COMPUTER VISION-BASED ANOMALY DETECTION METHOD AND APPARATUS, AND ELECTRONIC DEVICE

摘要

A computer vision-based anomaly detection method and apparatus, and an electronic device. The method comprises: S110: dividing, according to different region features of a target image, the target image into at least two feature regions, and forming training sets by using the feature regions corresponding to each target image; S120: selecting a generative adversarial network (GAN) as a network model used, and training the GAN network model by using the training sets of different feature regions to obtain GAN network models corresponding to the different feature regions respectively; and S130: performing, when executing anomaly detection, the same feature region division on a target image to be detected, respectively inputting different feature regions of said target image into the corresponding GAN network models to obtain a generated image, and performing pixel value-based difference detection on the generated image and said target image.