Appearance defect detection method and device for industrial products

Abstract

The invention relates to the technical field of industrial quality inspection. In order to solve the technical problem of troublesome detection of industrial product appearance defects and poor detection effect, a method and device for detecting industrial product appearance defects are provided. The method includes: segmenting the image of the industrial product to be inspected into multiple sub-regions; according to the similarity between adjacent sub-regions, the sub-regions are merged to obtain multiple sub-graphs to form a sub-atlas; the sub-atlases are clustered to divide multiple sub-graphs in the sub-atlas into Multiple size categories; adjust the submaps in each size category to the corresponding fixed size; input the adjusted multiple submaps into the convolutional neural network in sequence to output the corresponding feature maps, and flatten each feature map The corresponding feature vectors are obtained to form a sample set, in which the convolutional neural network includes a global average pooling layer; spectral clustering is performed on the sample set, and the samples in the sample set are divided into defect categories and good categories to judge industrial products to be tested Whether the image has cosmetic defects.

Description

technical field [0001] The invention relates to the technical field of industrial quality inspection, in particular to a method for detecting industrial product appearance defects, a device for detecting industrial product appearance defects, a computer device and a non-temporary computer-readable storage medium. Background technique [0002] Industrial intelligent quality inspection has been widely used in recent years and has attracted great attention. At present, in industrial intelligent quality inspection, commonly used methods are based on traditional image processing algorithms, deep learning algorithms, and combinations of these two algorithms. Traditional image processing algorithms are mainly effective for fixed position defects, structural defects, and defects that need to be measured, but for general appearance defects, such as scratches, crushes, bruises, etc., the effect is not ideal or even completely invalid . The deep learning algorithm performs well in ap...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above technical problems, the present invention provides a method and device for detecting appearance defects of industrial products, which realizes the detection of appearance defects of industrial products through an unsupervised learning method, without manual labeling, and obtains by dividing and merging images to be detected The data set, combined with the advantages of spectral clustering itself, not only has lower requirements on the amount of data, but also has a better detection effect, especially for new forms of defects.

Images