This PDF file contains the front matter associated with SPIE Proceedings Volume 13269, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Iron ore concentrate is highly prone to liquefaction during sea transport by its own physical properties and external loads. This would cause serious property losses and casualties. Therefore, an in-depth study on the liquefaction mechanism of iron ore concentrate is needed. The fundamental reason for liquefaction is that the microscopic structure of iron ore concentrates changes, and its ability to resist external loads is continuously weakened. To investigate the structure characteristics of iron ore concentrate before and after liquefaction, an indoor six-degree-of-freedom shaking table test and CT scanning are used to conduct a detailed study. The shaking table test simulates the dynamic loads experienced during sea transport, while CT scanning provides detailed images of the internal structure of the iron ore concentrate. Then Avizo software is used to reconstruct pore structure and the parameters of iron ore concentrate before and after liquefaction are analyzed. The results show that the porosity of iron ore concentrate decreases, and the number of small pores increases. Pore water migration conditions will be somewhat improved. However, it is difficult for the pore water to drain completely from the cargo. The liquefaction of iron ore concentrate becomes a mixture of solid-liquid phases. The surface of the cargo appears fluidized.

In recent years, with the rapid development of hyperspectral remote sensing technology, hyperspectral remote sensing data has witnessed progressive utilization in the subway transportation industry. Due to the large amount of redundant information in hyperspectral images and the high correlation between band information, traditional classification algorithms often find it challenging to obtain more accurate classification information. Convolutional neural networks in the field of subway transportation have achieved great success in hyperspectral remote sensing image classification. These networks operate quickly and offer high classification accuracy, thereby greatly advancing hyperspectral remote sensing image classification technology and benefiting the subway industry. This article analyzes and compares the classification accuracy of two hyperspectral remote sensing image classification methods based on three-dimensional convolutional neural networks, capitalizing on the Pavia University and Indian Pines datasets. It specifically examines the classification of transportation infrastructure, particularly subway lines, within these datasets. The two models differ in data input sizes and convolutional layer parameters. By varying the sample size and epochs for analysis, we assess the results using the Kappa coefficient and accuracy to evaluate the models’ classification performance. The findings indicate that increasing the data input size and period can effectively enhance the classification accuracy of subway lines and other land features. Notably, increasing the data input size leads to a more significant improvement, while the effect of increasing the number of epochs has a certain threshold beyond which improvements diminish.

To tackle the formidable challenges that adverse weather conditions pose for image object detection, this paper presents an innovative approach grounded in the Image Adaptive YOLO (IA-YOLO) framework. The framework has introduced a series of advanced strategies to address the challenges to the accuracy and reliability of object recognition under extreme weather conditions. In environments where visibility is reduced due to factors like rain, fog, or low light, traditional object detection methods often struggle to achieve satisfactory results. However, IA-YOLO aims to overcome these limitations by incorporating adaptive image enhancement techniques that can effectively improve the quality of captured images. By embedding a unique image refinement mechanism within an efficient convolutional neural network, IA-YOLO empowers the system to autonomously acquire superior parameters for image refinement through a minimally supervised learning approach. This approach ensures that the images are enhanced in a way that is specifically tailored to improve object detection performance. To encapsulate, the article presents IA-YOLO as an influential instrument for tackling the obstacles posed by inclement weather in the realm of image object detection. By leveraging adaptive image enhancement techniques, IA-YOLO aims to provide more accurate and reliable detections, even in the most challenging weather scenarios.

To enhance the quality and contrast of color polarization images, this paper proposes a novel color polarization image fusion algorithm based on wavelet transform (WTCPI). The method begins by preprocessing four images captured at different polarization angles to generate a polarization intensity (I) image and a degree of linear polarization (DoLP) image using the Stokes vector method. Wavelet decomposition is then applied to these images to obtain their high-frequency and low-frequency components. The low-frequency components are fused using a weighted average method, while the high-frequency components are fused by taking the maximum of their absolute values. Finally, the fused images are reconstructed. Experimental results demonstrate that the proposed algorithm significantly improves subjective visual comfort and outperforms existing methods across all objective evaluation metrics.

As the demand for intelligent and autonomous welding becomes more and more extensive, in order to realise the degree of autonomy of welding robotic arm for welding, thus the research of weld seam positioning technology is carried out. The weld positioning system is built by combining the line structured light vision sensor and the robotic arm, and the Gaussian filtering method, Otsu threshold segmentation method and image morphology closed-circuit algorithm are carried out sequentially by acquiring the image of the weld seam light bar, solving the interference caused by the high reflective noise of the welding piece to the narrow weld seam, and obtaining the complete weld seam characteristic topographic information. Determine the scanning speed of 6mm/s, realize the global scanning of the curved weld seam of the non-regular weldment, obtain the pixel coordinates of the light bar image of the weld seam in each frame, assign the value of 0 to the Z coordinate, obtain the linear model of RANSAC, obtain the linear model with the highest number of collected points for feature point extraction, and judge the current frame image, obtain the weld seam centreline position point set and remove the weld seam reconstruction of the workbench. For the extracted non-regular curve weld centreline, NURBS curve interpolation and control of two-point gap straight line interpolation are combined to obtain a smooth and dense complex weld centreline. From the experimental analysis, the error range between the extracted weld centreline and the actual position point in each axis direction is 0.2~0.7mm, which meets the welding positioning accuracy requirements.

Traditional bone and meat segmentation relies on operators' experience and manual operation, resulting in low efficiency and high cost. Introducing three-dimensional information to assist in bone and meat segmentation can help improve the efficiency of bone removal work and save costs. The three-dimensional information comes from the bone contours in the X-ray image sequence. In this paper, an improved YOLOv5 algorithm and point cloud reconstruction method are used to achieve accurate segmentation of the femoral contour. The improved algorithm incorporates Conv convolution and CA coordinate attention mechanism in the neck, enhancing the local feature extraction ability of the algorithm; At the same time, C3 was replaced with SENetV2 module in the backbone, which improved the precision of feature expression. Using motion recovery structure algorithm to reconstruct sparse point clouds from segmented contours. Compared to the original model, the segmentation accuracy mAP@0.5 and mAP@0.5 ~0.95 has been improved, laying a good foundation for subsequent point cloud processing and cutting path planning

This paper presents a comprehensive review of image calibration and distortion correction techniques based on internal threads, focusing on their principles, methods, applications, and challenges. Internal threads serve as unique calibration markers due to their invariability and high detectability, providing essential parameters for image distortion correction. The review highlights the effectiveness of these techniques in enhancing accuracy and reliability in various high-precision applications, such as industrial inspection and medical imaging. It also addresses the computational complexity, sensitivity to initial parameters, and the need for high-quality images. The paper aims to offer valuable insights and references for future research and development in image processing technologies, promoting their broader application and advancement.

At present, infrared image can show good detection effect in complex environment because of its unique imaging characteristics, which brings a lot of convenience to people's production and life. However, because the infrared image contains a great lot of unnecessary information, it has great requirements for computing power and memory capacity, so it is challenging to satisfy the demands of real-time detection on resource-constrained portable mobile devices. To solve this problem, this paper proposes a refined lightweight algorithm built upon the YOLOv5, which changes the backbone network of the model to GhostNet network, introduces CA attention mechanism into the model to improve the detection effect of small targets, and changes the loss function of the model to Focal-EIoU loss function to improve the location effect of the boundary box and quicken the model's rate of convergence. Then replace the conventional convolution used in the sampling process of the model with the space-to-depth convolution module, in order to enhance the model's detection performance on low-resolution images and enable the extraction of more feature information. According to the experimental results, the enhanced detection algorithm is just 1% less than the initial YOLOv5 algorithm, and the model volume and floating point calculation are reduced by 31.9% and 36% respectively. This is better suited for deployments that are lightweight.

It may achieve auto segmentation to construct 2 D histogram utilizing 2 D entropy. Habitual approach practices center gray scale and average scale of points in the same 4 neighbor. Actually, the horizontal or vertical change of the gray scale will not stand for the whole natures among neighbor points also. A maximum value way is constructed to established 2 D histogram by utilizing of center gray value and maximum value of 4 other point gray values except 4 neighbor points in the same 3×3 neighbor. The effects of multi-level experiments to almost all images in Matlab image toolbox manifest the validity of the constructed way.

Lane line segmentation technology is crucial for extracting traffic parameters at urban road intersections. Currently, manually delineating intersection lane lines can no longer meet the demands of constructing digital traffic systems. Due to significant environmental interference and the diverse distribution of lane lines in the perspective of intersection video surveillance, existing lane line segmentation algorithms are ineffective for intersection scenarios. To address this issue, we propose an improved intersection lane line segmentation model based on YOLOv8n-seg. Firstly, inspired by the architecture of the Vision Transformer (ViT), we optimize the C2f module and incorporate attention mechanisms to enhance the network's ability to integrate features and capture global information. Secondly, we replace strided convolutions with a Haar wavelet downsampling module, which preserves all feature map information during downsampling. The experimental results show that mAP@0.5 and mAP@0.5:0.95 of the improved YOLOv8n-seg in segmentation are 5.7% and 1.5 % higher than the original model, and the advantages of low parameters and high FPS are maintained, which can effectively realize segmenting intersection lane lines.

In the context of power construction, image caption generation leverages deep learning-based encoding and decoding techniques to comprehend image information and convert it into textual descriptions. This approach enhances traditional image analysis by providing preemptive safety warnings and diversifying the output formats. Conventional methods for image caption generation lack controllability and detailed descriptions, and there is a paucity of research focused on image descriptions in power construction scenarios. To address this, this paper proposes a controllable image caption generation optimization method based on an encoder-decoder architecture. A novel feature extraction model is introduced, utilizing the FVCR-CNN (faster and visual commonsense region-convolutional neural network) as the encoder to extract salient and visual commonsense features from the images. The activation function is improved to develop an enhanced M-tanh-based long short-term memory (MT-LSTM) neural network for feature decoding. Finally, a multi-branch decision strategy is employed to optimize the output. The proposed method was trained and tested on a dataset of power scene descriptions using the Ubuntu 16.04 and PyTorch deep learning framework. Experimental results demonstrate a significant improvement in the accuracy of image caption generation, as well as enhanced controllability of scene descriptions, thereby substantially improving the intelligent level of safety management at power construction sites.

In this study, we introduce a novel dataset distillation method applied to medical imaging, specifically using the OCTMNIST dataset. Our approach incorporates advanced techniques such as class centralization and covariance matching constraints to efficiently synthesize smaller, yet representative, datasets from a larger corpus of optical coherence tomography images. The efficacy of our method is quantitatively validated against standard distillation techniques such as DC, DM, and DSA, showcasing superior performance in terms of classification accuracy and model stability. Additionally, we emphasize privacy protection by employing metrics such as the L2 norm and Structural Similarity Index Measure (SSIM) to ensure that the distilled images do not retain sensitive information. The results illustrate our method's ability to maintain high accuracy with fewer images while significantly enhancing data privacy, making it ideal for medical imaging applications where data sensitivity is paramount.

Aiming at the problem of improving the security of inspection image in the case of limited computing resources in substation intelligent patrol, this paper proposes a secure encryption algorithm for substation inspection image based on chaotic mapping and DNA coding. The algorithm uses the hash value of the image as the original key, realizes the secure sharing of the original key, and expands the key space. The algorithm uses one-dimensional chaotic map to generate chaotic sequence, and uses chaotic sequence to block encryption of image. Then, on the basis of DNA coding, the algorithm performed DNA scrambling encryption and DNA diffusion encryption on the image in turn. Experimental results show that this algorithm can effectively resist exhaustive attack, brute force attack and statistical analysis attack, and can effectively improve the security of inspection image. It has a broad application prospect in the field of substation intelligent patrol.

Target detection technology in remote sensing images holds significant importance in fields such as signal processing and radar imaging. However, due to the substantial influence of electromagnetic wave scattering characteristics on remote sensing image, the size of objects to be detected undergoes considerable changes, increasing the difficulty of detection. At the same time, the extensive background information contains confusing geographical elements, severely affecting the detection performance. We use YOLOv8 as the baseline model and employ Swin Transformer as the backbone structure to enhance the model's ability to connect context. Additionally, Coordinate Attention is added at the end of the backbone to deal with complex backgrounds, concentrating more weight on our points of interest. Finally, to detect smaller targets, we incorporate a smaller scale detection head to capture more subtle points of interest. Through experiments, our model's mAP@0.5, Recall, and Precision reached 0.628, 0.583, and 0.691, respectively, far surpassing several existing advanced models.

To improve the efficiency of the home design process and reduce time consumption, the experiment proposes an intelligent generation method for home design images based on an improved Generative Adversarial Network (GAN). First, the basic principles and structure of GAN are introduced, and then the concept of MSSF is introduced to improve the detailed performance and overall consistency of the generated images. The results show that experiments were conducted on the 7 Scenes data set and the LaMAR data set. When the running time is 0.845 s and 0.812 s, the AIGC-GAN algorithm proposed in the experiment has the smallest RMSE value, and the corresponding values are 4.32 and 0.874. In addition, the average PSNR value obtained when running on the 7 Scenes data set shows that when the system running time is 0.312 s, the average PSNR value of the AIGC-GAN algorithm is as high as 63.89 dB. The above results all show that the AIGC-GAN algorithm can effectively assist home design and contribute new ideas to the development of image generation technology.

In micro-area spectroscopic analysis, precise identification of characteristic X-ray peak positions holds paramount importance. Analyzing the energy information of these characteristic X-ray peaks, enables the determination of elements types present in the sample and their relative abundances. However, spectral graphs often present intricate challenges such as overlapping, weak spurious peaks, complicating the peak-finding process significantly. To tackle these hurdles in micro-area X-ray spectroscopy, this paper proposes an adaptive threshold symmetric zero-area peak-finding algorithm. In comparison to traditional methods like the simple comparison method, derivative method, and Gaussian fitting method, the symmetric zero-area peak-finding algorithm demonstrates superior advantages in detecting overlapping, weak, and spurious peaks. Through peak-finding experiments conducted on original spectral data from various samples, and comparing results with the standard characteristic X-ray element tables and other classical algorithms, it’s evident that symmetric zero-area algorithm not only accurately distinguish each individual peak but also effectively distinguishes real peaks from noise-induced false peaks, proving feasibility of the symmetric zero-area peak-finding algorithm in micro-area spectroscopy.

With the rapid development of high-precision and high-speed ADC devices, traditional analog processing methods for nuclear pulse signals have gradually exited the historical stage. Digitizing nuclear pulse signals has become the main research direction in multi-channel nuclear energy spectrum measurement systems. The trapezoidal shaping filtering algorithm, as the most used nuclear pulse shaping algorithm, has a certain noise suppression capability. This paper studies the double-exponential trapezoidal shaping filtering method based on FPGA in practical applications and explores its role in further eliminating pile-up effects under high counting rates.

The layout design of millimeter-wave radar is the outset in its application to fields such as multi-UAV collaborative three-dimensional positioning and deformation monitoring. This involves the introduction of Position Dilution of Precision (PDOP), an indicator in the Global Navigation Satellite System (GNSS) used to evaluate the satellite constellation, as the optimization object for system geometric layout, and the construction of an optimization objective function combining the relevant characteristics of millimeter-wave radar. In the present work, the layout of millimeter-wave radar was optimized using the Incomprehensible but Intelligible-in-time (IBI) Logics Optimization Algorithm. Leveraging this algorithm, the optimal layout solution for subway tunnels was achieved after around 50 iterations, obtaining a weighted PDOP value of 1.815 for the target points in this layout. The findings show that the proposed method yields a layout result resembling that obtained by the traversal algorithm, but with significantly improved efficiency, demonstrating its practicability in the layout optimization of millimeter-wave radar.

Aiming at the restoration of traffic sign images in foggy weather, this paper proposes a traffic sign defogging algorithm based on an improved Conditional Generative Adversarial Network (CGAN). Firstly, the algorithm introduces a generator with a symmetric skip connection structure and designs a novel loss function. Secondly, the TT100K high-definition image dataset is utilized to construct pairs of clear-day and simulated foggy-day images, and the atmospheric scattering model is adopted to simulate different levels of haze for training. Finally, through comparison with other mainstream defogging algorithms, the experimental results demonstrate that the improved CGAN achieves superior defogging effects on the test set, significantly improving the clarity and recognition rate of traffic sign.

The decision tree classification algorithm identifies category patterns in data by constructing a tree like structure, while the decision tree regression algorithm predicts the trend of continuous numerical changes based on the tree model. This article uses decision tree classification and decision tree regression algorithms to analyze the market for mini program users. Through classification algorithms, we have gained a profound insight into the activity classification characteristics of mini program users at different time periods. These analyses not only reveal the evolution trajectory of user behavior patterns, but also provide us with an advanced perspective to understand market dynamics. At the same time, regression algorithms, with their precise predictive ability, accurately depict the growth and decline of the market size used by mini programs, injecting strong momentum into predicting market trends. This comprehensive analysis method that integrates classification and regression algorithms endows mini program developers with the super ability to insight into user activity in the time dimension, thereby helping them formulate strategies more accurately and efficiently, and promoting the stable and sustainable growth of mini program user scale.

Automatic sleep stage classification plays a crucial role in assessing sleep quality and diagnosing sleep disorders. While many automated methods exist for sleep stage classification, most rely solely on single-channel electroencephalogram (EEG) signals. In contrast, multi-channel polysomnography (PSG), which includes EEG, electrooculogram (EOG), electromyogram (EMG), and electrocardiogram (ECG) signals, provides more comprehensive and accurate sleep staging information. These signals not only reveal complex physiological changes during sleep but also aid in understanding the characteristics and variations of different sleep stages. This capability offers valuable support for both sleep research and clinical practice. Therefore, we propose MCTSleepNet, a Transformer-based model for multi-channel sleep staging. It takes signal images transformed via time-frequency features as input. The core architecture uses a Transformer encoder to capture joint features across multiple channels, complemented by residual connections to retain original input information. The experimental results demonstrate that our model outperforms other state-of-the-art techniques on the publicly available ISRUC-S3 dataset. MCTSleepNet is an effective multi-channel sleep staging method that significantly aids in clinical sleep staging tasks.

This study aims to design and implement a financial risk control system based on machine learning algorithms and big data technologies to improve the risk management capability of financial institutions. The system is designed using key steps such as data collection and collation, feature engineering, model selection and training, risk assessment and prediction, risk reporting and monitoring, and automated decision making and trading.In the data collection and collation phase, the system collects a large amount of financial market data, transaction data and customer data, and ensures the accuracy and consistency of the data through data cleansing and collation. Next, the system performs feature engineering, using machine learning algorithms to extract and transform features from the data to better express the characteristics and relationships of the data. This includes statistical features, time series features, technical indicators, etc. The system then selects machine learning models suitable for financial risk control, such as decision trees, random forests, support vector machines, etc., and uses historical data to train and optimise the models. The trained models are used to perform risk assessment and prediction on current data, such as assessing the risk of investment varieties such as stocks and bonds, predicting market volatility, and detecting abnormal transactions. The system also generates detailed risk reports and monitoring indicators to help financial institutions identify and respond to potential risk events in a timely manner. This includes the calculation of risk indicators and the generation of risk alerts, etc. Finally, the system can automatically trigger risk management strategies and trading decisions based on risk assessment results to reduce risk and improve investment returns. The design and implementation can effectively improve the risk management capability of financial institutions and promote the stable development of the financial market. However, challenges such as data quality, model interpretability and supervision still need to be addressed during implementation. Further research could explore how to optimise the accuracy and efficiency of the model and improve the application scope and stability of the risk control system.

Power grid line fault diagnosis is a key task to ensure the reliability and stability of the power system. Traditional fault diagnosis methods rely on centralized data collection and processing, which are often constrained by data privacy concerns, security issues, and high transmission costs. This paper proposes a method based on federated learning for power grid line fault diagnosis to address the challenges of data dispersion, privacy protection, and efficient computing. The study designed a federated learning architecture that includes multiple power grid substations. Each substation trains on local data and aggregates model parameters on the central server through a variety of aggregation algorithms (such as FedAvg, FedProx, etc.). The experiment evaluates the performance of five different neural network architectures (MLP, RNN, LSTM, GRU, and CNN) in processing non-independent and identically distributed (non-iid) data. Results show that the federated learning model is comparable to centralized learning in prediction accuracy while significantly reducing computational and communication costs. In addition, the accuracy of fault diagnosis is further improved through appropriate preprocessing techniques such as outlier processing and global normalization. This paper verifies the potential of federated learning in power grid line fault diagnosis and provides new ideas for efficient operation and maintenance of smart grids.

This study presents the design and simulation of a financial services risk control system based on an improved genetic algorithm (GA). The system aims to effectively manage and control the risks associated with financial services such as investment portfolios, loans and insurance. The improved genetic algorithm employs several improvements to provide more accurate and efficient risk control. Firstly, a new fitting function is introduced which takes into account multiple risk factors and their associated weights. This allows the system to prioritise and allocate resources based on the importance of each risk factor. In addition, the improved selection operator enhances the genetic algorithm to increase population diversity and prevent premature convergence. This ensures that the algorithm explores a wider range of solutions and avoids falling into local optima. In addition, a local search operator is incorporated into the genetic algorithm to refine the solutions obtained from the genetic operator. This local search operator helps to improve the quality of the solution by making small adjustments to the individuals in the population. In order to evaluate the performance of the proposed system, we have prepared simulation code using Python language to perform extensive simulations on real-world financial data. The results show that the improved GA-based system outperforms traditional risk control methods in terms of accuracy and efficiency. In conclusion, this study presents a new approach to financial services risk control using improved GA. The proposed system provides a more comprehensive and efficient method for managing financial services risk, which contributes to the stability and sustainability of the industry.

In order to analyze and solve the problem of reconfiguration of power system after fault occurs during ship sailing, a fault self-healing model of ship ring power network is established, which takes load power, switch action and generator efficiency as optimization objectives, and an improved binary particle swarm optimization algorithm is proposed. Aiming at the disadvantages of traditional binary particle swarm optimization, chaotic mapping is added. Secondly, the inertia weight is changed dynamically according to the number of iterations, and the contraction factor is improved by sinusoidal function, which makes the contraction factor decay slowly in the early stage of the algorithm and accelerate the decay speed in the later stage. Finally, the particle position is discretized to make it suitable for the fault reconstruction of ship power system. In the case of generator fault and load fault, the improved algorithm has better load recovery and switching times than the traditional algorithm, and has some advantages in iteration times and convergence speed. It shows that the proposed method can obtain a better reconstruction scheme, overcome the problems of poor population diversity and local optimality of the traditional binary particle swarm optimization algorithm, and ensure the safe operation of the ship better.

The system designed in this paper uses machine learning algorithms to construct the portrait of excellent talents in digital economy, and the system includes steps such as data collection, data preprocessing, feature selection, model training and model evaluation. First, the system collects relevant personal and professional information data, including education, work experience, skills, etc., through web crawlers and API interfaces. Next, the system uses a feature selection algorithm to select features from a large number of features that have an important impact on digital economy excellence talent. Finally, the system assesses the quality and effectiveness of the model by evaluating its performance metrics, such as accuracy, recall, and F1 value. In order to verify the effectiveness of the system, this paper uses Python programming code to conduct simulation experiments. The experimental results show that the system is able to predict the outstanding talents of the digital economy more accurately and identify the key features that have an important impact on the development of the digital economy. These results validate the feasibility and effectiveness of the system, and provide an effective method for enterprises and organisations to identify and cultivate excellent talents in the field of digital economy.

Visible Light Communication (VLC) technology solves the problem of limited spectrum in traditional mobile communications, and the data transmission rate can reach hundreds of megabits. However, nonlinear distortion also exists in visible light communication systems. Nonlinear distortion will cause a nonlinear relationship between the transmitted signal and the received signal, resulting in a higher bit error rate when the signal is decoded at the receiving end. In order to improve the performance of the VLC system, it is necessary to suppress the nonlinear distortion generated by the signal during the transmission process. Based on the understanding of commonly used VLC system models, LED nonlinear models and two common neural network models, this paper studies a post-distortion compensation algorithm based on a hybrid algorithm of genetics and particle swarm and a Back Propagation (BP) neural network. The received signal of the system is sent to the network, and the original transmitted signal is used as the desired output. The initial weight of the BP network is pre-optimized through a hybrid algorithm of genetics and particle swarm, and then the transmitted signal is estimated with the help of the trained network to achieve distortion compensation. Experimental results show that the improved optimization algorithm. It can effectively suppress the nonlinear characteristics of the VLC system, and appropriately expand the hidden layer of the network to improve the nonlinear mapping ability of the network to a certain extent, so that the compensated signal can better fit the original input signal, which improves the system performance.

The underwater world has always been a source of curiosity, and with technological innovations, travelling into the deep sea on a submersible has become a popular tourist activity. But with that comes the potential risks associated with technical failures. In order to better cope with unexpected situations, including loss of propulsion, we have built predictive models to help find the potential location of the submersible and aids in search and rescue operations.The model successfully achieved all of the goals. It was not only fast and capable of handling large quantities of data, but also provided the desired flexibility.

Mud leakage poses a persistent challenge in drilling operations, necessitating innovative solutions for improved sealing and circulation efficiency. This paper leverages data from the Mullen Field to categorize mud loss severity levels. Employing the Optuna framework, hyperparameter optimization enhances model performance metrics such as precision, recall, and F1-score. Our study underscores the efficacy of hyperparameter tuning in refining predictive models for mud loss severity.

Extreme weather events create crises for homeowners and insurance companies. As insurance companies adjust coverage, property insurance is becoming more expensive and harder to get. This study aims to examine the sustainability and related issues of property insurance by estimating the risks of climate change to cash crop production and creating weather index insurance programs for four regions in Canada. In this paper, genetic algorithm, particle swarm algorithm and error back propagation algorithm are used to find the optimal solution. The effectiveness of weather index insurance to protect against crop production losses under climate change scenarios in 2030, 2060 and 2090 is analyzed, highlighting current challenges affecting social inclusion and sustainability of insurance schemes.

In order to improve the coverage of Lora gateway communication, taking the coverage of Lora gateway as the maximum target, an improved firefly algorithm based gateway coverage optimization method (IFA) was proposed. Firstly, the Lora gateway communication coverage model is constructed according to the coverage area and the number of the gateway. Secondly, based on the firefly algorithm, particle swarm optimization algorithm is applied to improve the location of the gateway node during its movement, which can effectively improve the coverage uniformity of the gateway, and the coverage of the gateway area can be achieved by simulating the number of Lora gateways through the number of fireflies. Finally, the coverage area of the gateway can be maximized by the improved firefly algorithm. The simulation results show that compared with the basic firefly algorithm and genetic algorithm, the proposed algorithm can consume less resources to achieve better optimization effect, and the coverage rate is improved by 22.54% compared with the previous algorithm, and the coverage rate of gateway nodes is improved better.

This paper explores the integration of strategic optimization methods in the context of search advertising, focusing on ad ranking and bidding mechanisms within e-commerce platforms. Employing a combination of reinforcement learning and evolutionary strategies, we propose a dynamic model that adjusts to varying user interactions and optimizes the balance between advertiser cost, user relevance, and platform revenue. Our results suggest significant improvements in ad placement accuracy and cost-efficiency, demonstrating the model’s applicability in real-world scenarios.

This article proposes a cross-border e-commerce customer file processing system based on an improved genetic algorithm. The system utilizes improved genetic algorithms and big data technology of artificial intelligence to process and analyze cross-border e-commerce customer data, create customer profiles, and provide personalized services and recommendations for enterprises. It also introduces more advanced machine learning technology to improve the system's ability to predict and identify customer needs, thereby providing more accurate and diverse personalized recommendation services. This article provides a detailed introduction to the design and implementation process of the system, and conducts simulation experiments. The results indicate that the system can effectively process cross-border e-commerce customer data and generate accurate customer profiles. Therefore, a more precise and accurate understanding of customers becomes the key for cross-border e-commerce enterprises to build core competitive advantages.

The study focuses on the perforation process of ultra-long drilling and grouting piles under the complex geological conditions of the Yellow River alluvial plain. Spectral clustering analysis is employed to conduct an in-depth investigation of various regional geological attributes. These geological attributes include soil heterogeneity, fluctuations in groundwater levels, and the accumulation of sediments, which significantly influence the perforation behavior of ultra-long drilling and grouting piles. After collecting geological attribute data from 300 sample points, the spectral clustering algorithm is applied to categorize these points into five distinct geological categories. Through a detailed analysis of the clustering results, distinct characteristics of different geological categories and their spatial distribution within the research area are obtained. For regions with high sediment water content, it is recommended to use underwater construction techniques, utilizing suspended construction platforms to enhance foundation stability. Directional drilling using rock drills and foundation reinforcement through grouting piles is suggested in areas with low rock water content. In clayey soil regions, the construction technique of mixing piles is recommended to improve the strength and stability of the clay. Vibratory piles may be a practical choice for sandy soil regions, with the consideration of using jet grouting piles to enhance pile stability. In areas with complex strata, a multi-technique approach, such as combination piles, is advised to adapt to diverse geological conditions. This study not only provides an in-depth understanding of the geological attributes under the complex geological conditions of the Yellow River alluvial plain but also offers practical guidance for the construction techniques of ultra-long drilling and grouting piles. This guidance holds substantial significance for the foundation engineering design and construction in the Yellow River alluvial plain region.

With the rapid advancement of object detection technology, its role in intelligent transportation systems has become increasingly significant. However, two major challenges persist in complex traffic environments: 1) the need for further optimization of real-time detection and 2) limited detection accuracy under multi-scale and multi-target conditions. To address these issues, we propose a Multi-Scale and Multi-Target Thermal Infrared Traffic Detection method based on an improved YOLOv8. A novel Parm_C2f module is introduced, which increases the number of parameters without compromising the inference speed, thereby enhancing the model's encoding capability. Additionally, the DNC_Detect Head, leveraging deformable convolution, is incorporated to improve the model's performance in multi-scale and multi-target detection. The experimental results demonstrate that the proposed method outperforms the YOLOv8 algorithm on challenging thermal infrared radiation datasets, showcasing its robust object detection performance in the transportation domain.

In order to study the capture of heavy metal ions by Salophen, we used the complexes formed by Salophen with five heavy metals: Pb, Hg, Cr, Cd, and As, and conducted coordination studies using the quantum computing software Gaussian 16. The results indicate that Salophen is indeed capable of capturing these five heavy metal ions and forming stable complexes. Moreover, Salophen exhibits strong binding energies with all five heavy metals, enabling effective capture of these ions. More importantly, Salophen can perform quantitative and qualitative detection of these heavy metal ions through ultraviolet-visible absorption spectroscopy. Notably, for Cr and As, detection can be carried out using both ultraviolet and visible spectroscopy.

Plant diseases have a great impact on global agricultural production and ecological security, which can cause the reduction of agricultural production and quality, and also bring food security problems. Traditional methods for detecting crop diseases rely heavily on manual observation and judgment, which suffer from issues such as long processing times, low efficiency, and dependence on expertise. In contrast, deep learning-based detection methods automate the identification and detection of crop disease images through training neural network models, offering advantages such as rapid processing speed, high accuracy, and enhanced efficiency. Therefore, this study improves the yolov8 network on the basis of deep learning to establish an accurate, stable and fast network detection model to provide a fast and efficient disease diagnosis solution for disease detection.

With the promotion and application of LNG loading skid automatic control technology and information management technology, LNG loading stations are gradually developing towards unmanned stations. In the unmanned stations, in addition to automation equipment, inspection robots used for station inspection and maintenance are also essential. This article focuses on the research of dual arm inspection robots for unmanned stations, and conducts research on the coordinated motion control of dual arms for inspection robots in loading stations. Firstly, perform D-H parameter kinematic modeling on the designed robot; Secondly, design kinematic solving algorithms and dual arm collaborative motion algorithms based on the D-H model; Finally, a simulation platform is built in the ROS robot system for gripping simulation testing to verify the feasibility of the algorithm.

Gesture recognition, as a non-contact and intuitive human-computer interaction, has a unique advantage in the field of home control. This paper proposes a gesture recognition for home control based on YOLOV8, and the YOLOV8 model is one of the modern advanced object detection models. YOLOV8 model has great potential in the field of gesture recognition due to its advantages of high precision, real-time performance and adaptability. Specific gestures are collected using OPENMV for data acquisition, and pre-processed by color smoothing, canny edge detection algorithm, adaptive threshold algorithm, morphology closing operation, sharpening, etc. The processed images were imported into the YOLOv8 model for training, the trained YOLOv8 model was used for gesture recognition, the recognition results were converted into string format, and the recognition results were sent to the STM32F103 development board through serial communication to further control the home appliance.

With the continuous rise of intelligent manufacturing, pre-weld seam positioning has become an important step in achieving intelligent welding. The method of seam positioning using models can quickly locate the seam information. Firstly, the CAD model of the workpiece is point-cloudified using VTK to obtain the point cloud of the workpiece CAD model, and the seam parameters are picked up using VTK based on the CAD model. Secondly, the registration time is reduced by obtaining the key points of the workpiece point cloud and model point cloud, and the feature descriptors of the key points are used to describe the point cloud features. After rough registration using SAC-IA, the initial pose of the workpiece point cloud and model point cloud is correct. Finally, the workpiece point cloud and model point cloud are registered through iterative closest point matching, and the conversion matrix from the picked model seam parameters to the actual seam parameters is obtained. The registration error is 1.07-1.75mm. The experimental results show that the maximum positioning error after registration is 2.02mm.

Given the problems of a large number of flame and smoke detection parameters and poor detection effect in the current fire warning system, this paper proposes a lightweight YOLOv8n improved algorithm. First, in terms of data sets, the Mosaic data enhancement method is used to increase the diversity of data. Secondly, the C2f module in the Backbone part is fused with the LSK attention mechanism to form a C2f-LSK module, which can increase the ability to extract flame and smoke features in complex fire scenes, thereby improving the detection accuracy. Then, for the Neck part, the weighted Bidirectional Feature Pyramid (BiFPN) is used to replace the original Path Aggregation Network (PANet) of YOLOv8, which promotes the fusion of feature maps of different scales, thereby improving the detection accuracy of the model. Experiments show that the improved YOLOv8n has high accuracy in flame and smoke detection, with precision and mAP@0.5 reaching 80.8% and 75.3% respectively, which are 3.8% and 3% higher than the original model. The improved YOLOv8n network model has higher accuracy and lower false alarm rates in flame and smoke detection.

In order to achieve the possibility and probability of discovering the violation of operation tasks through previous data, this paper proposes an intelligent identification algorithm for safety risk of transmission line maintenance operations based on deep learning. Firstly, a large amount of transmission line maintenance operation image data is collected, secondly, the size of image chunks is determined by using a priori information and experiments to reduce the number of layers of convolution and reduce the computing time. Finally, a standard library of image samples of transmission lines is successfully constructed, which achieves the standardised collection of inspection defect samples with key features of multi-source data and the standardised management of transmission equipment information, unified labelling transformation and efficient storage and retrieval of defect information. The experimental results show that real-time monitoring and early warning of potential safety risks in transmission line maintenance operations can be achieved through the deep learning-based intelligent identification algorithm for safety risks in transmission line maintenance operations, thus improving the safety and efficiency of maintenance operations.

As the demand for high-precision underwater measurements of nuclear fuel assemblies (NFA) continues to rise, there is a concurrent improvement in underwater optical measurement technology. The sheet of light, an optical measurement technology based on a straightforward principle, has progressively emerged as a focal point of research. However, the process of conducting high-accuracy measurements underwater presents significant challenges due to the impact of scattering, refraction, and intense radiation. This paper presents the development of a reflective underwater sheet of light system. It proposes an enhanced refraction correction method, utilizing ray tracing. Additionally, it suggests an improved center extraction method for underwater laser lines, based on the grayscale center of gravity, to mitigate random speckle interference. The comprehensive experimental results demonstrate that the system achieves an underwater measurement accuracy of 0.2mm to 0.4mm at a distance of 850mm from the NFA. These measurements confirm the effectiveness of the underwater measurement system and provide a methodical verification for the safety assessment of NFA.

To improve the efficiency of diagnosing retinal diseases, deep learning-based object detection methods are employed for auxiliary diagnosis. Given the excellent performance of the You Only Look Once (YOLO) series models in the field of object detection, this paper proposes an improved YOLO model based on the YOLOv8n for detecting fundus images. Firstly, to enhance the model's ability to detect small objects, the Concat in the neck network of YOLOv8n is replaced with BiFPN. Additionally, considering that lesions in fundus images often have irregular and blurred boundaries, the original CIoU loss function in YOLOv8n is replaced with the Inner-IoU loss function to improve the model's recognition accuracy. Experimental results show that the Mean Average Precision at IoU=0.5 (mAP@0.5) of the improved YOLO model reaches 0.963, which is an improvement of 2.2% and 13.8% compared to the baseline model YOLOv8n and the classic object detection model Faster-RCNN, respectively

This paper presents an enhanced YOLOv8 model incorporating Squeeze-and-Excitation (SE) Attention and Sub-Pixel Convolution aimed at improving traffic sign detection under various environmental conditions. Through rigorous evaluation on the CCTSDB2021 dataset, our model demonstrates superior performance over the standard YOLOv8 and YOLOv7 models, particularly in challenging low-light scenarios. The introduction of SE-Attention allows for refined feature recalibration, while Sub-Pixel Convolution effectively increases resolution, enhancing the detection accuracy. The findings confirm that our architectural enhancements significantly boost the precision, recall, and F1-score, thereby enhancing the model’s robustness and reliability. This study sets a new benchmark for traffic sign detection systems and proposes directions for future research to further advance the field.

Detection of dim target in star image background poses significant challenges due to the presence of noise and variations in object intensities. This paper presents an improvement to conventional pipeline filtering algorithms through the implementation of energy accumulation technique. By accumulating energy from multiple sources, the improved precision achieved in identifying dim targets within intricate celestial environments. The improvement also integrates background suppression and wavelet transform denoising methods to enhance detection efficacy. Through a combination of comprehensive simulations and empirical validation, the enhanced algorithm demonstrates higher success rates and greater robustness in dim target detection compared to conventional methods. The results indicate that the improved algorithm offers enhanced precision in identifying dim target in star image background, thereby advancing capabilities for astronomical research and space exploration endeavors.

This paper explores deep learning-based algorithms for electrocardiogram (ECG) signal processing and their applications in cardiac health monitoring. Initially, we provide an overview of the fundamental principles of ECG signals and the suitability of deep learning for ECG signal processing. Subsequently, we discuss data preprocessing techniques and the selection and construction of deep learning models aimed at enhancing the accuracy of ECG signal processing. Finally, the focus is on the application of these algorithms in the detection of arrhythmias and the diagnosis of myocardial ischemia, validated through experimental results. The findings of this study offer new technological approaches for cardiac health monitoring and foster the application of deep learning in the field of ECG signal processing.

When ice covers power lines, it can cause safety problems in power transmission. To ensure stable operation and reliable power supply of power system, automatic filtering and icing detection are realized using automatic algorithm such as deep learning technology, so that time consumption and human resources are greatly reduced.Therefore, this study designed a transmission line ice covering detection network RPN-YOLOv8 based on improved YOLOv8. First, the network uses the receptor field attention convolution and coordinate attention to improve the C2f module of the network backbone, so as to improve the network's position information capture and eliminate the problem of convolutional parameter sharing. Secondly, for the detection head with a large number of parameters in the original network, lightweight heavy design is carried out to reduce the number of parameters and improve the detection speed on the basis of ensuring the detection accuracy. Finally, combining WIoU and normalized Gaussian Wasserstein distance reframe loss function wnwd is constructed to enhance the detection accuracy of small size targets. The experimental results show that RPN-YOLOv8's parameters are reduced 12.6% compared to YOLOv8's, while the average average precision improves to 74.8%, improving 3.7% compared to the improved algorithm.

Monitoring cuttings migration in boreholes is crucial for ensuring the safety and efficiency of oil drilling operations. This paper details the design of a borehole cuttings migration monitoring simulation system based on Electrical Resistance Tomography (ERT) technology. The system includes a cuttings circulation module that ensures the continuous migration of cuttings within the system, and a cuttings monitoring module that employs ERT technology. By arranging an array of electrodes around the borehole, the monitoring module obtains real-time resistivity data and its variations, processes this data, and generates images on the host computer. The aim is to quantitatively and qualitatively analyze the state of cuttings migration in the borehole through measurement technology. Experimental results demonstrate that the system can accurately and effectively capture the trajectory of cuttings migration, providing reliable data support and a theoretical basis for drilling safety management and cuttings migration research.