Lightweight multidimensional feature enhancement algorithm LPS-YOLO for UAV remote sensing target detection.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Detecting small targets in Unmanned Aerial Vehicle (UAV) remote sensing images presents significant challenges for conventional lightweight methods.
• Difficulties arise from inadequate feature extraction and substantial background interference, hindering accurate identification.

Purpose of the Study:

• To develop an effective and computationally efficient lightweight model for small target detection in UAV remote sensing images.
• To improve feature extraction capabilities and reduce computational complexity compared to existing methods.

Main Methods:

• Proposing LPS-YOLO, a novel architecture that replaces the standard Convolutional Neural Network (CNN) backbone with SPDConv for enhanced fine-grained feature retention.
• Integrating the SKAPP module for superior feature fusion and employing E-BiFPN and OFTP structures for efficient information preservation and transfer from the backbone.

Main Results:

• On the VisDrone2019 dataset, LPS-YOLO achieved a 17.3% increase in mean Average Precision (mAP) and a 42.5% reduction in parameters versus the baseline.
• Experiments on the DOTAv2 dataset showed a 14.5% improvement in F1 score and a 14.9% increase in mAP compared to YOLOv8-n, demonstrating robustness.

Conclusions:

• LPS-YOLO provides an effective solution for multi-target detection in UAV remote sensing, outperforming existing lightweight models.
• The model's design addresses key challenges in small target detection, offering improved accuracy and efficiency.