Unsupervised learning image anomaly detection method based on auto-encoder

Abstract

The invention discloses an unsupervised learning image anomaly detection method based on an auto-encoder. The method comprises the steps: dividing samples into a training sample and a test sample, respectively carrying out preprocessing of the training sample and the test sample, inputting the preprocessed training sample / test sample into the auto-encoder for reconstruction, and obtaining a reconstruction result; respectively calculating reconstruction loss, weighted feature consistency loss, feature discrimination loss and adversarial loss between corresponding layers of an encoder and a decoder in the reconstruction process; carrying out weighted summation on the losses to serve as a total loss function; and finally, calculating the abnormal score of the test sample. Then the abnormal score of each sample is mapped to [0, 1] by using feature normalization, and the area under the receiver operation feature curve is received to serve as an evaluation index. According to the method, theaccuracy of unsupervised anomaly detection is improved by utilizing the auto-encoder and the discriminated potential spatial features, and the method is applied to industries, security and protectionor other unsupervised environments.

Description

technical field [0001] The invention relates to an unsupervised learning image anomaly detection method based on an autoencoder, and belongs to the technical field of visual anomaly detection. Background technique [0002] Anomaly detection is the discovery of data in data that does not conform to expected patterns of behavior, hence the name outlier detection. Anomaly detection is an important field in machine learning, involving practical applications in many fields, such as network intrusion detection, security inspection, medical diagnosis, and video surveillance. In many applications, there is a lack of labeled data to distinguish outliers from normal ones, and they need to be detected in an unsupervised or semi-supervised manner. [0003] An autoencoder is an unsupervised representation learning model consisting of an encoder and a decoder that can be used for data compression and feature extraction. The idea is to map the input data to a latent feature space through...

AI Technical Summary

Problems solved by technology

This type of method basically focuses on the reconstruction error to make the reconstruction result of the sample closer to the input sample, and does not make full use of the characteristics of the latent space of the autoencoder.

Images