A Deep Neural Network for Detecting Broken and Scattered Power Lines in Complex Aerial Images

Object detection algorithms constitute a pivotal task in the realm of computer vision, aiming to precisely identify and locate objects within images across diverse scenarios. In power systems, especially in aerial images used to detect wire breaks or scattered anomalies, object detection faces significant challenges. The distribution of wires is often uneven, appearing in different sizes in high-resolution images, and the diversity of fault shapes and sizes makes it difficult for traditional object detection models to recognize accurately. Moreover, the detection of small targets also limits the precision of traditional methods. To address the issue of diversity in fault shapes and sizes in high-resolution images, which makes it difficult for traditional object detection models to recognize accurately, first, we collect fault images of wire breaks and scatter under common scenarios, constructing the “EWF _detection” (Electric wire fault) dataset, and propose a Pyramid Multi-scale Object Detection Model, establishing the Hybrid Spatial Attention Pooling module, HSAP, which significantly enhances the multi-scale feature extraction capability. It not only improves the model's ability to focus on key features, but also suppresses unnecessary features, enhancing the expressiveness of the network, and effectively adapting to complex backgrounds, variable lighting conditions, and different line scales and directions for transmission line detection needs. This improvement makes our model perform exceptionally well in detecting rare faults in wires and accurately locate fault positions. These achievements provide reliable technical support for wire anomaly detection in the power industry and indicate a broad prospect of application in future power scenarios. Our method not only improves detection accuracy, but also maintains efficient detection speed, reducing reliance on a large number of fault line images. These advantages make our model an important technical tool for power system maintenance and monitoring.