ESDDNet: efficient small defect detection network of workpiece surface. (1st October 2022)

• Record Type: Journal Article
• Authors: Chen, Guodong; Xu, Feng; Liu, Guihua; Chen, ChunMei; Liu, Manlu; Zhang, Jing; Niu, Xiaoming
• Is Part Of: Measurement science & technology. Volume 33:Number 10(2022)
• Publication Date: 2022-10-01
• Subjects: deep learning -- workpiece surface -- small defect -- efficient detection; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/ac793d
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: Defect detection of a workpiece surface is a basic and essential task in the production of products. Although significant progress has been made in workpiece surface defect detection, traditional methods still find it difficult to detect small defects efficiently. To deal with this problem, we propose an efficient small defect detection network with a novel parallel convolution module, serial convolution module and feature fusion module. First, a lightweight backbone network is used to extract the preliminary defect features. Second, the parallel convolution module and serial convolution module are used to obtain the abundant defect features. Then, the feature fusion module is used to fuse the shallow features with deep features, to enhance the features of the small defects. Finally, the obtained features are put into the corresponding detection head to get the final prediction results. The experimental results on a local cable dataset and a public printed circuit board dataset show that our method achieves a remarkable performance in detecting small defects and achieves a favorable trade-off between accuracy, speed and model size, which meets the requirements of industrial applications.