Surface Defect Detection Method of Hot-Rolled Steel Strip Based on Improved SSD Model

EasyChair Preprint no. 4681

Abstract

In order to reduce the influence of surface defects of hot rolled steel strip on product performance and appearance, a surface defect detection method combining attention mechanism and multi-feature fusion network is proposed. This method takes the traditional SSD network as the basic framework, selects RESnet-50 as the feature extraction network, and integrates the low-level features and the high-level features to complement each other, so as to improve the accuracy of detection. In addition, the channel attention mechanism is introduced to filter and retain important information, which reduces the network computation and improves the network detection speed. The experimental results on NEU-DET data set show that the accuracy of this method for surface defect detection of hot-rolled steel strip is obviously higher than that of traditional SSD network, and it can meet the real-time requirements of industrial detection.

Keyphrases: Channel attentionmechanism, feature fusion, hot rolled strip, ResNet-50, SSD network, surface defect

@Booklet{EasyChair:4681,
  author = {Xiaoyue Liu and Jie Gao},
  title = {Surface Defect Detection Method of Hot-Rolled Steel Strip Based on Improved SSD Model},
  howpublished = {EasyChair Preprint no. 4681},

  year = {EasyChair, 2020}}