanwhile, the fractional-order global terminal sliding mode surface based on fractional-order function can effectively weaken the chattering phenomenon of the integer order, simulation studies show the effectiveness of the proposed method.

With the great success of airworthiness in ensuring the safe operation of civil aviation products, military aircraft products have begun to gradually introduce the concept and method of airworthiness. Under the premise of focusing on the function and performance of military aircraft, the safety of products is pursued to the maximum extent. However, in the actual process, due to the huge difference between military products and civil products, in the process of promoting airworthiness, military aircraft products often have an unbalanced contradiction between airworthiness verification and performance verification. Based on this, this paper proposes a dual fusion scheme and process based on airworthiness and performance verification, designs the complex diversification mapping method of airworthiness verification and performance verification test subjects in detail, and describes the verification method of airworthiness clauses in performance verification process. The scheme and method have been applied in actual model tasks and achieved good results.

The outlet cross-sectional area of standard nozzle affects the air mass flow value of TPS. Through the jet simulation of three standard nozzles with different outlet sizes under different boundary conditions, the relationship between outlet diameter and boundary conditions and jet turbulence intensity and near-field and far-field noise is studied. The simulation results show that the outlet cross-sectional area affects the value and range of the turbulence intensity, and has a great change in the mixing region of the turbulent boundary layer. However, under the excitation of the sound source quadrupole, it has no great impact on the propagation trend and extreme value of the sound wave.

Deepfake open source technology has lowered the threshold for AI face swapping to a very low level, making it possible to swap faces with one click. The cost of "disinformation" is greatly reduced, so that some deeply faked pictures and videos can be spread on social networks The social network can spread explosively. However, in the defense layer, there are almost no standardized and automated detection tools for deepfake. There is no such tool. Therefore, whether for individuals or platforms, the time window for fighting fake and disinformation is very short, but it is very difficult. In this paper, we use the Transformer model as a base, improve the model and optimize the structure of the model, so that the model can extract the depth features of the video and build a more accurate and efficient deepfake inspection method.

Space application facilities have complex systems and high operation requirements, making induction maintenance applications difficult and time-consuming based on augmented reality technology. To solve this problem, this paper puts forward the agile manufacturing technology framework and key technologies of induced maintenance application combined with the product characteristics of space application facilities and verifies the application effect of this technology in the actual scene through practical cases. The realization of the agile production technology of induced maintenance effectively improves the efficiency of application production. It provides a new auxiliary solution for the astronauts to solve emergencies and sudden missions in orbit.

Lung cancer is one of the most serious cancers, which has high death rate. In much research, researchers find that preventing lung cancer more effective for people to against it. In this paper, we aim to predict the possibility of lung cancer for the test individuals and exploit the main factors. We apply three machine learning models, including linear regression. Polynomial regression and bootstrap for this task. In the experiment. We find the linear regression achieves the best performance, with the lowest MSE (0.11). Furthermore, we find that the age, smoke and alcohol take important role in lung cancer. The author provides a comprehensive prediction and analysis for lung cancer precaution.

With respect to the 16 characteristics of the workers, the objective of this study is to investigate how employee turnover can be classified using various machine learning algorithms (Support Vector Classification, Decision Tree Classifier, AdaBoost Classifier, Random Forest Classifier, Extra Trees Classifier, Logistic Regression and Gradient Boosting Classifiers). The information comes from the Employee Turnover dataset by E. Babushkin. Seven distinct classification models were developed and contrasted, including naive Bayes, random forest, logistic regression, support vector machines, and XGBoost. Numerous experiments validate the effectiveness of machine learning model. Among all the models, we find that the random forest model achieves the best results, which can be furtherly utilized in real-world prediction.

According to statistics from the World Health Organization, Colorectal Cancer (CRC) is the third most commonly diagnosed cancer in the world. The detection of CRC in an early stage is crucial for on-time and proper treatment, which may significantly increase the patient's survival rate. Although computers are not qualified to replace human experts at the moment, having a referential result from CRC auto-detection and saving the time of manual diagnosis is still very meaningful. This paper compares the performances of two different neural networks classifying CRC based on a set of histology images. The labeled dataset is publicly available on the Tensorflow website, and the two neural networks are tested on the same dataset separately. The first type of neural network in this study is Convolutional Neural Network (CNN), and the second type is a Deep Neural Network (DNN). As the dataset splits into training, testing, and validation sets, the loss, accuracy, and training time are recorded by the end of each epoch. The study result shows that the CNN method is better than the DNN method in terms of CRC image classification. It takes a long time but has better performance.

Breast cancer is common in women, ranking first in the incidence of cancer in women and occupying first place in the mortality rate of cancer in women. Because of the seriousness of breast cancer, researchers and institutions worldwide are making unremitting efforts to find the perfect diagnostic and therapeutic solutions. The increasing maturity of image processing technology has led to the growing use of computer-based pathological diagnosis in diagnosing various diseases, and researchers have done much research on this. This paper presents some studies on breast cancer histopathological images based on hematoxylin-eosin staining. Currently, the diagnosis of breast cancer is based on hematoxylin-eosinstained histopathological images. First, the surgeon will take a piece of tissue from the patient's lesion and make a histological section. Next, the pathologist will observe the histological section and diagnose the results. In this way of diagnosis, the patient's diagnosis depends more on the subjective judgment of the pathologist, which requires a high degree of professionalism and is not very efficient. Therefore, for hematoxylin-eosin-stained breast cancer histopathology images, there is a need for a computer-assisted automatic diagnosis method that can reduce the pathologist's burden and make the patient's diagnosis objective and efficient with the help of image processing technology. To this end, this paper compares the performance of three standard machine learning algorithms for comparing hematoxylin-eosin-stained breast cancer histopathology images.

One of the most important aspects of Computer General Force (CGF) is to understand what the opponent is doing and predict their possible future actions. In this paper, we propose an extended event graph named Probability Event Graph (PEG) to predict the opponent’s future events. Compared with the basic event graph model, the element of event node, logical node, causal edge and time window is redefines in PEG. Through these novel elements, PEG can describe the event and causal relationship about the system comprehensively. The PEG model is the fundamental of forecast analysis. Firstly, the behaviour characteristics of opponents are analysed and the corresponding PEG model is established according to domain knowledge. Then, the parameters are acquired by training data generated by simulation. Finally, the reasoning algorithm based on PEG model is proposed, and the possibility and principal analysis are carried out.

As one of the major diseases, cancer has always been a hidden danger to human health. There has been a considerably improvement in the therapy for patients all over the world, as research and technology advance, medical care becomes more effective. In this regard, the cure rate and survival probability have increased positively compared with the last century. However, the incidence rate of cancer has not been effectively controlled, and lung cancer and breast cancer are still more common. Predicting the probability that a cancer patient will survive at their initial appointment is extremely important according to this report. In this case, doctors can not only have a more detailed understanding of the situation of patients, but also make the allocation of medical resources more reasonable; Secondly, it can also promote the improvement of medical treatment in cancer. This article will first import the relevant data sets and analyze the variables contained. Then, the next step will use logistic regression analysis and linear regression analysis to predict the survival probability of patients. Furthermore, completed the judgement which variable has a greater impact by comparing the data that affect this probability. By comparing the accuracy of these regression analysis, the accuracy of logical regression (93.14%) is higher than that of linear regression (77.12%). In this case, logistic regression analysis will be more applicable. Finally, this paper compares the influence of related variables. According to the findings, a patient's probability of survival is determined by the amount of lymph nodes inside the system.

One of the most common malignancies worldwide is breast cancer. Early screening and diagnosis are important to the reduction of mortality rates of patients. In order to improve the performance and accuracy of breast cancer image screening, researchers have made significant progress in Computer-aided diagnosis (CAD) systems built on convolutional neural networks (CNN). In this research, several recent CNN models of breast cancer diagnosis are discussed and explained, and multiple public datasets of breast cancer images are introduced. The detailed performances of the models are presented and compared. The limitations and potential improvements of current CNN-based CAD are discussed. Convolution neural network-based CAD are still facing challenges of shortage of public dataset and the problem of implementation in the clinical scenario. Conclusively, using a convolutional neural network to diagnose breast cancer is still at its early stage, and further developments are required to apply convolutional neural network-based cancer diagnosis to clinical practices.

This paper is to predict the presence of recurrence for breast cancer patients by citing data. As a first step we will collect relevant data on breast cancer patients from the internet. Next, we will use decision trees in Scikit-learn to determine if there will be a recurrence of breast cancer in patients who have been cured. Through a series of calculations and predictions, the accuracy of our experimental model finally reaches 0.75 accuracy. These data can help us to accomplish our target prediction well.

With the rapid development of information technology and the transformation of manufacturing services, the overhaul activities of complex equipment are receiving more and more attention. Locomotives, as an important part of the rail transportation industry, are typical complex technical equipment with a long-life cycle and require multiple overhaul work during the life cycle to ensure safe operation. At the same time, because of the complex structure of locomotive products, high operational reliability requirements, large downtime losses, and many levels of maintenance, simple maintenance processes are not applicable. The Intelligent Transportation System (ITS) can monitor and detect the traffic information parameters in real time, so the ITS is used to troubleshoot locomotives and enable them to operate safely. It can be seen from this study that the application of intelligent transportation system in Locomotive Maintenance (LM) can be used as locomotive maintenance engine and monitor the maintenance process to keep the locomotive running well.

Diabetes is a very serious worldwide chronic disease that affects people's life and health. Patients require insulin injections to maintain blood sugar balance exogenously. Methods to detect diabetes are time-consuming and labor-intensive. With the popularity of machine learning algorithms, we expect to predict and analyze diabetes through deep learning methods. In this paper, we utilize machine learning methods for data analysis and prediction. Our method was tested on public datasets and found that the random forest algorithm performed best, and that BMI and gender were the most important factors affecting the prevalence of diabetes.

The dense small objects detection is a challenging task in the scenario of UAV aerial surveillance. This paper proposes an improved YOLOv5 detection method for the dense small objects in high resolution images. To augment the dataset, a 20% overlap crop is used for the UAV aerial photography training set. In order to detect the tiny objects in the aerial photos of UAV, a tiny detection head is added on the basis of YOLOv5. The SPP and CBAM modules are introduced in the head of the model, SPP for feature fusion at different scales and CBAM for adding attention to spatial and channel dimensions. Multiple experiments are conducted on the VisDrone 2019 dataset, the results show that the mAP of 12 classes detected by the model is 30.4%, and 3.1% higher than the original YOLOv5.

A brain tumor can negatively affect basic bodily functions and when malignant can result in low survival rates. Many studies were conducted to detect and classify brain tumors in MRI images using a convolutional neural network (CNN) and other techniques like image preprocessing and transfer learning. However, few studies have explored the effect of specific hyperparameters on the performance of such CNNs. This study aims to investigate how the input size affects the CNN’s accuracy in brain tumor detection. Brain MRI datasets were collected and split into training, validation, and test sets. Four models with identical architectures but different input sizes of 256px×256px×3, 224px×224px×3, 128px×128px×3, and 64px×64px×3 were built using TensorFlow Keras, trained on the training set with data augmentation, and evaluated using the test sets. Of these four models, the one with 64px as input size has the best performance, yielding the highest test accuracy, 99.16%, and lowest test loss, 0.0282, whereas the 224px model has the worst performance, with the lowest accuracy, 98.06%, and highest loss, 0.0976. Accordingly, it appears that larger input sizes do not necessarily result in higher accuracy of the CNN performing brain tumor detection. Future studies on this topic may consider using a smaller input size, not only maintaining high accuracy but also significantly reducing the required time to train and the space to save the model.

In stations, airports and other places, contraband detection faces many problems such as false positives, omissions and slow detection speed caused by object background interference and human factors. This paper proposes an improved network based on YOLO-lightweight. The attention mechanism module is embedded in the backbone network, focusing on the important features from different channels. CBAM-FPN (Convolution Block Attention Module and Feature Pyramid Networks) structure is adopted in the network neck to reduce the loss of network features. Attention mechanism module is added in the bottom-up feature fusion process. Finally, CIOU is used as the edge optimization loss function to accelerate the network convergence and optimize the network model. Compared with YOLOv4-tiny, the precision is improved by 3.8%, reaching 87.5%. The detection speed reaches 60.3fps. The improved network only occupies 23.4M memory, which is convenient for embedding mobile devices. The improved network meets the real-time detection requirements.

Aiming at some key test positions of the installed shielding vehicle that do not meet the test distance, the paper analyzes the prominent problems that are difficult or impossible to measure during the shielding effectiveness test, and analyzes the factors that cause the problem. The hole-cavity coupling theory is used to establish the theoretical correction factor models of different frequency bands in the installed state are established, and the correctness of the proposed method is verified through theoretical calculations and actual case tests, and the theoretical correction factor of the installed state is used to correct the shielding effectiveness of the known empty vehicle, to obtain the application shielding performance of key positions where the installed shielding vehicle does not meet the test distance, avoiding the testing of the installed shielding vehicle, thus greatly improving the ability to obtain the shielding performance of the installed shielding vehicle.

The traditional valve control system is prone to pressure shock which is harmful to the efficiency and reliability of the hydraulic system. In view of this, simulation model of valve control system has been established based on AMESim to analyze the characteristics of pressure shock. Through designing a set of load-sensitive controlling system instead of valve control system, the self-adaptive control strategy has been implemented in the system. The results show that the self-adaptive control system can effectively improve the pressure shock for multiple users working in different condition.

The OSNR monitoring based on machine learning has achieved some results in coherent optical communication system, but it is not widely researched in intensity-modulation and direct detection system. In this paper, an electrical domain signal processing scheme based on deep neural network is proposed for monitoring link OSNR of intensity-modulation and direct detection system. We successfully estimate the OSNR of the 4GBaud OOK signal with the mean absolute error less than 0.81dB in the range of eight to 18 dB by a five layers deep neural network using 550,000 datasets.

There is a serious coking risk in 660MW boiler mixed with high sodium coal in a power plant, In order to effectively prevent and control the boiler coking problem and ensure the safe and stable operation of the unit, the optimization and adjustment of pulverizing system was studied, the experimental research was carried out from the aspects of loading force optimization of coal mill, cold and hot primary air leveling, wind-powder ratio and separator optimization, according to the test results, optimization and adjustment are made. By tuning of the coal pulverizing system, improve the uniformity of four root powder tube carrying powder, to eliminate the vibration problem of the initial, reduce the effect on the thermal load in the early years of the start, avoid the problem of heating surface ultra gentle coked, ensures the safety of the equipment, and gives the gasification of different coal pulverizers guidelines export air temperature.

In order to study the hydrodynamic performance of an aquaculture fish cage, the hydrodynamic model of the cage was established based on the potential flow theory and Morrison equation, and the motion response amplitude operator (RAO) of the cage was analyzed. Based on the quasi-static method, the full chain and three-segment mooring systems are designed respectively. The motion response of the cage under the combined action of wind, wave, and current was calculated and the two mooring schemes were compared. This study shows that the three-segment mooring scheme can effectively reduce the maximum tension of the mooring line and reduce the weight of the mooring line under 50-year extreme environmental loads.

To ensure the reliable operation and efficient maintenance of the air conditioning ventilation system of Fuxing EMU and reduce the cost of the whole life cycle, RAMS was designed for the ventilation system in the conceptual design stage. According to the structural composition and technical characteristics of the air conditioning ventilation system, the risk analysis model of the air conditioning ventilation system is established, and the design of the safety, reliability, maintainability, and availability of the air conditioning system is completed one by one. This design result can provide strong data support for the optimization design and maintenance plan revision of the air conditioning and ventilation system of the EMU.

Side pole impact is a common form of road traffic accident. Due to the narrow side energy absorption space, the development of the vehicle side restraint system has put forward strict requirements. However, the reappearance accuracy of side pole impact sled in the industry is always at a low level, and enterprises can only choose expensive real vehicle tests or CAE methods for development. In this paper, a multi-cylinder intrusive side pole sled impact test method is proposed. Through the coupling and decoupling of the pulses between the main sled and the side impact sled, the interaction process between the door and the dummy in the real vehicle impact test is well simulated, and the V-type intrusion is realized. The data after the test is analyzed in detail. The chest ribs and abdominal ribs of the dummy show good consistency, and the deviation can be controlled within 15%, which greatly reduces the development cost of the side pole impact. At the same time, there are also deviations in the compression amount of the upper rib and the force of the pubic symphysis force, which can be continuously optimized in the follow-up work. This research provides a new idea and solution for the development of the side restraint system.

Hard target penetration weapon is a trump card weapon for attacking multi-layered structural targets such as underground fortifications and command posts. The key to achieve the efficient damage to targets is the precise layer counting of fuse penetration process. Considering the oscillation aliasing and identifying difficulty in the traditional overload layer counting, a method of penetration layer counting based on the geomagnetic characteristic signal is proposed. During the warhead penetration of the reinforced concrete multilayer structure such as multi-story buildings, due to the additional magnetic field generated by magnetized ferromagnets in the geomagnetic field, the magnetic induction intensity measured by the sensor will change with its position related to the floors, so the layer can be counted by detecting the layer penetrating signal. The multi-layer target plate and warhead models are established, and the principle of penetration process, signal characteristics and layer counting strategy under the multi-parameter condition are analyzed through the COMSOL, a finite element analysis software. The purpose of this paper is to break through the large error and identifying difficulty of characteristic signals in the traditional layer counting method with accelerometer, and to provide theoretical reference for the design and damage efficiency improvement of penetration weapons.

Aiming at the current research status and shortcomings of the mechanical characteristic condition monitoring system of high voltage circuit breaker, a condition monitoring system of high voltage circuit breaker based on DSP is developed in order to find the operation state of circuit breaker in time and improve the reliability. The system uses TMS320C28346 as the control core and AD7606 as the sampling chip, which can realize the collection, conversion, transmission and display of monitoring data, and can monitor the closing and opening coil current and contact stroke of the circuit breaker in real time. Through experimental tests, the results show that the system meets the design requirements of each parameter error below 2%.

In recent years, domestic commercial vehicles have developed rapidly, and many series of products have been developed from light, medium to heavy. However, the vibration and noise problems of the cab of commercial vehicles during driving still fall far short of the requirements expected by customers. In order to improve the comfort of commercial vehicle cab during driving, the mounting system and vibration isolation and noise reduction system of commercial vehicle cab are designed. The cab mounting structure is an active electromagnetic mounting structure. Through the blind source separation algorithm built in the cab mounting ECU, the impact of vibration on the cab during driving can be quickly eliminated. The vibration isolation and noise reduction system adopts double-layer sound insulating glass and ANC active noise reduction system to minimize the noise transmitted to the cab. Through the self-designed active electromagnetic mounting structure and the vibration isolation and noise reduction system composed of double-layer sound insulating glass and ANC active noise reduction system, the comfort of commercial vehicles during driving is improved, which meets the expectations of customers.

SAR systems have played a huge role in ocean monitoring. However, the fast and reliable interpretation of SAR images is still a challenge. SAR image simulations of sea surface for different states facilitates a deeper understanding of the intrinsic scattering mechanism in SAR images. In this paper, based on an improved semi-deterministic facet method, SAR images of different marine environments are simulated by fast calculation of the sea surface scattering field. This approach is no longer sensitive to changes in both cutoff scale and surface element size and is suitable for SAR imaging of large size sea surface. The simulation results reasonably show the effects of the ocean scattering mechanism on the SAR images, which is helpful for the image analysis and interpretation.

In order to further enhance the active power regulation capacity of China power grid, this paper analyzes the necessity of constructing accident reserve capacity of power grid under the new power system from the aspects of receiving terminal characteristics, new energy development and load fluctuation. In order to further improve the active power regulation ability of my country's power grid, the necessity of constructing the emergency reserve capacity allocation of my country's power grid under the construction of a new power system is analyzed from the aspects of receiving terminal characteristics, new energy development and load fluctuation.

The lighting control system in the market still has many problems, such as complex wiring, insufficient intelligence of the operating system, weak anti-interference ability and so on. This design uses wireless communication technology to solve the above problems. Firstly, a wireless communication network based on ZigBee module is constructed. Collect environmental information through various sensors and upload it to ZigBee terminal for intelligent logic judgment. The terminal can upload the received sensor data to the network to realize remote monitoring. In addition, the design of lowpower single live switch of RCC switching power supply is divided into open power supply and closed power supply. This design scheme can effectively take power and control the load. The controller adopts polysilicon solar charging scheme to effectively supply power to the main circuit.

The reliable operation of wind turbines is the foundation of efficient application of wind energy, and it can promote the development of new energy. Drive system is the key link to realize wind power generation, which affects the power generation efficiency, economic benefits and safe operation of wind farms. This paper introduces the components of the wind turbine drive system, analyzes the common failures of its components, analyzes the causes of the failures, and puts forward some suggestions to improve the reliability of the wind turbine drive system, which can provide a reference for wind farm operators.

In this paper, the capacity planning and operation optimization model for integrated energy system is established. Taking the lowest economic cost of integrated energy system in the whole life cycle as the optimization goal and considering the constraints of material and energy balance, the optimal hourly scheduling optimization of equipment power in typical weeks is realized; Based on the operation scheduling, the classical particle swarm optimization algorithm is selected as the method to solve the double-layer model, the results are analyzed, and the expectation of the lowest economic cost is returned to the upper genetic algorithm to find the optimal capacity allocation, so as to realize the two-stage integrated stochastic optimization strategy.

In this paper, a research was conducted to analyse and predict the impacts of COVID-19 on public transportation ridership in the U.S. and 5 most populous cities of the U.S. (New York City, Los Angeles, Chicago, Houston, Philadelphia). The paper aims to exploit the correlation between COVID-19 and public transportation ridership in the U.S. and make the reasonable prediction by machine learning models, including ARIMA and Prophet, to help the local governments improve the rationality of their policy implementation. After correlation analyses, high level of significant and negative correlations between monthly growth rate of COVID-19 infections and monthly growth rate of public transportation ridership are decidedly validated in the total U.S., and New York City, Los Angeles, Chicago, Philadelphia, except Houston. To analyse the errors of Houston, we consult the literature and made a discussion of Influencing factors. We find that the level of public transportation in quantity and utilization is terribly low in Houston. In addition, the factors, such as the lack of planning law and estimation of urban expressways, the high level of citizens’ dependence on private cars and pride of owning cars play a considerable roll in the errors. And the impacts can be predicted to a certain extent through two forecasting models (ARIMA and Prophet), although the precision of our models is not enough to make a precise forecast due to the limitations of model tuning and model design. According to the comparison of the two models, ARIMA models' forecasting accuracy is between 6% and 10%, and Prophet's forecasting accuracy is between 8%-12%, depending on the city. Since the insufficient stationarity, periodicity, seasonality of time series, the Prophet models are hard be more refined.

The current UAV inspection multi-link data congestion control method based on link capacity uses a cache queue model to regulate the data throughput at the sending end, which leads to low control performance due to the lack of monitoring of data sending nodes. In this regard, the transmission line UAV inspection multi-link data congestion control method is proposed. The state of the UAV network data nodes is sensed using an ant colony algorithm, data scheduling flows are selected according to the bandwidth load, and data congestion is alleviated through data allocation as well as route maintenance. In the experiments, the control performance of the proposed control method is verified. The analysis of the experimental results shows that the proposed method is used to construct a multi-link data congestion control technique with a low data congestion rate and its control performance is high.

When the robot grasps the U-shaped snap on the automatic production line, the pose detection and the positioning of the gripping point of the snap should be solved. To solve this problem, we propose the improved algorithm of YOLOv5, which can obtain the rotation angle and gripping point coordinates of the U-shaped snap. Firstly, the training sample angle information is obtained by roLabellmg. Secondly, in order to obtain the predicted angle, the algorithm adds a new angle prediction dimension and replaces the original positive box IOU with the minimum external rectangle IOU of the rotating box containing the angle information when calculating the IOU. Finally, the gripping point coordinates are determined on different poses of the U-shaped snap according to the robotic gripping rules, respectively. On the homemade U-shaped snap data set, the mAP value reaches 91.2%, which proves the effectiveness of the proposed method.

Existing lane line detection algorithm for identification of a straight line in good condition, to solve the problem of curve, however, failed to find a good strategy, especially in large curvature of curve, the visual field to extract the lane line produces by the two become one, resulting in a wrong calculation, in the case of real vehicle test, bend by camera height, visual field, etc. The indoor robot car is used as the carrier for the test, and a turning strategy is proposed to recognize the lane line at the corner, and the lane line detection algorithm based on sliding window is improved to make it less affected by the environment. The algorithm is simple and efficient, which is suitable for the indoor robot car visual line inspection. The experimental results show that the lane detection algorithm proposed in this paper improves the passing rate and stability of the robot car under large area rate curves.

Skin cancer is a common illness that claims thousands of lives annually in the United States alone. Accurately identifying malignant tumors is crucial to survival but can be challenging as the visual distinctions between benign and life-threatening tumors are minimal. The purpose of this project is to explore deep learning algorithms that can be trained to systematically classify skin cancer images, create a program to execute the algorithm, and exemplify an optimization process for the program that can serve as a reference for future works. The project will adopt a transfer learning algorithm based on previous studies on the subject and select a pre-trained model given practical restraints. Then a program will be coded in Python to retrieve datasets, process images, train the model, and evaluate accuracy. Finally, the algorithm will be optimized by tuning model parameters and training restraints. The experiments revealed that the algorithm was able to perform the task with an accuracy of around 70%. Model parameters such as optimizer choice and learning rate and training restraints such as batch size and epoch count have significant impacts on the training results and require precise values for maxima accuracy and minimal overfitting.

Aiming at the problems of insufficient labeled samples and high missed detection rate in common textured surface anomaly detection, the paper designs a self-supervised learning model based on masked Autoencoder, which can realize accurate detection and location of anomalies without providing mass anomaly samples. Autoencoder is widely used, but it is difficult to detect and locate anomalies by reconstruction error due to its strong generalization ability reconstructed anomalies with small errors. Then, masked reconstruction method is proposed to reduce the generalization performance. First, each input image is masked to obtain multiple masked input images which are sequentially reconstruct by the Autoencoder. Second, these reconstructed images are complementarily masked and recombined to obtain the final reconstructed image. Finally, anomaly detection and localization are achieved by evaluating the reconstruction error between the input and reconstructed image. The experiment results indicate that the anomaly detection rate of this method is 95.09 % and the anomaly location rate is 93.32% under the anomaly detection standard metric，and the performance can be significantly improved.

In the process of practical application, constrained by the UAV's own performance and other conditions, some transmission line UAV inspection data low latency back transmission method has the defect of high probability of code element error. In this context, a new method of low latency return transmission line UAV inspection data is designed. Construct the transmission line inspection task allocation model, derive the mathematical expression formula of the distance interval between the stationing point and the target tower in the area to be inspected, describe the probability of priority allocation in the inspection data transmission process, detect the effective bandwidth of the UAV channel, decompose the time delay between the UAV transmission end and the ground receiving end, and design the data low-latency back transmission method. Test result: The mean value of code element error probability of the designed transmission line UAV inspection data low latency return method is 2.214%, which has a higher performance advantage compared to the other two transmission line UAV inspection data low latency return methods.

In recent years, UAV networking technology has become a hot topic, and it has been used in industry, agriculture, emergency, fire protection, communications and other fields. Communication technology is the key technology of UAVs, which has a decisive impact on the development of UAVs. The 5G cellular network also brings more possibilities to UAVs. With the rapid development of 5G cellular-connected drones, the security issues of their sensor networks have also attracted widespread attention. This paper studies the key management technology in the 5G network-connected UAV sensor network. Although public key encryption schemes such as RSA and elliptic curves can provide sufficient security, their applications are limited due to their extremely high demands on computing power, which conflict with the resource constraints of sensor nodes. For the security management of cluster keys, this paper introduces the concept of bivariate polynomials into key management. This not only ensures pairwise keys between any two nodes, but also performs intercluster key distribution on the generated bivariate polynomials, tests various characteristics of the generated bivariate polynomials in key management, and Simulation experiments are carried out under different orders of polynomials.

This paper introduces and analyses the method of applying Neuroscience methods to Boltzmann Machine, involving a combination of cognitive psychology, information theory, and dynamical systems. We utilized the emergent property of the probability of hidden layers to find the pattern of how units are behaving when stimulated by the visual layer and research into enhancing the predictive encoding capability of the encoding layer. We measure the connections and links between the units of the encoding layer by approximating it with the probability distribution of two units' activation behaviours. For example, the portion of the Auditory cortex responsible for processing auditory information, such as music, differs from the sections responsible for processing visual information, although they can still be linked and active concurrently. Besides, Neurons can modify their connections by learning new information and reinforcing the connections that have been utilized more frequently, and forgetting the connections if the probability distributions of two units diverge much. The Boltzmann machine is the probabilistic inference machine for ground truth using the free energy principle. The latter has stepped further from the concept to interpret cortical responses as a fundamental of intelligent agency. With simple and random interactions of each neuron, this 'intelligent agency' could achieve sophisticated functions in a specific area of a brain. Randomness is also a vital aspect of learning since it may achieve balance and embrace regularities according to Ramsey's Theory.

Pear recognition is one of the key technologies of pear picking robot, and the pear recognition algorithm based on convolutional neural network has high computing cost and large parameters, which is difficult to be deployed on pear picking robot with low computer resources. This paper presents a lightweight pear real-time detection method based on YOLOv5. This method designs a lightweight feature extraction network based on Ghost bottom-leneck, and embeds the SE module into the designed network, which improves the ability of feature extraction while reducing the amount of network parameters. The experimental results show that compared with YOLOv5l, the parameters of the improved lightweight model are reduced by 48.17 %, mAP is increased by 0.9 %, and the recognition speed is increased by 36 %. The improved model is more suitable to be deployed on the picking robot with limited computing power and provides a solution for the vision system of pear picking robot.

Implicit discourse relation classification, identifying relationships between arguments without explicit linguistic cues, is a challenging task. Previous studies have shown that connectives are important for recognizing implicit discourse relations. Most previous works applied connective prediction as an auxiliary task to promote knowledge transfer from connectives to labels which did not make full use of the relational mapping information of connectives. In this work, we propose an innovative Connective-aware Interactive Attention (CAIA) joint learning approach. Specifically, we use BERT to predict connectives and incorporate connective information into the interaction of the attention mechanism. Our experimental results on the PDTB dataset show that our approach achieves competitive results compared to recent state-of-the-art systems.

Statistical learning methods require large-scale data to make the significant generalized probability and observation error close to each other. Few-shot learning can alleviate this situation, but it cannot break through the limitations of statistical learning methods. The training model only depends on the correlation between data distributions. There may be potential risks in applying these models to decision-making in the natural environment. This paper studies feature selection in small sample regression analysis based on AutoMPG and MOP The performance of the two datasets on the regression task is first verified through three classical regression analysis models. Then, through the causal inference method, this paper analyzes the causal effect of the relationship between the features in the dataset and finds that two groups of features do not have a causal relationship. Finally, by setting up a simulation environment, this paper illustrates the potential risks of not considering the causal effect in feature selection.

The experimental measurement of the strength of low-alloy steel is very cumbersome, but it is also essential to knowledge its strength. In this study, two machine learning methods, random forest (RF) and support vector machine (SVM), were used to study the strength of low-alloy steels on the existing data samples of low-alloy steels, so as to make relevant predictions on their strengths and find the most influential factors. Comparing the measured results with the predicted values shows that RF outperform SVM in predicting results. And by calculating the correlation coefficient, the two features that have the greatest influence on the strength are the temperature and the content of V, respectively. This result can be used to optimize the properties of low-alloys.

The repeated failure of a plug-in capacitor solder joint occurs in an equipment in the process of long-term use, which has certain effect on the function and performance of the equipment. This paper analyzes the cause of the failure, by SEM、 EDS and anatomy of the solder joint, it is found that the legs of the capacitor are oxidized, resulting in cracks during soldering. By oxide layer removing and tining, it can effectively improve wettability and reduce the risk of defects, achieve the purpose of improving product quality.

Light and small UAVs have attracted widespread attention due to their good adaptability, low cost, and high temporal resolution. With the continuous development of technology, multi-UAV cooperation has become a research hot spot. The route planning problem of multi-UAV cooperation can be decomposed into two sub-problems: task allocation and route planning. In this paper, a task allocation method based on reinforcement learning is proposed for multi-UAV cooperation. Considering the task requirements, the capabilities of the UAV, the influence of the environment and the conflict of the task, we construct a MDP process include the state space, action space, reward function and discount factor with the constraints and optimization functions. In this paper, the task allocation process is combined with the trajectory planning based on maximizing information throughput, and a large number of simulation tests are carried out to verify the stability of the method.

The conventional single carrier frequency antenna pattern has a narrow main lobe level, but its side lobe level cannot be greatly reduced, which is not conducive to the improvement of the anti-interference effect. Based on this, this paper applies the sinusoidal weighted multi-carrier frequency scheme to the Frequency Diverse Array (FDA) and compares the performance of the antenna patterns of several FDA structures using the sinusoidal weighted multi-carrier scheme. It can be seen from the simulation results that compared with the single carrier frequency and other multi-carrier frequency schemes, the performance of the FDA regime radar such as SL-FDA is significantly improved after applying the sinusoidal weighted multi-carrier frequency scheme. The main lobe width in the range dimension is narrowed, and the side lobe level is also effectively suppressed. The sinusoidal weighted multi-carrier scheme is significantly better than other multi-carrier schemes.

In order to exclude differences in the perception of vibration by different testers and to help the relevant authorities to plan maintenance and determine the timing of line repairs, the vehicle jitter threshold was analysed to make it one of the predictors of sustained vehicle jitter and to exclude to a certain extent the interference of the testers' physical sensations. According to the acceleration signal collected by the test, spectrum analysis and ride and comfort index calculation. When it is close to the threshold and the trend continues to increase, i.e. ride index is greater than 1.6-1.8, comfort index is greater than 0.7-0.9, lateral main frequency 6-9Hz, vertical main frequency 6-9Hz and 12-15Hz, the vehicle and line should be checked in advance and targeted management, if close to the threshold but the trend of change is gentle, the means of governance can be temporarily not taken, but need to strengthen the monitoring, collection of various types of monitoring indicators exceed the threshold and there is a continuous trend of increase, it is recommended that the vehicles and lines are inspected.

Visual ship image object detection has essential applications for near-shore ship management and military object location. In recent years, object detection technology based on a deep learning algorithm has been widely studied in object detection of visible ship images, and achieved outstanding results. However, due to the difference and overlap of nearshore ship objects, the object loss rate is high. Aiming at the above problems, this paper proposes an improved RetinaNet ship object detection algorithm. Firstly, channel attention is added after the residual network, and used to enhance the attention to low-frequency information. Secondly, the cyclical focal loss and the CIOU loss function are used to increase the training times of negative samples in the middle of training, which effectively improves object detection accuracy. The experimental results show that the improved RetinaNet algorithm improves the recognition accuracy of ship objects by 2.5%.

In view of the low efficiency and high cost of the current line patrol method, as well as the cumbersome technology and weak operability of the helicopter power patrol inspection, this paper describes the UAV system in detail. At the same time, combined with the application of UAV in the line operation and maintenance management, it introduces the process of UAV patrol inspection in detail, and focuses on the path planning, line fault detection and line evaluation and prediction in the transmission line patrol inspection. It is concluded that UAV inspection can effectively improve the efficiency of inspection and maintenance of transmission lines, and promote the process of intelligent operation and maintenance of transmission lines.

Logo recognition technology can be used to identify the authenticity of logos, and logo substitution technology can be used to add watermarks to images, print anti-counterfeiting, effect generation, image composition, and even document signing. It can facilitate specific people and protect the rights of the author. This paper is a study of Logo recognition and substitution based on the SIFT algorithm, using the SIFT description to recognize the presence or absence of a pre-stored Logo in the Logo Library. The logo is replaced by a logo from the logo library.

To combat the threat posed by "low-small-slow" UAVs, this paper examines the attack modes of UAVs, focusing on single attack and swarm attacks. Based on the performance of the existing mainstream anti-UAV system, the type of UAV attack, and the features of defense, "triangular joint defense" tactics are presented for the usual scene of key point defense. Through UAV single machine and swarm deduction of critical point attacks, it evaluates the defense potential of an anti-UAV system at various attack distances. It designs an emergency response mode to deal with the challenge of short countermeasure time swarm close-range attacks in mind. This research serves as a technical reference for the deployment of key point defense power and the development of anti-UAV measures.

To solve the dangerous behavior of passengers, this paper studies a light warning device for escalators. First, the dangerous behaviors of passengers are graded, and then the dangerous behaviors of passengers are photographed and identified through the binocular cameras. Finally, the passengers are discouraged from the dangerous behaviors by the light warning. Through this device, the dangerous behaviors of passengers on escalator can be detected, prevented and dealt with as early as possible, and the occurrence of safety accident can be fundamentally eliminated.

The handrail is an important part of the escalator, which moves synchronously with the steps to ensure the safety of passengers. It is one of the common failures of escalator equipment that the handrail is out of synchronization with the step operation. If the handrail is not synchronized with the step operation or even stops running, passengers will fall down. In this paper, a failure case of the driving chain of the escalator handrail is analyzed macroscopically and microscopically. The research results show that the serious wear of the pin shaft and sleeve of the escalator handrail drive chain is the main reason for the handrail drive chain.

The weight reduction of the frame longitudinal beam directly affects the weight reduction of the frame. The intelligent algorithm is applied to the lightweight problem of the frame, and aimed to the area of the longitudinal beam section is optimized. The genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm and ant colony optimization are compared. The static working condition of the optimized frame is analyzed. The results show that all four algorithms can get the solution satisfying the constraint conditions, and the particle swarm optimization algorithm is the fastest, the simulated annealing algorithm is the slowest, and the other two algorithms are moderate. All four algorithms reduced the weight of the frame, the ant colony optimization reduced by 4.1%, and the other three ways reduced by 6.8%.

With the continuous improvement of radar performance, it is very difficult to break through the enemy's radar defense net by relying solely on trajectory planning. In actual penetration, fighters use a variety of electronic jamming methods to achieve penetration. Based on the dynamic changes of RCS, this paper model the discovery probability of penetration fighters under active jamming and passive jamming respectively, and studies the factors that affect the penetration probability and survival rate of fighters under different jamming situations. By comparing the radar detection probability with and without jamming, it is concluded that the fighter's active and passive jamming can effectively reduce the radar detection probability. This article has certain guiding significance for the fighter's dynamic penetration.

This paper analyzes a feasible spacecraft flight plan that uses gravitation assistance to transport the spacecraft from Earth to the circular orbit around Saturn (the spacecraft is in a circular orbit around the Earth, with an orbital period of 90 minutes and a total mass of 5000 kg, including fuel) by establishing a low thrust transfer orbit model and calculates the minimum amount of fuel required, which is 1878.73kg. There is also an attempt to evaluate different options for controlling the ion thrusters during the journey, and one of the schemes inspired by the Cassini Huygens spacecraft is proposed and considered optimal. Adopting this plan, the total journey time is calculated to be 14.2 years.

Font generation is a challenging problem. To address the existing problems of poor font style conversion models, which have missing structure, blurred glyphs and require paired datasets, this paper proposes a Chinese font style migration algorithm based on the improved CycleGan. The model introduces deformable convolution in the encoder part of the generator, which can learn the font features adaptively. A skip connection module, which fuses global and local features, was added to the model, and the features in the encoder are projected to the decoder using this module to avoid the structural error problem by reducing the information loss of the decoder. Meanwhile, using the attention mechanism, we can quickly and efficiently obtain the key information of the target region. On this basis, we can further complete the local and global feature fusion. According to the research results, this method can better achieve font generation in practice, so it has high application value.

The seed word-driven approach based on weakly supervised text classification (WTC) is the dominant approach. In existing seed word-driven methods，using metrics such as Term Frequency (TF), Inverse Document Frequency (IDF) and its combinations to update the seed words. the method assigns the same weight to all metrics, leading to the selection of common or poorly differentiated words as seed words; In addition most of the text classifiers used in the study have difficulty in capturing the correlation and global information between text information. In order to solve the above problems, Using Transformer as a text classifier first, The multi-headed self-attention mechanism allows capturing longrange dependencies while computing in parallel and fully learning the global semantic information of the input text. Then an improved TF-IDF method is proposed to increase the weight of IDF so that some common words that affect the classification can be filtered out. Its experimental results are improved on 20News and NYT datasets.

Aspect-based sentiment analysis is crucial for Internet applications such as social networks and e-commerce, where the previous deep learning methods cannot process long-range semantic information in parallel. This paper proposes an aspectbased sentiment analysis method based on multiscale convolution and a double-layer attention mechanism. The technique uses pre-trained BERT to obtain the hidden semantic information of the context from the training set, then uses multiscale deep convolution and double-layer attention to process the long-distance semantic information between the target word and the context in parallel, and finally uses softmax for sentiment classification of the target word. In this paper, we use the public dataset of SemEval 2014 and the Twitter Dataset to validated the improved accuracy and F1 of the model.

Aiming at the problems of low precision, slow speed and the detection problems when the text lines are arranged in any direction in the traditional natural scene text detection method, based on the target detection algorithm YOLOv5, a rotated text detection method with angle classification is proposed——YOLOv5-R, by defining the representation of the rotating rectangle, calculating the method of rotating the IoU, and designing a new loss function to achieve accurate detection of horizontal and oblique text, and tested the effectiveness on the scene text datasets ICDAR2013 and ICDAR2015, after the transformation The algorithm realizes the function of rotating target detection, but there is still some room for improvement in arbitrary shape detection.