A Method for Steel Surface Defect Recognition Based on Deep Learning and Receptive Field Block

Abstract

Surface defects are an important factor affecting the steel quality, and their classification is crucial for detecting the steel surface defects and analyzing the causes of the damage. Recently, computer image technology has achieved remarkable recognition rates in image classification tasks. And the traditional steel defect image detection algorithm due to the low contrast between background and characteristics, can not meet the detection requirements. Although the accuracy has improved, there is still a great potential for optimization. This paper deeply investigates the image classification algorithm and proposes a residual network based on the optimization initial module and Receptive Field Block(RFB). The entire network is optimized based on a residual network model and establishes a fast connection between the network modules. Residual structure is suitable for deep network, and RFB module is helpful for extracting detailed features, enhancing feature discrimination and improving network quality. Experimental results show that compared with some classical methods, this method can effectively improve the accuracy of steel surface defect classification.