Research On Image Detection Of Solder Joint Defects Based On Deep Learning

With the development of the Internet of Things technology and intelligent information technology,a variety of electronic products have flooded into people’s daily life.It is the basis of intelligent perception and information collection.Therefore,the reliability of electronic products is directly related to whether the system can extract information successfully.Solder joint defects are common factors that affect the hardware reliability of various electronic products.Hence,intelligent detection of solder joint defects is an important step to raise production speed and improve product quality quickly.With the development and application of machine learning,the advantages of various algorithm applications based on deep learning have gradually manifested.In this paper,starting from the detection of solder joint defects of electronic components,the detection algorithm is studied as follows:Firstly,analyze the single-level target detection algorithm YOLOv3,and make it better in order to improve the detection accuracy of small solder joints defects.The shallow feature map contains more information of tiny target,in order to obtain more feature information of tiny defects and increase the detection accuracy,the shallow information is fused on the basis of the original YOLOv3 feature fusion and the multi-scale prediction scale is changed.So as to raise the detection accuracy further,the bounding box regression loss,confidence loss and the strategy of removing redundant prediction boxes are modified.Experiments on public data sets prove that the improved algorithm has better detection results in defect detection,especially in the detection of tiny targets.Secondly,the two-stage target detection algorithm Faster R-CNN principle is studied.In order to effectively extract the defect features of the solder joints,Res Net50 is used as the feature extraction network of the model to extract the features of the tiny defects;to fuse the deep and shallow information of the defect target,using the feature pyramid structure for reference to fuse features at multiple scales.Experiments on public data sets verify that the improved algorithm effectively enhence the accuracy of defective target detection.Finally,This paper detects the defect image of solder joints.The obtained electronic component solder joint defects are preprocessed and annotated with the defect image data set to obtain a standard data set.The defect image data sets are respectively applied to the improved YOLOv3 and Faster R-CNN algorithms to detect the solder joint defect targets.The experimental results verify that the improved two algorithms have a certain degree of improvement in the detection accuracy of solder joint defect images,and provide effective ideas for many industrial applications of defect detection.