Development Of Robotic Arm Sorting System Based On Deep Learning Object Detection

Sorting operation is a very common scenario in the manufacturing industry.Due to its repet-itive nature,sorting operation has become one of the important application scenarios of industrial robots.Conventional automatic sorting systems use traditional image methods which manually struct features for specific workpiece and use templates to match workpiece positions and then classify them.This method has low accuracy,poor robustness and weak portability.At the same time,the lack of image talents among domestic manufacturing practitioners has raised the threshold for the application of vision-based automatic sorting in manufacturing.Therefore,for the algorith-mic effect of the vision-based automatic sorting system and its application threshold,this paper designs and implements an automatic sorting system based on deep learning and a cloud platform for object detection algorithm training of the automatic sorting system.Firstly,based on the situation of large amount of deep learning and the specific scene of the automatic sorting system,the overall architecture of the automatic sorting system based on deep learning is designed,and the performance level of the automatic sorting system(grab the workpiece and acquire the image),The control plane(image processing and robotic arm control)and the back-ground level(target detection algorithm training and selection)are organically separated,so that the modules of the automatic sorting system exhibit a loosely coupled relationship,facilitating the development of subsequent deep learning cloud platforms.At the same time,combined with the automatic sorting system using the depth detection-based target detection algorithm,the hardware selection,model selection and communication mode selection are targeted.For the camera and the robot arm,the design and implementation of the hand-eye calibration and the manipulator control instruction scheme were carried out.Secondly,the image processing module of the automatic sorting system is designed.The tradi-tional image processing algorithm and the deep learning target detection algorithm are organically combined to reduce the use delay of the image processing module of the automatic sorting system.At the same time,because the training of deep learning model requires a large number of data sets,combined with the fact that the amount of data in the manufacturing factory is small,and the data is time-consuming and labor-intensive,the migration learning technology is used to improve the per-formance of the training model under a small number of labeled data sets.Then combined with the actual hardware configuration of the automatic sorting system,select YOLOv3-tiny as the target detection algorithm model configuration of the automatic sorting system.Thirdly,a cloud platform for training deep learning target detection algorithms is designed and implemented.Combined with the actual use scenarios and user requirements of the cloud platform,a Web-server development solution was designed.Completed the development of web front-end pages and server-resident programs.The user can obtain a customized deep learning target detection model through a simple web page operation.Finally,for the image processing module of the automatic sorting system,the design experi-ment verifies the superiority of the configuration selection of the target detection model,and collects the success rate of the automatic sorting system for the workpiece and the success rate of sorting,and verifies the automatic sorting system.Practicality and stability.The performance of the web front-end page and the server background of the deep learning cloud platform was tested.The experimental results prove that the deep learning cloud platform has good stability.The deep learning-based automatic sorting system designed in this paper not only improves the target detection algorithm of the traditional automatic sorting system,but also optimizes the hard-ware selection and model training,and proposes the deep learning cloud platform.The application threshold of the system in the manufacturing industry is greatly reduced.