Research On Digital Character Detection System Based On Machine Vision

In recent years,with the proposal of "Made in China 2025" plan,all industries in our country are accelerating the pace of intelligent manufacturing and industrial transformation and upgrading.At present,although the degree of automation and intelligence in the process of electronic products processing is also gradually improving,but in the process of quality inspection,it still needs to rely on a lot of manpower to achieve,which causes the waste of labor and increases the cost of products.Aiming at the problem of digital tube detection commonly used in electronic products,I propose a scheme to realize the intelligent detection and recognition of digital tube by using machine vision technology.Recently,although there are more and more researches on image target detection and many practical applications,but most of them rely on high-performance computer,GPU,high-precision image sensor or cloud computing.Due to the high cost,such a scheme is ususlly unacceptable for the cost sensitive electronic industry.Therefore,I choose to research in the field of micro embedded,which has a significant advantage in cost performance.I try to combine image processing technology with embedded development to improve the ability of edge computing,so that the micro embedded system can realize the location,segmentation,character recognition.Finally,the detection system is completed.1.In this system,I design an embedded vision detection unit with STM32F7 as the main controller,and then complete the camera calibration according to the machine vision related principle.Because ABB Robot is used as the actuator of the system to complete the positioning and screening task of the digital tube module,it is necessary to determine the corresponding relationship between the robot coordinate system and the visual coordinate system by hand-eye calibration method.Then the basic experimental environment of the system is completed.2.This system uses the embedded visual detection unit to collect the digital characters,and then uses the traditional image processing algorithm to segment the digital characters in the image.After comparing the accuracy and portability of threading algorithm,template matching algorithm,three-layer full connection neural network algorithm and classic lenet-5 convolution neural network algorithm,lenet-5 algorithm is selected and optimized and deployed.