CF-YOLO for small target detection in drone imagery based on YOLOv11 algorithm.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Drone imagery presents challenges for target detection due to scale variations and small, low-detail objects.
• Existing algorithms struggle with information loss and feature redundancy in multi-scale feature fusion for small targets.

Purpose of the Study:

• To develop an improved small target detection algorithm for drone-based remote sensing.
• To address information loss and feature redundancy issues in multi-scale feature fusion.

Main Methods:

• Proposed a Cross-Scale Feature Pyramid Network (CS-FPN) to mitigate information loss in hierarchical convolutional structures.
• Introduced a Feature Recalibration Module (FRM) and Sandwich Fusion Module for effective multi-scale feature fusion.
• Optimized the model with RFAConv module and LSDECD detection head.

Main Results:

• CF-YOLO achieved significant improvements in mean Average Precision at 50% IoU (mAP50) on VisDrone, TinyPerson, and HIT-UAV datasets.
• Demonstrated superior performance compared to baseline models and other state-of-the-art methods.
• Specific improvements: 12.7% on VisDrone, 10.1% on TinyPerson, and 3.5% on HIT-UAV.

Conclusions:

• The proposed CF-YOLO algorithm effectively enhances small target detection in drone imagery.
• The novel CS-FPN, FRM, and Sandwich Fusion modules contribute to improved feature representation and fusion.
• CF-YOLO offers a robust solution for challenging remote sensing small target detection tasks.