Aerial Image Detection Based on DCB-YOLOv8-SOD: An Improved Object Detection Model

Due to the complex backgrounds, multi-scale target distributions, and high-resolution characteristics typically found in aerial images, traditional target detection algorithms often struggle to accurately capture the detailed features of small targets, leading to missed detections and false positives, which affect the accuracy and stability of detection. To address these issues, this paper proposes an improved target detection model, DCB-YOLOv8-SOD. Based on YOLOv8, this model introduces depthwise separable convolutions and attention mechanisms to enhance the network’s feature extraction capabilities and optimize the feature fusion process, while reducing computational complexity. This improves the model’s adaptability to complex aerial imagery and enhances its ability to accurately capture and fuse features of small targets. Additionally, the model employs an adaptive target focus loss to optimize weight allocation, improving the learning capability for hard-to-detect targets, thereby further enhancing the model’s robustness and stability. To validate the effectiveness of the proposed model, extensive experiments were conducted on multiple public aerial image target detection datasets. The experimental results show that DCB-YOLOv8-SOD has achieved excellent performance in small target detection tasks, with a significant improvement in detection accuracy. Although it is slightly slower in inference speed compared to some other models, its other performance metrics are much higher than those of other models, making the slightly slower inference speed worthwhile. Additionally, to further verify the contribution of each module, an ablation study was conducted to analyze the impact of depth wise separable convolution, attention mechanism, and adaptive target focus loss on the model’s performance, demonstrating the rationality and effectiveness of the design of each component.