Aircraft Detection Approach Based on YOLOv9 for High-Resolution Remote Sensing

The rapid advancements in Cube and Nano satellites have made high-quality remote sensing images easily accessible, leading to crucial object recognition tasks with extensive applications range. The particular task of aircraft detection is placed as one of the most promising applications of computer vision systems for aerospace data processing, primarily due to its broad application prospects starting from aircraft traffic planning and controlling for potential threats detection, environmental monitoring to managing CO2 emissions all the way to logistics supply chain optimization based on spatial finance insights. However, analyzing high-quality remote sensing data presents a significant challenge, particularly in the detection tasks of densely located aircrafts, aircrafts with various size ranges, etc. In this regard, this article focuses on addressing these challenges through a deep learning-based approach based on an optimized version of YOLOv9 for aircraft detection on high-resolution remote sensing data. The proposed approach has been trained and evaluated based on the images from Airbus dataset and demonstrates 3 % increase in accuracy detection compared to existing state-of-the-art approaches, achieving 0.987 in average precision (AP) and 0.746 in mean average precision (mAP).