Deep Learning Network for Vision-Based UAV Recognition

Unmanned aerial vehicles (UAVs) could be used carelessly or intentionally, which poses a serious risk to aviation safety. This makes automated UAV detection and tracking a crucial component of aerial security systems. Radar, radio frequency, and visible- and thermal-infrared imaging are often used technologies for UAV detection. Deep neural networks (DNNs) for the image-based object recognition have recently made advancements that make it possible to leverage For this detecting and tracking challenge, visual data. Through the encoder-decoder structure network, the suggested technique first estimates the distribution line area. The interference is eliminated using sampling clustering and image processing operations. Finally, based on the discovered distribution line, the UAV direction of tracking of the power line is determined. The approach can achieve a frame inspection speed of over 78 ms, a heading error of deviation range of (−1.53°, 1.36°), and a tracking rate of almost 99%. The results of the network and inspection technique dataset test demonstrate that the approach presented in this research may be used to check UAV distribution lines efficiently, swiftly, and accurately.