Military Aircraft Detection in Remote Sensing Images Based on Improved YOLOv8
Year: 2026
Type: conference-paper
Abstract: Remote sensing military aircraft detection plays an important role in aerial surveillance and intelligent reconnaissance. Due to complex background interference, diverse target scales and dense distribution of aircraft objects in remote sensing images, traditional detection methods often suffer from low accuracy and poor generalization. To address these problems, this paper proposes an improved YOLOv8 detection method for multi-class aircraft recognition based on the MAR20 dataset. Three effective optimization strategies are adopted: composite data augmentation is used to enrich sample diversity and suppress overfitting; transfer learning with pre-trained weights accelerates network convergence and extracts robust visual features; and SIoU loss function replaces the original CIoU to optimize bounding box regression and enhance localization accuracy for small aircraft targets. Comprehensive experiments are conducted on the MAR20 dataset. Ablation experiments verify the effectiveness of each improved module. Comparative experiments among YOLOv8 series models (YOLOv8-N, YOLOv8-S, YOLOv8-M) and different YOLO versions (YOLOv5, v8, v10, v11) are carried out from multiple evaluation metrics including Precision, Recall, [email protected] and [email protected]:0.95. Qualitative visualization results further demonstrate the superior detection performance of our method. Experimental results show that the improved YOLOv8 achieves favorable detection accuracy and strong robustness, which can be well applied to remote sensing military aircraft detection tasks.
Institution: Shenyang Environmental Protection Bureau
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202394985