Analysis of Deep Learning-Based Object Detection Across Infrared Bands for UAV Imagery in Military Applications

In modern military operations, object detection plays an important role in enhancing situational awareness and tactical decision-making. However, the use of object detection in complex military environments faces challenges, including poor visibility and the need for real-time decision-making. Infrared imaging technologies, which operate outside the visible spectrum, are effective for detecting military targets, particularly in low-visibility conditions. Despite the advantages of infrared imaging for object detection, comparison of performance across different infrared spectra in military applications remains limited. This limitation is primarily due to the varying object visibility across different infrared bands and the lack of accessible military datasets, which are crucial for analyzing how these bands influence military object detection and the performance of deep learning models under such conditions. This study provides an analysis of object detection across various infrared bands, including near-infrared, short-wavelength infrared, mid-wavelength infrared, and long-wavelength infrared, using popular deep learning-based object detection models such as YOLOv8, YOLOv9, YOLO Ghost P2, and RT-DETR. The models were trained separately on each infrared band using simulated data and evaluated for their ability to recognize military tanks, as this capability is crucial for unmanned aerial vehicles in modern defense applications. Our findings reveal that the near-infrared band is the most suitable for detecting tanks, and the YOLOv8 model provides the best performance.