YOLO11-LKAConv: optimizing UAV image multi-target detection based on improved YOLO architecture

Abstract

This paper presents YOLO11-LKAConv, an improved model based on the YOLOv11 framework, to address the issue of low detection accuracy for small targets in UAV aerial images. By replacing standard convolutions with lightweight large-kernel attention convolution (LKAConv), the model integrates dynamic large-kernel receptive fields and channel-spatial attention mechanisms, enhancing its ability to capture long-range contextual features for small targets. Experiments on the VisDrone2019 dataset show that the improved model achieves an mAP50-95 of 0.173, a 1.2% increase over the baseline YOLOv11n, with improvement in both P and mAP50 while maintaining almost the same inference time. The results indicate that LKAConv effectively balances detection accuracy and computational efficiency through its lightweight large-kernel design, offering a better solution for real-time UAV target detection tasks.