Low-Contrast Defects Recognition Using Low-Order Residual Network

Low-contrast defects recognition is a dramatically difficult issue in the field of image recognition. The traditional machine vision method is mainly suitable for defects with obvious feature differences. In recent years, machine learning techniques have been successfully applied to the image analyses, and the deep learning methods provide new solutions for challenging problems in many areas. In this paper, a deep learning network framework based on the low-order residual network is proposed to detect low-contrast defects. Especially, a low-order feature extraction module is designed in order to effectively extract target features with low contrast and small size. The low-contrast watermark defects on complementary metal-oxide-semiconductor transistor (CMOS) camera modules are collected as the test objects to validate the effectiveness of the proposed method. The gray differences between the watermark defects and their adjacent areas are generally several gray-levels. The experimental results show that compared with the existing advanced classification neural network algorithms, the proposed method can effectively identify the watermark defects with a recognition accuracy of over 89%.