Defect detection method based on variational automatic encoder

Abstract

The invention discloses a defect detection method based on a variational automatic encoder, and the method comprises the steps: collecting a product data set, and dividing the data set to obtain a training set and a test set; generating image blocks disorganized according to labels Si from the pictures in the training set; constructing a variational automatic encoder network, and sending the disordered image blocks into the encoder network to obtain potential features; constructing a decoder network, inputting potential features output by an encoder into a decoder, assisting in solving a puzzle, capturing global and local information so as to reconstruct a high-resolution image, and carrying out back propagation on a model by combining a preset loss function to update network parameters so as to obtain a trained model; and testing on the test set to complete defect detection. Experimental results show that the model has excellent generalization ability and defect detection ability.

Description

technical field [0001] The invention belongs to the field of defect detection, in particular to an industrial defect detection method based on a variational automatic encoder. Background technique [0002] Defect detection is a very important link in manufacturing. It can monitor the quality of products, so that workers can find problems in time, so as to screen out substandard products and improve technology, which penetrates into industrial defect detection in each production step. It plays a vital role in reducing production costs, improving production efficiency, and improving product quality qualification rates. However, the traditional artificial naked eye detection has various limitations and cannot meet the needs of high-speed and accurate detection. [0003] Deep learning algorithms are widely used in industrial defect detection. However, some traditional methods, such as One-class SVM and SVDD, rely on manually labeled datasets, and when directly applied to high-d...

AI Technical Summary

Problems solved by technology

However, the approximate posterior distribution is often too simple and differs from the true posterior distribution

Images