Abstract：With the development of power inspection technology, using drones and infrared thermal imaging technology for inspection has become an important mode of power inspection operations. A target detection method for distribution equipment based on infrared images is proposed to address the issues of low recognition accuracy and difficulty in deploying large model parameters in current network models. Firstly, to address the problem of large parameter count and complex model in the original YOLOv8s model, it is proposed to replace some traditional Conv convolutions with GhostConv convolutions in the backbone network and Neck section to achieve model lightweight. Aiming at the problem of poor small target recognition ability in the original YOLOv8s model, it is proposed to add a small target detection layer to improve the detection ability of small targets. Finally, in response to the problem that the original YOLOv8s model loss function is not conducive to the prediction and regression of ordinary quality samples, a Wise-IoUv3 loss function is used to focus on the prediction and regression of anchor boxes that are difficult to fit during the training process. The research results show that the improved model has an accuracy of 87% which is 4.1% higher than that of the original model, a recall rate of 79.1% which is 3% higher than that of the original model, and a mean average precision (mAP) of 83.5% which is 1.5% higher than that of the original model. The inference speed is 62 ms/sheet. It can be effectively used for component detection in distribution equipment.
2024, Vol. 25 
