Mobile Deployment Method for Defect Recognition Algorithms in Power Transmission Scenarios

This paper presents a research study on the integration of artificial intelligence (AI) method in digital power grids to achieve comprehensive observability, precise measurement, and high controllability. The study addresses the challenges associated with handling large amounts of video and image data collected from online and mobile monitoring terminals. To overcome limitations in network bandwidth and cloud computing resources, the research proposes the transplantation of cloud-based algorithms to edge devices, specifically unmanned aerial vehicles (UAVs). However, the limited computing power of edge chips poses efficiency and endurance issues. To address this, the paper introduces model compression methods, including model structure optimization, pruning, channel compression, and distillation, to reduce the size of neural network models without compromising accuracy. The compressed models are deployed on UAV edge devices within a cloud-based system, enabling real-time data processing and analysis. The proposed approach enhances the efficiency and effectiveness of digital power grid systems, enabling comprehensive observability, precise measurement, and high controllability in power grid operations.