This research introduces LPANet, a novel detection framework that uses Large Language Models to guide feature alignment between different modalities in UAV object detection systems.

• Solves critical semantic gaps between visual sensors through progressive alignment
• Leverages LLM-extracted semantic features to improve detection accuracy
• Achieves superior performance compared to existing multimodal detection methods
• Enables more reliable object identification from aerial platforms

Why it matters for Aerospace: Enhanced detection capabilities on UAVs directly improve reconnaissance, surveillance, and monitoring applications, allowing for more reliable autonomous operations in diverse environments.

Large Language Model Guided Progressive Feature Alignment for Multimodal UAV Object Detection