Detection Of Silver Surface Defects Of Metal Workpieces Based On YOLOv4

With the development of computer vision,more and more automated quality inspection solutions are applied in the industrial field.The detection scheme based on digital image processing is excellent for the detection of specific scenes,but has certain limitations in performance such as robustness and generalization.The detection scheme based on deep learning has better robustness,but its disadvantage is that model training takes up a lot of computing resources,which makes it difficult to deploy in resource-constrained scenarios.This paper takes the silver-plated metal workpiece produced by an electronic enterprise as the research object,and studies an effective surface defect detection scheme from the two fields of digital image processing and deep learning.The main research work includes:1.This paper proposes a method for detecting surface defects of metal workpieces based on traditional image processing methods.First,in order to reduce the computational complexity,the collected metal workpiece image is grayed and the workpiece area is extracted.Second,use the emphasize algorithm to filter out noise and enhance image details,and use the Sobel edge detection algorithm and global automatic threshold segmentation method to obtain the region of interest for surface defects.Then,in order to further confirm the defect area,the gray co-occurrence matrix was introduced for feature extraction,and the features such as energy,contrast and correlation of each area were extracted and analyzed respectively.By taking the contrast as the characteristic index,the defective area is obtained by regional screening,and whether the product corresponding to the original image is qualified or not,the final detection rate reaches 82.6%.2.In this paper,a deep learning-based method for surface defect detection of metal workpieces is constructed.First,YOLOv4 is selected as the detection model.Considering reducing the training cost of the model,a depthwise separable convolution module and a lightweight network MobileNetV2 are used to replace the standard convolution module and feature extraction network in the original model.Second,an image prediction method is proposed.The processing method locates and crops out the area of the workpiece in the image before input to the detection network.Then,in view of the poor performance of the original mosaic data enhancement method in this dataset and to prevent training overfitting,a new data enhancement method is introduced.Finally,in order to speed up the training convergence,the model is pre-trained by means of transfer learning.The experimental results tested on the data set collected from the field show that the detection accuracy of this method for workpiece defects reaches 90.63%,and the detection speed is 33 frames per second.Compared with the original YOLOv4 model,the model size is reduced by82.1%,and the detection accuracy is improved by 1.8% as well as the detection speed by150%.Compared with classic models such as SSD and Faster R-CNN,the model has excellent comprehensive performance in detection speed and detection accuracy,which can efficiently detect silver surface defects of workpieces.The actual industrial application has a certain reference function.