关键词: 缺陷检测, 压敏电阻, 无监督学习, 空间注意力, 变分自编码器

Abstract: To solve the problems of defect samples collection and poor detection accuracy in the surface defect detection of varistor,an unsupervised surface defect detection method using Deep Convolutional Variational Autoencoder (DCVAE) was proposed.At the pre-processing stage,the original images were processed by background elimination,normalization and image difference technology,and then the difference images were extracted accurately to eliminate the influence of irrelevant data features.To improve the accuracy of defect detection,the proposed algorithm integrated spatial attention mechanism into the variational autoencoder to effectively extract the features of good products.The residual image obtained by subtracting the reconstructed image and the original one was used to highlight the defect area and attenuated the nondefect area..compared with two popular unsupervised defect detection methods,the experimental results showed that the F1 value of the proposed method was improved by at least 7-4%,and the accuracy rate was 98.58%.

Key words: defect detection, varistor, unsupervised learning, spatial attention, variational autoencoder