Development Of An Intelligent Vibration Sorting System Based On Improved YOLO V7

With the development of the semiconductor industry,the accuracy of electronic component sorting has become increasingly demanding.Due to the small size,high productivity and great variety of electronic components,manual sorting will face issues of inefficiency and error prone.This paper developed an intelligent vibration sorting system based on an improved YOLO v7 algorithm,which was combined by three modules,image acquisition,deep learning network construction and vibration sorting.The data sets of electronic components,target posture detection and positioning,vibration testing and automatic sorting of materials were realized as shown as follows.(1)Researches on image algorithms.A target detection algorithm of YOLO v7 was improved to identify and locate electronic component targets in this paper.The attention mechanism was introduced into the YOLO v7 model by replacing the channel attention mechanism with the attention mechanism Coordinate Attention(CA),in response to the small size and large number of electronic components.The loss function of YOLO v7 was also improved by replacing the CIOU with the Focal-EIOU loss function as the bounding box prediction loss function.With the above improvements,the improved YOLO v7 model was tested on the VOC2007 and DOTA public datasets.It was shown that the average precision mean of the improved model was 91.79%,which was a 4.31%higher than that of the original model.Its detection accuracy was improved by 0.92% and the recall rate was improved by 4.95%.The model was visualised using the DOTA dataset,which showed the improved YOLO v7 model had a better ability in recognizing small targets.(2)Design of vibration scheme and dataset.As the objective of this paper is to sort out workpieces in one position,the workpieces in the vibrating plate need to be turned over and vibrated away in order to obtain the target workpiece information.The images of the vibrated workpiece were first divided into tables,and the proportion of the table occupied by electronic components was used to assess whether the workpiece was fully vibrated.The vibration sequence was then obtained by adjusting the vibration commands to obtain the most suitable parameters with combining the different commands with each other.The camera was calibrated using the Zhang’s calibration and the image acquisition of the SMDs,triodes,chips and resistors was completed in combination with the vibration sequence of the vibrating plate.According to the different positions of electronic components in the vibration plate,the electronic components used in this paper were divided into 11 categories.(3)Development of sorting experiment.Firstly,the hardware platform was built and the information transfer between the systems was completed by calibrating the XYZ module.Then,the vibration software interface,the motion control software interface and the visual sorting software interface were designed with the improved YOLO v7 model deployed.The data set of electronic components was tested on different network models.It was shown that the mAP value of improved YOLO v7 was 99.85% with an improvement of 1.36%,and the FPS value was 30.80.Its speed and accuracy of detection met the expected requirements and was able to accurately locate the workpiece in the target position.Finally,whole vibratory intelligent sorting system was finally tested to complete the sorting of the target workpiece.