Detecting Object And Direction For Polar Electronic Components Via Deep Learning

In recent years,with the development of computer vision,deep learning has been gradually applied to some fields that replace traditional algorithms,such as object detection and image classification.Among them,industrial intelligent manufacturing is one of the important application fields of deep learning object detection,which can improve the efficiency of industrial production.The classification,orientation recognition and positioning of polar electronic components play a vital role in the fields of industrial production,welding and inspection.The application of deep learning object detection to the identification and positioning of polar electronic components has great research significance and engineering application value.In this thesis,the method that detecting object and direction for polar electronic components via deep learning is proposed.By analyzing the deficiencies of the current research algorithms of electronic components,deep learning is applied to component recognition to achieve polar electronic components classification,polarity discrimination and positioning at the same time.The main research contents of this article are as follows:1.Extend the research on the classification of components to be able to identify the type,the location and the direction of the polar components.Change the problem of direction identification into a classification problem.2.The deep learning object detection algorithm is used to study the object detection and direction recognition of polar electronic components.The Faster RCNN and YOLOv3 algorithms do well in classification,direction recognition and positioning of polar electronic components.The mAP of the two algorithms reaches 0.9705 and 0.9922,respectively.3.By studying the length and width distribution of the object box of the training set,selecting the number of cluster points and the location of the initial cluster point.And then using the Kmeans algorithm to improve the design of Faster RCNN and YOLOv3 anchor box,so the mAP is increased by1.16% and 0.1% respectively,which reduces the training difficulty of the model to a certain extent and improves the accuracy of the model.4.Based on the YOLOv3 algorithm to achieve accurate classification,direction identification and precise positioning of polar electronic components,combined with the characteristics of the length and width distribution of the object box of the dataset,a network structure for large object detection named YOLOv3-Big Object is proposed.While improving the detection accuracy rate,the time for detecting a single picture is greatly reduced to about half of the original detection time.5.On the basis of 2,3,4,the stability of the above algorithm was studied.By adding different types of polar electronic component data with similar shapes,the performance of the algorithm is almost the same as when it is not increased;The same type of component device is soldered on the circuit board for testing,and YOLOv3-Big Object still achieves good results.In summary,it is of great research significance and engineering application value to develop a deep learning-based polar electronic component object detection and direction recognition method.