Research on intelligent detection system for surface defects of workpiece materials based on deep learning optimization model

Abstract Intelligent detection of surface defects of workpiece materials is a key link in quality control of modern manufacturing industry. Aiming at the limitations of YOLOv5 in small target detection, an intelligent detection system for surface defects of workpiece materials is proposed based on the improved YOLOv5 model, which firstly adopts the bidimensional empirical mode decomposition (BEMD) for the image threshold denoising preprocessing to effectively improve the image quality. Secondly, Hardswish activation function is introduced on the basis of YOLOv5 model and attention mechanism SimAM is added to enhance the model performance. Meanwhile, peak signal-to-noise ratio (PSNR) and mean Average Precision at Intersection over Union threshold of 0.5 ([email protected]) are adopted as the evaluation indexes of image processing and target detection respectively. Through the data processing of the workpiece material surface defect NEU-DET and bearing datasets, the proposed system improves the accuracy and recall while maintaining the detection speed, which significantly improves the workpiece surface defect recognition capability when compared with the performance of traditional models. The research results in this paper not only enhance the intelligent detection of defects, but also provide an effective solution for the automated quality control of industrial production.