Research On Image Segmentation Of PCB Board For Automated Recognition

PCB board(Printed Circuit Board)is widely used in modern electronics industry because of its standardization,light weight,and automatic production,which greatly improves industrial production efficiency.The early assembly of PCB boards is mainly done manually,efficiency and accuracy cannot be guaranteed,and the research of PCB board automated assembly technology has become an inevitable trend.With the rapid development of computer image processing technology,the use of machine vision technology in the automated assembly of PCB boards to achieve accurate and rapid identification of PCB boards has gradually matured.The automatic recognition of PCB boards is mainly based on image processing.The most critical step in the process is the accurate realization of image segmentation.This is of great significance for subsequent target detection,positioning,and improvement of assembly efficiency and accuracy.This paper takes the PCB hole and its elements as the research object.First,an image segmentation quality evaluation algorithm is proposed.On this basis,a variety of segmentation algorithms and image characteristics are matched based on the characteristics of the PCB image area and the image segmentation is completed..Finally,in order to improve the quality of segmentation,deep learning algorithms are used to optimize the region partition to achieve the relatively best segmentation effect.The main research contents are as follows:(1)Considering the light demand of the PCB board in the image acquisition process,and the influence of external interference,an experimental platform with a red ring light source,a light source controller,an industrial camera,an industrial lens as the main components was built.(2)Focuisng on the problem that the existing PCB image segmentation quality evaluation algorithm is scarce and the sensitivity is low,an image segmentation quality evaluation algorithm based on edge characteristics is proposed.Based on three performance evaluation indexes,edge roughness,edge discontinuity and edge dispersion rate of the segmented image,the algorithm extracts the parameter values of the corresponding criterias by the proposed methods,and obtains the quality evaluation results of the segmented images in combination with the specific weight distribution principle.The advantages in accuracy and stability of the algorithm is compared with the existing image segmentation quality evaluation algorithms.(3)The traditional image segmentation algorithm can realize the detection of the target in the process of intelligent manufacturing,but because specificity requirements of image segmentation and regional target detection are getting stronger,while the traditional image segmentation algorithms are not closely related to the characteristics of the image area,and the applicability of the traditional image segmentation algorithms in the industrial application scene is gradually decreasing.Therefore,combined with the characteristics of PCB regional images and traditional image segmentation algorithms,the matching of algorithms based on PCB image area characteristics is carried out.The paper first categorizes the scene and characteristics of a variety of traditional image segmentation algorithms,then partitions the image according to the image texture and outline,then proceeds features' evaluation of the fragmented image,and realizes the match of the characteristics of the segmented area and the traditional image segmentation algorithm,and finally completes the image segmentation process.In order to explore the factors affecting the accuracy of image segmentation,the PCB images are divided into more areas,and the image segmentation quality evaluation is carried out by using the image segmentation quality evaluation algorithm.(4)During PCB plug-in assembly process,jack positioning and recognition are particularly important.In view of the problem that the matching efficiency is low and the accuracy of target detection is not ideal,this paper applies deep learning convolutional neural network framework SSD(single shot multibox detection)to PCB board element detection.By obtaining a large number of original PCB board images,creating semantic label image datasets,feeding them into neural networks,and adjusting the hyperparameters,the relatively best classification accuracy and positioning accuracy are achieved.At the same time,compared with other deep learning network methods,the PCB image detection process of different algorithms is carried out.(5)The corresponding PCB image segmentation software system is built according to the corresponding algorithm and process.By using the corresponding graphical user interface,the process of image acquisition,parameter setting,and final realization of the algorithm can be completed to choose the best algorithm...