Defect Of PCB Image Detection Based On Deep Learning Method

In the process of industrial production,defect detection plays an increasingly important role in the continuous and large-scale automatic production process of modern industry.In the PCB production process,surface cracks,scratches,stains,pits,holes,burrs,bubbles and other defects affect the product performance,integrity,serious defects and even affect the safety of use.Therefore,it is of great significance to detect defects on PCB surface.At present,the main methods of PCB defect detection are still manual detection,mechanical detection,ray detection,etc.among the above several detection methods,manual detection is the most traditional and the least efficient.Due to the labor intensity of workers,the reliability of detection depends on many subjective factors,so the accuracy of detection is relatively low;The mechanical inspection usually uses the contact type inspection.In the inspection process,the position of the inspected products needs to be adjusted,so the efficiency is relatively low.The ray inspection can distinguish the defects of the products very well,but because of its complex equipment structure and high cost,it is not widely used in the actual production process.At present,with the development of PCB manufacturing industry,there are more and more types of defects on its surface,and the demand for detection is more and more complex.Therefore,several commonly used detection methods can neither meet the rapid production process of modern PCB nor the quality monitoring in the production process.Therefore,imagebased defect detection based on deep learning method is widely used in PCB defect detection.In this paper,a defect detection platform based on deep learning method is designed,which can quickly and accurately identify the defects and types of PCB in the production process.Firstly,this paper investigates the application of PCB detection system and the current research status,and on this basis,the lens and light source of high-speed camera are selected,and the hardware platform of the system is built;then,the collected image data are marked and classified to eliminate a small amount of fuzzy and difficult to identify image data,and at the same time,and the dataset is transformed into a coco dataset that can be recognized by the detectron 2 platform.Secondly,the image of PCB is identified and detected,and a decron platform is built on the server side to extract robust or invariable features of the image in view of various changes of defects in PCB,such as position,brightness and so on.Then the image features are extracted by convolution neural network,and the defect category is determined.Then the defect location is determined and marked by boundary regression algorithm.Now,many deep learning algorithms can effectively achieve this goal.Among them,RCNN series algorithm is the most widely used target detection algorithm based on deep learning.This paper will also take this kind of algorithm for target detection and analysis.Finally,on the basis of the above research,a software platform for PCB image-based defect detection is developed.The platform applies segmentation view,integrates the above algorithms,and can classify and identify the defects in the input PCB images.Through the test of the platform,the practicability and reliability of the platform are verified,which lays the foundation for the platform to carry out defect detection in the actual production environment.