End-to-end semi-supervised image surface defect detection method based on memory information

Abstract

The invention relates to the technical field of image surface defect detection methods, and discloses an end-to-end semi-supervised image surface defect detection method based on memory information. According to the technical scheme, the method is characterized by comprising the following steps of (1) simulating an abnormal sample, (2) freezing an encoder, (3) extracting memory information, (4) fusing multi-scale features, (5) making a space attention map, (6) importing a decoder, and (7) obtaining an abnormal area of an input image. According to the method, the problem that abnormal samples are difficult to obtain under a supervised learning framework is solved, and model training can be completed only by collecting normal samples; meanwhile, the defect that an existing detection method needs high calculation cost in the reasoning stage is overcome, and the real-time requirement of industrial scene defect detection can be better met; meanwhile, the defect that the generalization ability of a model based on reconstruction is too high is overcome, and high anomaly detection precision is achieved.

Description

technical field [0001] The invention relates to the technical field of image surface defect detection methods, in particular to an end-to-end semi-supervised image surface defect detection method based on memory information. Background technique [0002] The intelligent development of the manufacturing industry has put forward higher requirements for the quality inspection of industrial products, and the surface defect inspection of products is a key part of product quality inspection. Surface defect detection is a problem of locating abnormal areas in images. However, in practical applications, due to the low probability of abnormal samples, the small proportion of abnormal areas, and the high cost of data labeling, it is difficult to detect and locate abnormal images through traditional supervised learning. great difficulty. Therefore, the method of surface defect detection based on semi-supervised technology has great advantages in practical application. It only needs no...

AI Technical Summary

Problems solved by technology

Although this type of model requires almost no time consumption in the training phase, it needs to perform highly complex feature matching operations in the inference phase, which causes expensive time costs for the inference phase

At the same time, since this type of model does not train for a specific data set, it directly uses pre-trained parameters for feature extraction, and the extracted features are directly used for anomaly location, so the extracted features are not suitable for anomaly detection tasks.

Images