Research On Detection Technology Of LED Surface Defect Based On Machine Learning

As Industry 4.0 developing,our requirements for product quality keep growing,especially for the surface quality of Light Emitting Diode(LED).In the actual production line,detection technology of LED surface defect is one of the key steps to detect the LED surface quality,which has become a research hotspot in recent years.Most of the existing researches are only suitable for the single light source detection environment.While in the detection environment of a multi-light source,it is difficult to ensure that the detection speed and accuracy can meet the production requirements at the same time.Based on the detection environment of multi-light sources,this dissertation studies the detection technology of LED surface defects.The main work contents are as follows:(1)This dissertation investigated the research status of detection technology of surface defect based on machine learning,analyzed the problems existing in the current surface defect detection,and explained the research significance of surface defect detection.Furthermore,create a data set of surface defects of LED with multi-light source in multilight source detection environment,including blue light source image and ultraviolet light source image.(2)A classification algorithm of LED surface defects based on the Perceptual Hash Algorithm(PHA)and Support Vector Machine(SVM)is proposed.Firstly,PHA is used to select the extraction method of edge feature suitable for the blue light source image and ultraviolet light source image.Then,the edge features of two kinds of light source images are extracted by suitable extraction methods of edge feature.After that,SVM is used as a classifier for training and testing.Finally,the test results of two kinds of light source images are fused.Through the contrast experiment,it is proved that this method can obtain a better detection effect.(3)A detection algorithm of LED surface defects based on improved FCOS(Fully Convolutional One-Stage Object Detection)is proposed.Firstly,Res Net-50 is used to extract the features of the blue light source image and ultraviolet light source image,and then the extracted features are fused by the convolution layer.Then the feature pyramid algorithm is used to improve the detection ability of small defects.Finally,three loss functions are used to calculate the loss,and the effectiveness of the algorithm is verified by experiments.