Small-Target Detection Algorithm Based on Improved YOLOv11n.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Remote Sensing

Background:

• Small target detection in UAV aerial photography is challenging due to scale and background complexity.
• Existing algorithms often struggle with feature loss and cross-level conflicts during downsampling.

Purpose of the Study:

• To propose an improved YOLOv11n algorithm for enhanced small-target detection in UAV aerial photography.
• To address limitations in feature fusion, feature highlighting, and detection accuracy.

Main Methods:

• Implemented a detection head on the 160x160 feature layer and fused features using Asymptotic Feature Pyramid Network (AFPN).
• Integrated Spatial Channel Attention SPPF (SCASPPF) and enhanced the loss function with MPDIoU and InnerIoU.
• Utilized Inception Deep Convolution (IDC) to improve the C3k2 module for an expanded receptive field.

Main Results:

• The improved YOLOv11n algorithm achieved 39.256% mAP@0.5 on the Visdrone2019 dataset.
• This represents a 6.689% improvement compared to the benchmark YOLOv11n's 32.567% mAP@0.5.
• The modifications effectively alleviated feature loss and improved the model's ability to detect small objects.

Conclusions:

• The proposed enhancements significantly improve small-target detection performance in UAV aerial imagery.
• The integration of AFPN, SCASPPF, enhanced loss functions, and IDC contributes to higher detection accuracy.
• This algorithm offers a more robust solution for complex aerial surveillance and monitoring tasks.