Sorting And Placement Of Chip Bulk Materials Based On Machine Vision

With the development of integrated circuits in China,the industry has developed rapidly.The placement machine has become the key of the production of automated circuit boards.According to the existing placement machine,this paper designs a machine vision-based patch bulk sorting and placing control system.This paper firstly introduces and analyzes the current status of domestic placement machines,industry needs and related technologies.Secondly,according to the system performance and the actual needs of the project,the vision-based patch bulks sorting and placing control system is designed,mainly including the mechanical control module,target detection module,text recognition module and the control system module.The target detection system based on YOLOv2 predicts the position and kind informations of various patch components according to the processed images.In this paper,the dataset collected by ourselves is used for training the YOLOv2 network which parameters are modified.At the same time,the clustering algorithm is selected according to the distribution of the sample dataset to get clustering parameter values,and the clustering parameter values which is more suitable are used to perfect the YOLOv2 network.The OCR text recognition is used to predict the close-up patch image,and the patch text information is obtained which is used to get the specific information of the patch.This paper uses a modified CNN+SVM network structure to make the text recognition more acurrate.And the abnormal prediction samples are corrected according to the existing patch information.The overall control system is designed which uses the OCR and target recognition algorithm to predict,converts the image coordinates and actual coordinates,analyzes PCB text,and completes interactive interface.At the same time,the placement machine is driven according to the communication protocol.Finally,the improved placement system was repeatedly debugged and tested,then the target recognition system and OCR were tested separately,and the overall experimental observation was completed in the indoor environment.