Year: 2025
Type: article
Abstract: Target detection in UAV aerial photography scenarios faces challenges of small targets and complex backgrounds. Thus, we proposed an improved YOLOv11n small-target detection algorithm. First, a detection head is added to the 160 × 160 resolution feature layer, and non-adjacent layer feature is fused via Asymptotic Feature Pyramid Network (AFPN) to alleviate feature loss caused by downsampling and reduce cross-level feature conflicts. Second, the Spatial Channel Attention SPPF (SCASPPF) module replaces the original Spatial Pyramid Pooling-Fast (SPPF) module to highlight key features and suppress irrelevant ones. Moreover, the loss function is enhanced by fusing MPDIoU and InnerIoU to boost detection accuracy. Finally, Inception Deep Convolution (IDC) is adopted to improve the C3k2 module, expanding the model's receptive field and enhancing small-target detection performance. Experiments on the Visdrone2019 dataset show that the algorithm achieves 39.256% [email protected], 6.689% higher than 32.567% [email protected] of the benchmark model (YOLOv11n).
Source: Sensors
DOI: 10.3390/s26010071
Authors: Ke Zeng, Wangsheng Yu, Xianxiang Qin, Siyu Long
Institution: Air Force Engineering University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W7117151504