U-Shape Network for Chip Surface Defect Detection

Currently, deep learning methods and models have achieved significant progress in image segmentation and have been successfully applied in various practical applications. However, common methods and models are usually only suitable for situations where the segmentation target occupies a large portion of the input image and has a very obvious difference with the background. Additionally, they require high-quality datasets with balanced categories and large quantities. In the field of industrial defect detection, defects often occupy a small area of the image and are difficult to distinguish from the background. The edges of defects are often blurry, and the differences in shape and color between defect categories are relatively small. Moreover, in industrial settings, acquiring a large amount of defect data and labeling it is very expensive. Therefore, the available training data in practical applications is often small and imbalanced. These characteristics make it difficult to train common image detection models on industrial defect datasets. To address these issues, we propose a novel network architecture based on the Unet [25] segmentation model. This architecture effectively analyzes and utilizes image features extracted by commonly used backbone networks. As a result, the model can detect defects that closely resemble the background and occupy small regions. We also introduce a practical method for training segmentation models on low-quality datasets. Compared to existing segmentation models commonly used in image segmentation, our proposed architecture demonstrates better performance across various evaluation metrics. This research aims to provide a viable solution for industrial defect detection.