SSDD-Net: A Lightweight and Efficient Deep Learning Model for Steel Surface Defect Detection

Industrial defect detection is a hot topic in the computer vision field. At the same time, it is hard work because of the complex features and various categories of industrial defects. To solve the above problem, this paper introduces a lightweight and efficient deep learning model (SSDD-Net) for steel surface defect detection. At the same time, in order to improve the efficiency of model training and inference in the XPU distributed computing environment, parallel computing is introduced in this paper. First, a light multiscale feature extraction module (LMFE) is designed to enhance the model’s ability to extract features. The LMFE module employs three branches with different receptive fields to extract multiscale features. Second, a simple effective feature fusion network (SEFF) is introduced to be the neck network of the SSDD-Net to achieve efficient feature fusion. Extensive experiments are conducted on a steel surface defect detection dataset, NEU-DET, to verify the effectiveness of the designed modules and proposed model. And the experimental results demonstrate that the designed modules are effective. Compared with other SOTA object detection models, the proposed model obtains optimal performance (73.73% in mAP@0.5) while keeping a small number of parameters (3.79M).