Target Detection Algorithm for UAV Aerial Images Based on IRE-YOLO

Abstract

To address challenges in UAV aerial images such as a high proportion of small objects, complex backgrounds, significant scale variations, and limited computing resources, this study proposes an improved object detection algorithm based on YOLOv11n—namely, IRE-YOLO. In the backbone network, the algorithm introduces the inverted residual mobile block (iRMB) attention mechanism to enhance the model's ability to extract features of small objects in complex backgrounds. It replaces some standard convolutions with depthwise separable convolutions (DSConv) to reduce the number of parameters while maintaining feature expression capability. In the neck network, an efficient upsampling block (EUCB) is designed to recover detailed information through hierarchical feature reconstruction, thereby improving the representation quality of small objects' edges and textures. Experimental results on the DOTA dataset show that IRE-YOLO outperforms mainstream lightweight models in key metrics including precision (P), recall (R), mAP@50, and mAP@50:95, achieving 69.4%, 59.0%, 61.0%, and 34.2% respectively. Meanwhile, the model's parameter count is reduced to 8.7MB, realizing an effective balance between accuracy and lightweight design. The research demonstrates that IRE-YOLO has high practical value in small object detection tasks for UAV aerial images and provides a feasible solution for real-time deployment on embedded platforms.