A Lightweight Object Detection Algorithm for UAV Power Inspection

Deep learning algorithms have shown excellent performance in many fields, leading to their application in tasks such as fault detection, fault recognition, and operational safety monitoring in the power industry. With the advancement of technology, more and more edge embedded devices are deployed in power scenarios to handle data collection and processing. However, these devices typically have limited computing resources and storage capacity. The large number of parameters, computational requirements, and memory usage of most neural network models severely hinder their deployment on edge devices. This article proposes a lightweight drone detection algorithm based on YOLOv8. We (1) introduce a decoder module during the training process to improve detection accuracy through additional supervision information, which can be removed during the model inference process to avoid increasing inference workload; (2) Combining attention mechanisms to help the network focus on key regions and features, thereby improving performance with minimal additional computation. Our lightweight model can be effectively deployed on resource constrained embedded devices, providing robust detection results.