Research On Sorting System Of Plate Stacked Parts Based On Machine Vision

With the in-depth implementation of "Made in China 2025",my country’s manufacturing industry is also accelerating its transformation to intelligence.As one of the key technologies for intelligent production,machine vision technology has been widely used in industrial production processes such as identification and sorting,size measurement,and defect detection.In the sorting system,accurate identification and precise positioning of targets are the prerequisites for robots to automatically grasp.In order to explore the method of identifying and positioning plane targets,this paper takes plane board parts as the research object.Aiming at the problem that the graspable targets in plane stacking parts are difficult to accurately identify and locate,a part identification and positioning method combining deep learning and template matching is proposed..Based on the research of template matching algorithm and deep learning target detection algorithm,the intelligent sorting system of stacked plate parts is built,which realizes the accurate identification and precise positioning of the parts that can be grasped in the stacked parts,and guides the robot to sort the parts.The sorting experiment lays the foundation for the realization of intelligent sorting of plane targets in industrial production.Firstly,complete the selection of machine vision hardware equipment,study the classic template matching method in target recognition and positioning,perform part image preprocessing operations according to the template matching steps,and select Halcon image processing software for template matching experiments based on the processing results.Secondly,several template matching operators in Halcon are analyzed.Finally,the scale template in shape matching template is selected for part recognition and location.The CAD drawings of parts are used as template images to quickly create matching templates.By adjusting the relevant parameters of template matching operator,the recognition and positioning experiments of single and multiple parts are realized respectively,and the recognition and positioning results are analyzed.Then,in view of the problem that template matching matches the occluded parts when recognizing and positioning stacked parts,the deep learning target detection algorithm is used to complete the graspable target recognition task before template matching.The Yolo v3 algorithm in the one-stage detection algorithm is chosen as the part recognition model,and based on the basic principles of the Yolo v3 algorithm,the model is modified accordingly to meet the needs of part detection in this paper.After that,using the modified network to train on the constructed data set,the average accuracy of the model reached 94.7%.Finally,the combination of deep learning part recognition method and template matching part positioning method is realized by constructing a sorting system for board parts.On the Visual Studio software development platform,the main interface of the system and other functional modules,namely the camera module,the deep learning part recognition module,and the template matching part positioning module are realized.Using the built software platform to carry out part recognition and positioning experiments,and design a robot control program according to the experimental results to carry out part sorting experiments,verifying the rationality and feasibility of the part recognition and positioning method combining deep learning and template matching.