| With the rapid development of technology,miniaturization,precision and integration of products have become a development direction that cannot be ignored,and injection molded parts,as an important part of products,are facing the same problems.In this case,to ensure the quality of small and complex injection molded parts,to avoid the defective parts into the final product has become an important research topic.In this paper,we study the detection and identification of surface defects by using image inspection technology for small injection molded electronic connectors,which is important to ensure the overall quality of injection molded products.The image acquisition and detection system for surface defects of small and complex injection molded parts is established,the original image is compressed by digital interpolation,the compressed image is expanded in quantity by affine transformation data enhancement,the image features are enhanced by differential algorithm,the expanded image data is labeled with various types of defect targets by using Label IMG image labeling tool,and the data is processed based on PASCAL VOC data set format is used for processing,and the surface defect data set of injection molded parts is constructed.A two-stage deep learning-based defect detection system for injection molded parts is established,using residual structure as the basis of convolutional network,constructing lightweight network and residual network instead of traditional network structure to achieve feature extraction,proposing an improved target detection algorithm based on surface defects of injection molded parts,using area proposal network and clustering algorithm to achieve dimensional prediction and generation of anchor frame,transforming feature by bilinear interpolation The surface defect detection model based on residual structure and improved target detection algorithm is finally completed.Experiments were conducted on the traditional model and the improved model,and the training,validation,and testing results can show that both improved models have a certain degree of improvement in detection accuracy compared to the traditional target detection model,with the highest accuracy of the target detection model based on the residual network and the improved detection algorithm.Meanwhile,based on this,a QT-based graphical user interface is designed to integrate the process of detecting surface defects of injection molded parts. |
