YOLO-RLDW: An Algorithm for Object Detection in Aerial Images Under Complex Backgrounds

Abstract

Aiming at the challenges of low detection accuracy, susceptibility to complex background interference, difficulty in detecting small objects, and multi-scale object issues in aerial images, our proposed an improved YOLOv8-based object detection algorithm, named YOLO-RLDW. Leveraging the advantages of Receptive Field Attention Convolution (RFAConv), we designed a feature extraction module named C2f-RFA to enhance the feature extraction capability for small objects in aerial images. Inspired by the concept of Large Separable Kernel Attention (LSKA), we developed the SPPF-LSKA module, which effectively reduces the interference of aerial backgrounds in object detection. We replaced the YOLOv8 detection head with a Dynamic Head (DyHead), further enhancing the model’s generalization and adaptability. Finally, we employed as boundary box regression loss based on a dynamic focusing mechanism, WIoU, as the loss function, which accelerates model convergence while improving the localization capability for multi-scale objects. Experimental results demonstrate that on the VisDrone2021 dataset, the proposed algorithm achieves improvements of 5.5%, 3.9%, 5.4%, and 3.7% in precision (P), recall (R), mean average precision (mAP50), and mAP95, respectively, compared to the original algorithm. On our self-built remote sensing image dataset RSI, the accuracy, recall, and mean average precision reach 94.2%, 91.0%, and 95.4%, respectively, demonstrating good performance in detecting objects in aerial images. Comparison with other mainstream object detection algorithms validates the effectiveness and superiority of the proposed method.

Keywords

• Aerial image
• object detection
• deep learning
• complex background
• YOLO-RLDW