Infrared Thermography-based Temperature Detection of Key Components in Switchgear

With the continuous increase in the number of electric power facilities, relying on traditional manual inspections to maintain the efficiency and accuracy of the power grid is facing greater challenges. In the daily temperature detection of electric power equipment, advanced non-contact infrared thermal imaging technology is widely used in the field of switchgear state monitoring and fault diagnosis. To further improve the efficiency of temperature detection of key components in switchgears, this paper proposes an improved switchgear key component detection model based on YOLOv8. This improved model includes three key improvements aimed at enhancing the detection performance and real-time capability of the detection network. First, this paper introduces Tied Block Convolution to replace the convolutional layer in the original Bottleneck, significantly reducing the model's parameter count, making it more suitable for fast detection scenarios. Second, Large Separable Kernel Attention is introduced in the C2f module, which can capture long-distance contextual information, reducing the loss of critical information in low-resolution infrared images, thereby improving detection accuracy. Finally, this paper adopts the Shape-IoU loss function to replace CIoU, improving the model's accuracy in predicting bounding boxes and further enhancing detection precision. Through contrast experiment, the improved model showed a 1.1% increase in average precision (mAP) and a reduction of 5.5 million parameters compared to the original model. In the comparison of detection results on the test set, the improved model outperformed the comparison algorithms in scenarios where objects were occluded or backgrounds were blurry, further confirming the effectiveness of the improved model and demonstrating significant potential for practical applications.