Regular surface damage detection of large concrete structures is one of the important measures to ensure their stableness and reliability. Recent advancements in computer vision and deep learning have been increasingly applied to the high-precision detection of concrete surface damage. However, most damage monitoring and localization methods are based on high-resolution images taken from close range, and the images usually contain tiny areas of actual structures. This paper proposes a contrastive embedding model to detect and localize damage on a wide range of concrete images. To achieve high-definition imaging of damage to the surface of large structures, a multifocus, and high-resolution imaging system is designed and employed. Furthermore, a concrete surface damage dataset containing 1006 detection areas and 45,375 images with seven types is constructed by manual annotation based on the obtained multiscale and multiresolution images. In addition, a contrastive embedding model based on a deep neural network is then modified, trained, and tested using the constructed dataset. To the best of our knowledge, this paper is the first to jointly use contrastive embedding to deal with damage monitoring and localization. Moreover, an evaluation framework based on sliding window iteration and nonmaximum suppression is proposed to verify the robustness and accuracy of the proposed contrastive embedding model. Experiments show that the best model achieves the classification accuracy of up to 94.84% and localization of 97.57%.
Semiconductor wafer is elementary unit in semiconductor industry. In the fabrication of semiconductor wafer, surface defects such as dirties, scratches, burrs, chippings and holes may be generated which severely affect the quality of downstream production. Typical inspection of these defects mainly depends on human experts inspecting system which is time-consuming and low efficiency. With the fast development of digital imaging and processing technique, Computer vision automatic inspection method has shown vast potential for product quality test. Due to low contrast and weak context characteristics of wafer surface defects, the existing methods have difficulty to extract whole defect patterns. A novel algorithm for defect contour extraction is proposed based on multi-frame differential image summation. For each side of semiconductor wafer surfaces, multiple images are captured by high resolution digital camera. For each image the gradient is calculated using common used differential mask, and then the gradient is thinned based on edge extraction using Canny operator and smoothed using Gaussian smooth filter. All refined gradient images are added up to enhance defect features and smooth defect-free regions furtherly. Finally, the Canny operator is applied again to extract whole defect’s contour from gradient summation image. Experiments using real semiconductor wafers illustrate that the proposed algorithm can detect most of defects correctly and effectively.
3D foot digital models have great potential for application in ergonomics design and online virtual shoes try-on. Traditional techniques usually take 10 seconds to several minutes to acquire dense data, which is a critical limitation to large scale data collection. To enhance data collection efficiency, a novel approach which combines simulated laser speckle projection stereo with 3D silhouette is proposed. Laser speckle has more statistical advantages so that it can deal with lacking of texture of human skin and make stereo matching easier. With dark field lighting that strengthen foot contour, 3D silhouette can remove border noise caused by our active stereo reconstruction. Besides, all light sources in our work are infrared to avoid ambient light inference. In our design, five active stereo rigs are installed around 4π solid angle centered at the foot position to capture whole foot’s surface mesh data. Composed of a pair of stereo IP-cameras with visible cut filters, an infrared simulated laser speckle mini film projector and a cluster of infrared LEDs surrounding camera lens, each active stereo rig takes charge of obtaining 3D information of corresponding foot part. The five rigs are controlled by an MCU controller to successively capture one pair of speckle pattern images for stereo reconstruction and one pair of edge enhanced dark field lighting images for silhouette. The system design resolution is less than 0.3 mm per pixel. Data capture could be performed in less than 1 second for each foot and more than 500 thousand valid points are acquired as dense point cloud model. Finally, foot mesh model is generated using Poisson reconstruction algorithm.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.