Center of Ultra-Precision Optoelectronic Instrument Engineering and the Key Laboratory of Ultra-Precision Intelligent Instrumentation, Ministry of Industry and Information Technology, Harbin Institute of Technology, Harbin, China
Auto-generated profile, based on publicly available information
Zhao Zhang
Center of Ultra-Precision Optoelectronic Instrument Engineering and the Key Laboratory of Ultra-Precision Intelligent Instrumentation, Ministry of Industry and Information Technology, Harbin Institute of Technology, Harbin, China
Facing the demand of industry for metal surface detection, this article addresses the segmentation of metal surface defects. Metal surface defect segmentation has many technical problems, such as serious imbalance in quantity and distribution, low contrast, and weak boundary information. To solve these problems, this article proposes a multi-scale attention feature fusion module, normalized mean square frequency category weight strategy, biased weight training sampling strategy, and category-boundary loss calculation strategy. The multi-scale attention feature fusion module fuses different levels of feature information to obtain defect features in images more effectively. The normalized mean square frequency category weight strategy suppresses the excessive weight of small defect categories and false positive classification. The biased weight training sampling strategy allocates different training frequencies to training samples and increases the attention to defect samples. The category-boundary loss calculation strategy imposes weights on the defect boundary and increases the attention to the boundary information. On four public datasets (Severstal, SD900, Magnetic-tile, and Crack), the proposed method achieves improvements of 7.8%, 4.0%, 18.4%, and 9.4% (absolute quantity), respectively, compared with the baseline method, while the calculation amount and number of parameters remain almost unchanged. Hence, our method ranks as state of the art.
Accurate semantic segmentation is crucial for monitoring the quality of metal surfaces in industrial production. To solve the issues of the scarce quantities and uneven distributions of metal surface defects, challenging to achieve real-time detection and hardware integration, and hard to capture boundary information, this study proposes a dual-attention multi-scale residual aggregation network, category weight calculation method, defect migration topology method, and loss calculation method for dual boundary attention. The methods solved the technical issues by aggregating the multi-scale information of the original image and exerting attention, changing the weight coefficients of categories, expanding the datasets using the topology of the defects of defective samples to a defect-free image, and paying dual attention to the boundaries of ground truth and predicted image. Compared to the fifteen mainstream methods and our previous work, this study achieved a favourable performance on five public datasets with 5.1 M parameters and real-time inference speed of 37.5 fps. Additionally, this study demonstrates commendable robustness in the presence of noise. Our code locates at https://github.com/zz-ux/Metal-surface-defect-detection.
17 October 2023 in Multimedia Tools and Applications
Photometric stereo represents a sophisticated technique utilized to capture high-fidelity surface textures and morphologies under varying illumination angles, with substantial applications in fields such as quality assurance and cultural heritage preservation. However, contemporary implementations often grapple with challenges in achieving high-precision measurements that are both economically feasible and conducive to real-time processing. To address these challenges, we propose a Dual-Scale Feature Fusion (DSFF) photometric stereo network coupled with a Shadow-Fusing Supervised (SFS) training strategy. By aggregating dual-scale features and effectively leveraging shadow-occluded regions, the proposed approach overcomes limitations of existing techniques, enabling low-cost, real-time, and high-precision measurement of complex morphologies across diverse materials. Experimental results on four authoritative public datasets demonstrate that our methods outperform nearly 30 state-of-the-art approaches in key metrics such as measurement accuracy, memory consumption, and inference time. These results highlight the breakthrough improvements and the immense potential for application in various fields. The code used in this study is publicly accessible at https://github.com/zz-ux/PS-Measurement.