Hot-Rolled Steel Strip Defect Classification Based on ResNet34 Model

Currently, the classification accuracy of surface defect images of hot-rolled steel strips is not high. To improve the accuracy and rate of surface defect classification of hot-rolled steel strips, we propose a surface defect classification method for hot rolled steel strips by embedding the attention mechanism into ResNet34 model. The experiment will classify 6 kinds of hot-rolled steel strip defects. To better learn relevant information and extract relevant features, a CBAM attention module is embedded on the basis of ResNet34. The module can learn the characteristic information of channel and space better, so as to enhance the network performance. Results of the experiment indicate that this approach can enhance the classification accuracy and has a quicker convergence rate. This method has a certain degree of application value.