Self-Supervised Learning For Image Surface Anomaly Detection And Location

At present,anomaly detection methods based on deep learning are widely used in actual mass industrial production.These detection methods can be divided into supervised learning and self-supervised learning according to the quantity variance of labels.Since the self-supervised learning method only requires to employ a little number of normal samples for the model training,it can perform anomaly detection on the surface of the object,and it has the advantages of stronger versatility,low data cost and short training time.So it is more suitable for application in the industrial production process of product surface anomaly detection.In order to implement the smart contactless object surface anomaly detection and location,this paper has carried out the research of image surface anomaly detection and location based on self-supervised learning.The specific contents are as follows:(1)Aiming at the problems of low manual detection efficiency,high cost,insufficient automatic detection marker samples and high missed detection rate in ceramic tile surface anomaly detection,a detection model based on self-supervised distribution and enhanced contrast learning is proposed,which can realize the detection and location of common ceramic tile surface anomalies without a large number of defect samples.Negative samples are generated by sample expansion,and the distribution enhancement contrast learning is used to improve the data irregularity and expand the sample distribution,so as to reduce the uniformity of the contrast representation,so that the representation feature distribution is consistent with the classification target.Based on the self-supervised learning representation,a class of classifier is constructed to achieve accurate anomaly detection and location.The experimental results show that under the criterion of anomaly detection standard evaluation metric(AUROC),the anomaly detection rate of this method is 3.71 % and 2.74 %higher than that of the other two advanced methods,respectively.The anomaly location rate is increased by 1.22 % and 4.01 % respectively.(2)Aiming at the problems of insufficient labeled samples and high missed detection rate in common textured surface anomaly detection,the paper designs a self-supervised learning model based on masked Autoencoder,which can realize accurate detection and location of anomalies without providing mass anomaly samples.Autoencoder is widely used,but it is difficult to detect and locate anomalies by reconstruction error due to its strong generalization ability reconstructed anomalies with small errors.Then,masked reconstruction method is proposed to reduce the generalization performance.First,each input image is masked to obtain multiple masked input images which are sequentially reconstruct by the Autoencoder.Second,these reconstructed images are complementarily masked and recombined to obtain the final reconstructed image.Finally,anomaly detection and localization are achieved by evaluating the reconstruction error between the input and reconstructed image.The experiment results indicate that the anomaly detection rate of this method is 93.05 % and the anomaly location rate is 94.82% under the anomaly detection standard metric,It has strong universality and wide application value.(3)Pyqt5 is used to develop a product surface anomaly detection system based on self-supervised learning and applied to the surface detection of industrial products.The system constructs each function window of the system by using C + + language,Visual Studio2019 and Qt Creator 5.15 software development platform,and adds various appearance controls to realize the detection and location of static image anomalies,real-time camera capture video stream anomalies and other functions.The experimental results show that the system designed in this paper can run normally and stably the surface anomaly detection task of different objects.In summary,the self-supervised anomaly detection and location technology proposed in this paper can not only obtain the types of defects in the image,but also indicate the size and specific location of the defects.This information can assist in the quality grade evaluation,and has broad market application prospects.