Research On Surface Defect Detection Algorithm Of Flat Chip Based On Machine Vision

There will be scratches,scratches,holes,package damage and other defects in the packaging process of the chip.In order to ensure the quality of the chip,the defective chip needs to be screened out.At present,the manual screening method is inefficient and difficult to meet the needs of chip production.Machine vision method has the advantages of high speed and high accuracy.It has been widely used in the field of industrial defect detection.The research on chip surface defect detection algorithm based on machine vision has important practical application value.Combined with the characteristics of chip image,this paper studies machine learning detection algorithm and deep learning detection algorithm.The main contents are as follows:1.Aiming at the problem that the image edge pixels of the adaptive median filter algorithm are easy to be misjudged as noise and the detection speed is slow,the sampling window of the adaptive median filter algorithm is improved and the cross window is added to retain the image edge details;When the noise is dense,the mean filtering strategy is adopted to reduce the amount of calculation.In view of the small number of defective products in the chip production process and the time-consuming problem of defect identification and classification of all chips,a method based on Hough transform positioning and image difference is designed to quickly judge whether the chip has defects and reduce the detection time of good chips.2.Aiming at the problem that the existing defect detection algorithms have slow detection speed and are difficult to match the chip production speed,a flat chip target detection algorithm based on machine learning is studied.Firstly,the chip image identified as defective is segmented.Aiming at the problems of high time complexity and slow segmentation speed of two-dimensional Otsu threshold segmentation algorithm,particle swarm optimization algorithm is introduced to speed up the search for the segmentation threshold of two-dimensional Otsu algorithm,so as to achieve the purpose of fast segmentation.Secondly,the feature extraction of the segmented defect image is carried out.For the complex shape of defects on the chip surface,For the problem that a single feature is difficult to describe all the information of the defect,the gray,shape and texture features of the defect image are extracted and fused to comprehensively represent the defect information,and the fused features are reduced to reduce the design complexity of the classifier;Then the wolf swarm algorithm is used to optimize the SVM,and the optimized SVM is used to build dag-svm multi classifier;Finally,according to the algorithm studied in this paper,a defect detection model based on machine learning is designed to complete the classification of chip surface defects.3.Aiming at the problems of high missed detection rate and poor recognition effect of deep learning target detection algorithm for chip surface defect detection,the chip surface defect detection algorithm based on yolov4 is studied.Firstly,according to the characteristics of the chip surface image,K-means++algorithm is used to reset the initialization size of the anchor box of yolov4 to improve the applicability of the detection algorithm;Secondly,the feature extraction network of yolov4 is improved,the attention mechanism is introduced,and the senet structure block is added in the network structure to enhance the feature extraction ability of small targets,so as to improve the detection accuracy of small target defects;Then,in order to make the improved target detection algorithm adapt to the chip production speed,the deep separable convolution network is introduced to replace the convolution core in panet,so as to reduce the amount of parameters and speed up the detection speed of chip surface defects;Finally,the training method is given,and the improved yolov4 chip surface defect detection model is obtained.4.The flat chip image data set is established to train the designed DAG-SVM classifier and the improved YOLOv4 target detection algorithm.Build a defect detection experimental platform to verify the two chip surface defect detection algorithms designed in this paper.Using a unified evaluation index,the performance of the two algorithms is compared,and the algorithm with better performance is selected for flat chip surface defect detection.