AUTHOR=Wang Biao , Huang Yan , Yang Yongyue , Wang Yonghong , Li Hongli , Huang Bin , Chen Jianbin 

TITLE=CPDD-CLMM: a comprehensive lightweight mobile-optimized network for composite plate defect detection

JOURNAL=Frontiers in Physics

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2023.1264636

DOI=10.3389/fphy.2023.1264636

ISSN=2296-424X

ABSTRACT=Automatic defect detection technology based on deep learning is increasingly concerned with distinguishing production quality by more and more industry companies. But production lines are usually installed with lots of function modules, make it hard to integrate new modules. Common deep learning models run on PC platforms take lots of space with high cost, while ARM64 mobile platforms are much smaller with less cost, equivalent connectivity but also weaker performance.Considering these facts, ARM64 platforms with fully optimized model become the best solution for adding defect detection function for existing production lines. This paper focused on the mobile-optimized model to achieve higher speed and equivalent precision on the ARM64 mobile platform for detection. First, model's structure is simplified by reducing redundancy of feature maps to increase the network's inference speed. Second, a convolutional block attention module (CBAM) is attached to compensate for the decreasing precision caused by structure simplification. Furthermore, transfer learning method is adopted to improve training performance. Finally, the trained and compiled module is exported to the PyTorch Mobile format and deployed on the mobile platform application to execute its defect detection function. The results show that the optimized network realizes 2.124 fps, 210.7% speed compared with YOLOv5n of 1.008 fps on RK3399 ARM64 platform, and has an average mAP of 99.2%. The studied mobile optimized model has a better speed and equivalent precision, and can be available on many different ARM64 platforms regardless of processor manufacturer. It can satisfy the need for real-time defect detection and can be used in the similar scenarios.