Research On Identification And Location Of High-value Electronic Components For Disassembly Of Waste Circuit Board

The rapid increase in the annual scrap of electronic products in China has led to a large amount of environmental pollution and waste of resources.It is urgent to develop efficient disposal and utilization technologies and methods to realize the recycling of metal materials on circuit boards,especially those that are rich in gold,silver and other rare metals.Precious metal electronic components,if they can be dismantled separately and Centralized purification process,will effectively improve the disposal efficiency,and will also reduce the amount of chemical reagents in subsequent purification and reduce environmental pollution.However,due to the many types of electronic components on the circuit board,large target scale range,and complex background,the traditional identification and positioning algorithms have many problem about a large amount of calculation,low accuracy,and poor portability of the algorithm during the automatic identification and positioning of electronic components,which makes the circuit board intelligent disassembly project difficult.In view of this problem,this paper proposes a method for automatic identification and positioning of electronic components of discarded circuit boards.By setting up a deep learning network,constructing an electronic component detection and recognition model,and combining camera calibration methods,automatic identification and positioning of electronic components of discarded circuit boards is achieved,and the effect of identification and positioning is significantly improved.The main work of this paper is as follows:1)Research and experimental analysis of several mainstream deep learning model algorithms(SSD,Faster RCNN,YOLOv3).Under the Ubuntu system,use the Python programming language and tensorflow deep learning framework to build a network,and use the homemade electronic component data set and some KITTI data sets to construct the corresponding detection and recognition models,and do comparative experiment analysis.The results show that the SSD network model detection speed Fastest,but obviously insufficient for small target detection;Faster-RCNN detection accuracy is high,but the detection speed is slow;YOLOv3 network model has the best comprehensive detection performance,and its detection accuracy rate reaches 77.32%,but at the detection scale Larger sockets will still have low fitting and re-check,and problems such as missed tests and low scores will occur when detecting small IC chips and crystals.2)Network optimization and experimental analysis.Aiming at the problems of the YOLOv3 network model algorithm in detecting the electronic components of abandoned circuit boards,two different ideas were used to optimize the structure,parameters,and training of the YOLOv3 network.Firstly,the YOLOv3-Darknet61 with 4-scale detection combined with Res-Net connection structure and the YOLOv3-Darknet50 electronic component detection and recognition network with 4-scale detection combined with Dense-Net connection structure were set up respectively,and then using the circuit board electronic components were made by self and partial KITTI data set combining with a series of optimized training methods to construct corresponding target detection and recognition models.Through a large number of experiments,it has been proved that the performance of the optimized electronic component target detection model is superior to the original YOLOv3 network model.The rate reached 81.01%,and YOLOv3-Darknet50 not only achieved a detection accuracy rate of 79.7%,but also reduced the model parameter resource consumption by 98.93 MB compared to YOLOv3 during the training process.3)Positioning and system design of electronic components.According to the imaging principle of the camera,after the camera calibration and accuracy verification tests are implemented,combining with the deep learning target detection position border regression to obtain the coordinate parameters of the target in the image,and the actual spatial position coordinates of the target are obtained through coordinate transformation.Finally,the function and performance requirements of the disassembly and recycling of electronic components of waste circuit boards are analyzed,and a visual system for electronic component identification and positioning is designed.In combination with the electronic component identification and positioning algorithm developed in this paper,a simple and practical software system was developed.