Investigating image quality impact on remote sensing object detection

YOLO-series object detection methods have currently achieved significant success in remote sensing (RS) image analysis but depend on high-quality images, which are difficult to obtain due to payload limitations, atmospheric effects, and satellite motion. These challenges often lead to image degradation, including noise, occlusion, and blurring. Although quality enhancement can be applied, identifying the specific types of degradation to address and determining the best enhancement strategies remain unresolved issues. In this paper, we simulate various levels of RS image quality degradation and evaluate the impact on object detection performance using YOLOv5 models on prevailing RS image datasets. By analyzing how different types of degradation affect the detection performance of various object categories, we provide insights into which degradation recovery methods should be prioritized at specific quality levels. This work provides guidances on when and how to enhance the quality of RS images, aiming at improving further performance of object detection.