Transmission Line Anomaly Detection and Real-Time Monitoring System Combining Edge Computing and EfficientDet

Accurately identifying aberrant issues on transmission lines is crucial to guaranteeing the power system operates safely and steadily. Transmission lines are a crucial part of the power system. In response to the current issues of low recognition accuracy and low image contrast in abnormal detection of power transmission lines. This paper develops a transmission line anomaly detection and real-time monitoring system combining edge computing and improved Efficient Det based on the characteristics of foreign object images. Firstly, by collecting abnormal images of transmission channels, the Retinex algorithm is used to enhance the input images. Secondly, in order to improve the speed of model detection, the convolution operation process for obtaining data features based on Ghost lightweight module is reconstructed. In the target recognition part, an improved EfficientDet algorithm is adopted as the main body, and the aspect ratio of anchor boxes in the algorithm is optimized using K-means clustering algorithm. At the same time, a gradient equalization mechanism is added to the loss function. The experimental results show that the proposed method effectively improves recognition accuracy and recall rate, achieving over 88% on different types of transmission lines. The ablation experiment shows that adding K-means clustering algorithm and gradient equalization mechanism can significantly improve recognition accuracy.