An image anomaly detection method based on a generation antagonism network

Abstract

The invention discloses an image anomaly detection method based on a generation antagonism network, which comprehensively considers the characteristics of a normal image and an abnormal image, and generates an abnormal image and detects the abnormal image through the generation antagonism model. The method comprises the following steps: acquiring a training data set and constructing an implicit space; constructing a generating network to obtain a generating picture set; constructing a coding network to obtain a mapping of a generated picture set on an implicit space; through distinguishing network and detecting network and constructing shared parameter; the generation network, the coding network, the network discrimination network and the detection network constitute a generation antagonism network and perform antagonism training. The invention can fully utilize the relationship between abnormal data and normal data to generate an antagonistic network for detecting abnormal images, thereby effectively determining the normal data boundary and improving the accuracy of abnormal detection.

Description

technical field [0001] The invention relates to the field of anomaly detection, in particular to an image anomaly detection method based on generating confrontation networks. Background technique [0002] Anomaly detection in images refers to the classic problem of images that do not conform to expected normal classes. The characteristic of the data is that there are enough samples of abnormal images, and there are far more normal images in the existing data than abnormal samples. With the rapid development of technology and requirements, anomaly detection appears in different application areas, including security monitoring, traffic monitoring, medical image disease diagnosis and many other applications. [0003] The key to image anomaly detection is to model the distribution of normal and abnormal images, which are usually high-dimensional and complex. In recent years, generative adversarial models have shown promising results in modeling and synthesizing complex pattern...

AI Technical Summary

Problems solved by technology

However, all existing models focus on discovering normal patterns

Images