Infrared Image Recognition of Power Equipment Based on Improved YOLOv5

Addressing the challenge of achieving both high accuracy and real-time processing in infrared target detection tasks within complex electrical scenarios, a lightweight YOLOv5s-based algorithm is designed for the detection of infrared electrical equipment. Initially, MobileNetV3 is employed to replace the backbone network of YOLOv5s, accelerating detection and facilitating network light-weighting. Subsequently, in the Neck part, the GSConv convolutional module is designed to fuse multi-channel feature information, and the EMA attention mechanism is incorporated, enhancing the recognition capability of electrical equipment while maintaining a lightweight model. Finally, the MpDIou loss function is utilized in place of the original loss function, improving the network's prediction accuracy. The improved model achieves a mAP of 94.58% on a self-constructed dataset, reduces GFLOPs to 3.0, and reaches a detection speed of 96FPS, indicating a 1.29% increase in mAP, an 81.7% reduction in GFLOPs, and a 37.9% improvement in detection speed compared to the original YOLOv5s model. Experimental results demonstrate that the enhanced YOLOv5s model effectively improves the accuracy and speed of infrared electrical equipment recognition while being lightweight, making it easy to deploy and meeting the demands of practical inspection scenarios..