Lightweight component defect detection method based on single sample learning

Abstract

The invention relates to a lightweight part defect detection method based on single sample learning. The defect detection of the component is realized by utilizing a machine vision technology and using a component image acquired by the existing optical detection equipment, including qualitatively judging whether the component has a problem or not and giving a possible area with the defect. The method provided by the invention does not need manual sample labeling, is low in data acquisition cost, and can adapt to different illumination conditions.

Description

technical field [0001] The invention belongs to the technical field of machine vision detection, and in particular relates to a defect detection method for lightweight components based on single-sample learning. Background technique [0002] The purpose of industrial quality inspection is to detect objects with heterogeneous features in image data, also known as defect detection. Defect detection tasks play an important role in industrial production, traffic detection, medical diagnosis and other fields. For example, in the process of industrial production, detecting unqualified products in products is a defect detection task. In the traditional industrial production process, in order to complete the defect detection task, it often relies on the excellent abnormal detection ability of humans, but this requires experienced skilled workers, and different operators have different perceptions of defects, resulting in uneven detection results. low efficiency. [0003] In the c...

AI Technical Summary

Problems solved by technology

In the traditional industrial production process, in order to complete the defect detection task, it often relies on the excellent abnormal detection ability of humans, but this requires experienced skilled workers, and different operators have different perceptions of defects, resulting in uneven detection results. low efficiency

[0003] In the current defect detection scenario, the traditional method of using image labels to learn image features and then find general features of images is difficult to achieve

Because in the process of industrial production, a large number of products are non-abnormal and non-defective, and defective and abnormal substandard products are very rare, which leads to a small number of images with abnormalities used for training, and these abnormalities The defects of the target are very diverse, difficult to describe, and cannot be made into pixel-level labels, so that traditional network training often cannot obtain satisfactory results

At the same time, the existing optical inspection equipment needs to be equipped with a supplementary light device, and the traditional machine vision inspection method is easily affected by the light. Usually, the visual inspection method trained under a certain lighting condition is difficult to apply to other lighting conditions.

[0004] In summary, affected by factors such as few defect samples, difficulty in describing defects, and changes in illumination, the image data of components acquired by automatic optical inspection equipment is complex and diverse, and the existing component defect detection algorithms are difficult to meet the actual needs of industrial quality inspection.

Images