End to end polysemantic cooperative mixed task trainer for UAV target detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Unmanned Aerial Vehicles (UAVs) are increasingly used for urban surveillance.
• Accurate object detection in UAV imagery is crucial for 3D reconstruction and urban analysis.
• Existing methods struggle with subtle features, complex backgrounds, and small targets in UAV images.

Purpose of the Study:

• To develop a novel end-to-end network for enhanced target detection in UAV images.
• To improve detection accuracy for urban objects, especially small and subtle targets.
• To establish a new benchmark for UAV image detection methodologies.

Main Methods:

• Introduced the Polysemantic Cooperative Detection Transformer (Pc-DETR) network.
• Developed a Polysemantic Transformer (PoT) Backbone with a dynamic attention matrix for enhanced visual representation.
• Implemented a Polysemantic Cooperative Mixed-Task Training scheme with auxiliary heads to boost encoder learning.

Main Results:

• Pc-DETR demonstrated superior detection performance compared to existing state-of-the-art methods.
• Achieved a 3% improvement in detection accuracy over MFEFNet.
• The PoT Backbone effectively captures static and dynamic features using contextual information.

Conclusions:

• Pc-DETR offers a significant advancement in UAV image target detection.
• The proposed methods enhance feature representation and training efficiency.
• This work contributes to the development of intelligent UAV surveillance systems.