Steel Surface Defect Detection Based on Improved MASK RCNN

Abstract

The defect detection of steel is an important process to ensure the quality of steel. The traditional detection methods have low efficiency and poor accuracy. With the development of deep learning and computer vision technologies, this paper proposes an improved Mask RCNN model for steel defect detection. The feature extraction network of Mask RCNN is replaced by a more robust EfficientNet, the improved BiFPN structure is combined with EfficientNet to extract features of different scales, and a CBAM module is added to the mask branch to improve the quality of mask prediction. Experiments on the Severstal steel surface defect dataset show that the improved method not only significantly improves the accuracy of the model, but also greatly reduces the model parameters.