Research of Methods for Processing Analog Video Signals of FPV Drones for Identification Systems Based on YOLOV9

Abstract

The article addresses the problem of reduced performance of identification systems based on YOLOv9 when using analog video signals from FPV drones operating in the 5.8 GHz band. An analysis of the impact of noise, transmission artifacts, and digitization processes on input data quality and object detection accuracy is conducted. An approach to improving the informativeness of the video stream is proposed through the application of signal preprocessing methods and adaptation of the neural network to analog channel conditions. The obtained results confirm the feasibility of using a comprehensive preprocessing pipeline to enhance the reliability of computer vision systems for FPV drones.