Article Details

Abstract

Dense object distribution and severe occlusion remain significant challenges in remote sensing image object detection. To address the characteristics of densely arranged objects and multi-directional occlusions in the DOTA-HBB dataset, this paper proposes an improved YOLOv8n network to enhance the accuracy and robustness of dense object detection. Specifically, the Convolutional Block Attention Module (CBAM) is integrated into the backbone network and combined with the C2f structure to form a C2f_CBAM module, which strengthens key object features and suppresses background interference through channel and spatial attention mechanisms. In addition, a Coordinate Attention (CoordAtt) mechanism is introduced in the multi-scale feature fusion stage to capture long-range spatial dependencies among objects and improve location perception capability. Furthermore, the original CIoU loss function is replaced with the Scylla-IoU (SIoU) loss function, which introduces the vector angle between the predicted box and the ground-truth box to provide directional guidance for bounding box regression, thereby improving localization accuracy for dense and multi-oriented objects. Experimental results show that the proposed CS YOLOv8n model achieves an mAP of 71.7% on the DOTA-HBB dataset, which is 2.6% higher than that of the baseline model. The results demonstrate that the proposed attention mechanisms and improved loss function effectively enhance detection performance in complex dense scenarios and provide a feasible solution for accurate detection of high-density objects in remote sensing images.

Keywords

remote sensing image object detection; dense object detection; YOLOv8n; CBAM attention mechanism; coordinate attention; SIoU loss function

References

1. [1] Qin Lunming, Mei Wenquan, Cui Haoyang, etc Improved YOLOv8s Remote Sensing Image Object Detection Algorithm [J]. Advances in Laser and Optoelectronics, 2025, 62 (10): 317-325
2. [2] Guo Ruidong Research on Remote Sensing Image Object Detection Based on YOLOv8 [D]. Guilin: Guilin University of Technology, 2025
3. [3] Guo Xuan Research on Small Object Detection Algorithm Based on CEMA-YOLOv8 [D]. Ganzhou: Jiangxi University of Science and Technology, 2025
4. [4] Huang Weihe Research on Multi object Detection of SAR Images Based on Lightweight YOLOv8 Model [D]. Beijing: Beijing Jianzhu University, 2025
5. [5] Feng Zhaochen, Zhou Hu, Wu Zhongjun Remote sensing image object detection algorithm based on improved YOLOv8 [J]. Computer Engineering and Design, 2025, 46 (7): 1856-1863
6. [6] Yu Zhengbo Research on Small Object Recognition and Detection Algorithm Based on Deep Learning [D]. Chengdu: Chengdu University of Technology, 2023
7. [7] Li Jia, Wu Zongze Improved EfficientNetv2-YOLOv5s lightweight SAR remote sensing image target detection method [J]. Surveying and Spatial Geographic Information, 2026, 49 (1): 108-111115
8. [8] Fu Xianglin, Li Zongmei, Lv Xiangcheng Improved high-precision detection algorithm for aircraft remote sensing images using YOLOv5 [J]. Journal of Xiamen University of Technology, 2025, 33 (3): 10-18
9. [9] Wan Huiyao, Zhang Chaochen, Chen Jie, etc Vehicle detection in remote sensing images based on VG-YOLO [J]. Modern Applied Physics, 2025, 16 (3): 197-203
10. [10] Liu Junrong Target detection method for low altitude remote sensing images of unmanned aerial vehicles with high overlap based on improved YOLOv9 [J]. Computer Measurement and Control, 2025, 33 (8): 79-85

About the authors

Zhang Nannan (August 2000-), female, Han ethnicity, from Harbin, Heilongjiang, master's student, research interests include big data and artificial intelligence.