Defect Detection Of Spark Plug Image Based On Convolutional Neural Network

In recent years,with the rapid improvement of computer parallel computing capabilities,deep learning has become a key technology in the smart industry in the new era.The image classification,face recognition,security monitoring,and speech and text recognition industries have achieved remarkable results and reached commercial use.The industrial spark plug is the key to the ignition system of automobile engines,and its detection is particularly important.There are many kinds of target detection algorithms.Among them,there are two target detection algorithms combined with deep learning.One is a two-step algorithm for comprehensive region suggestion and feature extraction,and the other is only a classification regression algorithm.Spark plug defects are divided into welds and weld defects.The target is small,the shape is different,and the data is complicated.The traditional spark plug defect detection relies heavily on manual detection,the detection task is heavy,and it is not easy to find with the naked eye.The detection error caused by human experience detection is very large,and there is no unified standard.Modeling with SVM support vector machine,random forest and other techniques is difficult,with low detection accuracy,high false detection rate,and low detection speed.It is not suitable for small target multi-morphology target detection.In view of the above detection difficulty,this article carefully studies the spark plug image,combined with the deep learning convolutional neural network,proposes a target detection method based on Faster R-CNN(Fast Area Convolutional Neural Network)comprehensive region recommendation and feature extraction spark plug The image is inspected for defects.For the spark plug welding gap image,the image rotation method is used to enhance the data set,and the convolution neural network is used to accurately locate and extract the weld seam area,and record the position of the welding gap in the spark plug image.For the spark plug weld image,the data is expanded using the lateral shift method,and then the original image of the spark plug is labeled using LabImg software.After the neural network training platform is built,the convolution neural network is used to accurately locate and extract the weld area,and the extraction results are separated Welding out a small area of??the weld,using a graph theory method to separate the weld from other areas,using the iteration threshold to extract the weld,and then by thinning the straight line,the spark plug weld is a vertical line,converted by pixels and distance Out of its height.In this paper,the accuracy rate of industrial X-ray weld defect detection is 93.7%,and the rate of misjudgment and missed judgment is low,which effectively shortens the inspection time,gives the length of the weld seam,and improves the efficiency of industrial inspection.