Defect Recognition of Traction Wire Rope Based on Improved EfficientNet Network

This paper proposes an improved EfficientNet network-based method for recognizing defects in traction wire ropes, aiming to address the current issues of low efficiency and poor robustness in defect recognition. Based on the EfficientNet network, we introduced the ECA module instead of the original SE module and improved the loss function. The experimental results demonstrate that the improved EfficientNet network achieves a classification accuracy of 99.3% for defect recognition of traction wire rope, which is 1.1% higher than that of the original network. This improvement better meets the engineering application of defect recognition of traction wire rope.