Dynamic Focal Network: Advancing Real-world Electric Pylon Detection Using High-Resolution Satellite Remote Sensing Image

Due to its greater speed and efficiency, the utilization of high-resolution satellite remote sensing images for detecting electric pylons represents an emerging and significant research area within power inspection and emergency response. However, the current methods encounter the following three challenges: large variations in pylon size, complex backgrounds, and data set setting divorced from practical application scenarios. In this paper, we propose the DFN (Dynamic Focal Network) for real-world electric pylon detection. Firstly, we apply the Dynamic Convolution in detection network to improve the applicability of the model to different size pylons. Secondly, we use a Focal Mechanism to improve the detection model's ability to cope with complex background interference. Finally, we propose a new satellite remote sensing image dataset for electric pylon detection, which is more suitable for real application scenarios. We have carried out experiments on several public data sets, such as EPD, WHD, etc., and achieved a precision of 91.5%, 86.5% recall and 91.5% mAP@0.5, verifying the effectiveness of this method.