RN-YOLO: A Small Target Detection Model for Aerial Remote-Sensing Images

Ke Wang,Hao Zhou,Hao Wu,Guowu Yuan

DOI: https://doi.org/10.3390/electronics13122383

IF: 2.9

2024-06-19

Electronics

Abstract:Accurately detecting targets in remote-sensing images is crucial for the military, urban planning, and resource exploration. There are some challenges in extracting detailed features from remote-sensing images, such as complex backgrounds, large-scale variations, and numerous small targets. This paper proposes a remote-sensing target detection model called RN-YOLO (YOLO with RepGhost and NAM), which integrates RepGhost and a normalization-based attention module (NAM) based on YOLOv8. Firstly, NAM is added to the feature extraction network to enhance the capture capabilities for small targets by recalibrating receptive fields and strengthening information flow. Secondly, an efficient RepGhost_C2f structure is employed in the feature fusion network to replace the C2f module, effectively reducing the parameters. Lastly, the WIoU (Wise Intersection over Union) loss function is adopted to mitigate issues such as significant variations in target sizes and difficulty locating small targets, effectively improving the localization accuracy of small targets. The experimental results demonstrate that compared to the YOLOv8s model, the RN-YOLO model reduces the parameter count by 13.9%. Moreover, on the DOTAv1.5, TGRS-HRRSD, and RSOD datasets, the detection accuracy (mAP@.5:.95) of the RN-YOLO model improves by 3.6%, 1.2%, and 2%, respectively, compared to the YOLOv8s model, showcasing its outstanding performance and enhanced capability in detecting small targets.

engineering, electrical & electronic,computer science, information systems,physics, applied

What problem does this paper attempt to address?

The paper mainly addresses the problem of object detection in aerial remote sensing images, particularly how to improve the detection accuracy of small objects. The authors propose a new object detection model, RN-YOLO (YOLO combined with RepGhost and Normalized Attention Module NAM), which is an improvement based on YOLOv8.

Specifically, the paper addresses the following issues:
1. **Difficulty in detecting small objects**: Due to the significant size differences of objects in remote sensing images and the presence of a large number of small objects, this poses a challenge to detection algorithms. RN-YOLO enhances the ability to capture small object features by introducing NAM, which achieves this by recalibrating the receptive field and strengthening the information flow.
2. **Feature extraction efficiency**: To reduce the number of parameters while maintaining or improving detection performance, the paper adopts the RepGhost module to replace the original C2f structure, creating a new feature fusion network RepGhost_C2f, effectively reducing the number of model parameters.
3. **Localization accuracy**: The difficulty of localizing objects of different sizes, especially small objects, is significant. Therefore, the paper introduces the WIoU loss function to replace the original CIoU loss function in YOLOv8, improving the localization accuracy of small objects by assigning different weights to objects of different sizes.

Experimental results show that compared to YOLOv8s, the RN-YOLO model improves detection accuracy on the DOTAv1.5, TGRS-HRRSD, and RSOD datasets by 3.6%, 1.2%, and 2%, respectively, and reduces model parameters by 13.9%, demonstrating RN-YOLO's excellent performance and significant advantages in detecting small objects.