A Deep Learning Approach to Aircraft Detection and Classification in Satellite Imagery Using YOLOv8

Identification of civilian aircraft in high resolution satellite imagery is very important for air traffic management, infrastructure planning, and security applications. This paper presents a specific model for identification and classification of civilian aircraft utilizing YOLOv8 and the FAIR1M-2.0 dataset.The dataset was reduced to 11 different aircraft types for achieving high classification accuracy. The annotations are transformed into YOLO format for increased compatibility, and a YOLOv8 Nano model is employed for real-time detection with optimized computational efficiency.To improve the model’s generalizability, the training pipeline uses a multi-component loss function and data augmentation approaches. In the inference stage, the model detects objects, classifies them, and identifies aircraft, all of which are enhanced using the Non-Maximum Suppression technique. The results demonstrate that the suggested model allows quick real-time processing and has excellent classification accuracy, which advances aerial image analysis for tracking civilian aircraft and other related applications.