Anomaly detection method and device

Abstract

The present invention provides a method and device for abnormality detection. The method includes the following steps: acquiring a target detection image of a workpiece to be detected and a good product image of a good product; acquiring a first feature map according to the target detection image, and obtaining a second Feature map; obtain the cosine similarity according to the first feature map and the second feature map; perform the first abnormality detection on the workpiece to be detected according to the cosine similarity; if it is impossible to judge whether the workpiece to be detected is abnormal, then obtain the first Grayscale image, and obtain the second grayscale image according to the good product image; obtain the first segmentation image according to the first grayscale image, and obtain the second segmentation image according to the second grayscale image; obtain the second segmentation image according to the first segmentation image and the second segmentation image Obtain the index score; perform a second anomaly detection on the workpiece to be detected according to the index score. The invention can accurately detect the abnormality of the workpiece to be detected, has a wide application range, and does not need to consume a lot of manpower, material resources and time costs.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to an abnormality detection method and an abnormality detection device. Background technique [0002] In the field of industrial quality inspection, abnormal detection of defects is a key link. Compared with ordinary defects, the number of abnormal defects is small, and there is also a large gap in shape between abnormal defects and ordinary defects, such as: severe distortion of shape, large area of ​​material shortage, etc. Due to the extremely small number of samples of abnormal defects, this type of defects cannot be learned by the target detection model. [0003] In related technologies, operators of open source libraries such as OpenCV, or GAN (Generative Adversarial Networks, generative confrontation network) networks, or deep learning algorithms are generally used to achieve anomaly detection. However, the above technologies have the following problems: (1) Using O...

AI Technical Summary

Problems solved by technology

Due to the very small number of samples of abnormal defects, it is impossible to learn this type of defects by using the target detection model

[0003] In related technologies, operators of open source libraries such as OpenCV, or GAN (Generative Adversarial Networks, generative confrontation network) networks, or deep learning algorithms are generally used to achieve anomaly detection. However, the above technologies have the following problems: (1) Using OpenCV When the operators of open source libraries implement anomaly detection, the computing resource consumption is large, time-consuming, and the scope of application is small; (2) when using the GAN network to implement anomaly detection, it takes a lot of time to train the GAN network before anomaly detection. And for different faces of the same workpiece, different models need to be trained, which is time-consuming and labor-intensive; (3) When using deep learning algorithms to achieve anomaly detection, a large amount of training data is required, and the accuracy is low

Images