At present, unmanned aerial vehicle (UAV) is usually used to capture the ground images for power inspections, then transmit the images to the fixed ground station for analysis. That method is not conducive to the rapid positioning of key parts or timely treatment. At the same time, the automatic sensing of power inspections mainly adopts the targets detection and classification algorithms based on deep learning, which has a large amount of calculation, and the processor installed in the UAV terminal is difficult to achieve the effect of real-time detection. In order to improve the accuracy and real-time of key targets detection and classification of power facilities in the inspection process, a lightweight edge computing framework AirNet is proposed. In AirNet, simple linear iterative clustering (SLIC) algorithm and direct convolution method are used to optimize the UAV input image to simplify multiple granularity feature information and improve the accuracy of the algorithm. Real time intelligent analysis of the algorithm model is carried out in UAV terminal, and the key parts of transmission towers, houses and vehicles are selected for test. The results show that the algorithm can achieve 64 ms detection speed and 85% accuracy on Huawei Atlas 200 chip equipment.
Read full abstract