Artware appearance defect detection method based on unsupervised image segmentation

Abstract

The invention relates to a handicraft appearance defect detection method based on unsupervised image segmentation. The method comprises the following steps: generating a pre-training defect detection model; using the pre-training defect detection model to extract and establish a feature map of the normal product image; performing picture feature extraction by using a pre-training defect detection model, generating a low-rank feature matrix, and calculating an abnormal score value of each pixel point of the image based on a mahalanobis distance; and selecting all the pixel points of which the abnormal score values are greater than a preset threshold value as a defect segmentation result of the product image. The method has the beneficial effects that the high-efficiency and accurate segmentation of the defect area in the product image can be realized without an abnormal sample image by judging the separation degree of different pixel points of the image deviating from the normal sample.

Description

technical field [0001] The invention relates to the field of handicraft appearance defect detection, more specifically, to a handicraft appearance defect detection method based on unsupervised image segmentation. Background technique [0002] With the improvement of consumers' requirements for the quality of industrially manufactured products, the surface defect detection of handicrafts has become one of the links that manufacturers pay more and more attention to. The traditional method of distinguishing product defects by the human eye can no longer meet the increasingly stringent inspection needs. In addition, the subjectivity of manual inspection, as well as the constraints of the proficiency, efficiency and cost of new and old quality inspectors, make product appearance quality inspection a problem for related manufacturing companies. With the development of vision hardware and artificial intelligence technology, more and more automatic defect detection algorithms based ...

AI Technical Summary

Problems solved by technology

However, due to the influence of factors such as equipment and technology, the types of defects on the product surface are often varied, such as stains and damages in the production of fabrics, scratches, cracks, and unevenness on metal products. Various types of defects make the traditional It is difficult for machine vision algorithms to complete the modeling and migration of defect features, the reusability is not great, and it is required to distinguish working conditions, which will waste a lot of labor costs

In recent years, deep learning has achieved very good results in feature extraction and positioning. The industry has begun to introduce deep learning algorithms into the field of handicraft defect detection, but the deep learning model requires a large amount of defective product image annotation data to train the detection model, and Some special field applications (such as precision medicine, customized products, etc.) cannot provide enough accurate labeling data due to their particularity, and the lack of enough abnormal samples makes it difficult to use supervised learning models to achieve industrial quality inspection

Images