An Improved YOLOv8 Algorithm for Infrared Recognition of Train Running Gear Components

Aiming at the disadvantages of low efficiency, high cost and susceptibility to environmental factors in manual inspection of key components of train running gear, an improved YOLOv8 target detection method is proposed. Firstly, a light-weight method is introduced to replace the convolution modules in the Backbone and Neck networks with Ghost convolution, and the VoV-GSCSP module is introduced in the Neck network to replace the original C2f module, so as to obtain a lightweight YOLOv8 model and reduce the model complexity and computational consumption. Secondly, the SE attention mechanism is added to the previous layer of the SPPF layer in the Backbone network to enhance the feature perception of the running parts and further improve the accuracy of the lightweight YOLOv8 model. Experimental results show that compared with the original YOLOv8 method, the FPS of the improved YOLOv8 model reaches 144.83 and the mAP reaches 99.4%, while the number of parameters is reduced by 13.29% and the amount of computation is reduced by 16.05%. The model can effectively detect the key components of train running gear and provide technical support for intelligent detection.