Defect Detection of Crane Track Based on YOLOv5-CMT

In light of the challenges such as inaccurate detection results and time-consuming processes in traditional crane track defect detection, an improved crane track defect detection algorithm YOLOv5-CMT is proposed. Firstly, the lightweight GhostNet module is integrated into the main structure of YOLOv5, substituting the residual module within CSP1. This substitution serves to cut down on the parameter quantity and alleviate the computational burden of the network model.; Secondly, the fusion channel hybrid Transformer module empowers the network to focus more intensively on the crucial features of small-sized targets. This leads to an improvement in the network's feature extraction and localization abilities, consequently boosting the precision of detecting small-sized targets; Finally, the traditional GIoU loss function is replaced with the WIoUv3 loss function to accelerate the model's convergence rate and enhance its regression accuracy. Experimental findings indicate that, on the self-constructed dataset, the YOLOv5-CMT algorithm achieves a mAP value of 92.4%. This represents an increase of 3.5 percentage points compared to the original YOLOv5 algorithm. Moreover, the detection speed of YOLOv5-CMT reaches 49fps, demonstrating its suitability for practical application scenarios.