Enhanced Yolov4 Power Equipment Surface Detection Method Based on Deep Residual Network and Collaborative Attention Mechanism

The existing technology still lacks effective means to detect electrical characteristic problems, such as icing, corrosion, oil leakage, cracks, dirt, and other electrical characteristic defects. How to accurately detect internal defects of electrical equipment with uneven appearance under high voltage conditions where the characteristics and morphological changes are not obvious is a topic worthy of research. This article investigates a method for detecting infrared image defects in power equipment using YOLOv4 (You Only Look Once, Version 4) algorithm. Our goal is to improve the automation detection capability of power equipment and ensure the safe operation of the power grid. We use CSPDarkNet53 (Cross Stage Partial Network DarkNet53) as the basic backbone network and combine it with CA (Coordinated Attention) attention mechanism to improve the model's attention to key features. This experiment used an infrared image dataset from real power equipment inspections, including five types of defects: ice, rust, oil leakage, cracking, and dirt. The model was trained and evaluated by adjusting the learning rate, batch size, and data augmentation strategy. The experimental results showed that the configuration with a learning rate of 0.001, batch size of 16, and data augmentation achieved the best performance, with an accuracy of 98.5%, a recall rate of 97.2%, an F1 score of 97.8%, and an average detection time of 0.045 seconds. The method explored in this article achieves efficient and accurate detection of defects in power equipment in complex environments, providing an effective solution for automatic detection of power systems.