Semantically Aware UAV Landing Site Assessment from Remote Sensing Imagery via Multimodal Large Language Models

TL;DR

This work tackles safe UAV emergency landing by moving beyond geometric flatness to semantic risk awareness using remote-sensing imagery and multimodal LLMs. It introduces a coarse-to-fine pipeline with Stage 1 semantic filtering, Stage 2 iterative visual verification, and Stage 3 context-aware reasoning that fuses visual evidence, POI data, dynamic context, and regulatory constraints to rank landing sites with natural language explanations. The Emergency Landing Site Selection (ELSS) benchmark, consisting of 500 expert-annotated samples, enables robust cross-domain evaluation and demonstrates that semantic methods outperform purely geometric baselines in risk identification, while providing interpretable justifications for trust and auditability. Practical impact includes improved safety in autonomous UAV operations and a framework that can integrate real-time data and regulatory guidelines, with future work focusing on edge deployment and dynamic data fusion to reduce latency and increase reliability.

Abstract

Safe UAV emergency landing requires more than just identifying flat terrain; it demands understanding complex semantic risks (e.g., crowds, temporary structures) invisible to traditional geometric sensors. In this paper, we propose a novel framework leveraging Remote Sensing (RS) imagery and Multimodal Large Language Models (MLLMs) for global context-aware landing site assessment. Unlike local geometric methods, our approach employs a coarse-to-fine pipeline: first, a lightweight semantic segmentation module efficiently pre-screens candidate areas; second, a vision-language reasoning agent fuses visual features with Point-of-Interest (POI) data to detect subtle hazards. To validate this approach, we construct and release the Emergency Landing Site Selection (ELSS) benchmark. Experiments demonstrate that our framework significantly outperforms geometric baselines in risk identification accuracy. Furthermore, qualitative results confirm its ability to generate human-like, interpretable justifications, enhancing trust in automated decision-making. The benchmark dataset is publicly accessible at https://anonymous.4open.science/r/ELSS-dataset-43D7.

Figures (7)

• Figure 1: Conceptual comparison between conventional geometry-based approaches and the proposed MLLM-driven framework for emergency landing site selection.
• Figure 2: Scheme chart of the proposed framework.
• Figure 3: Comparison of remote sensing imagery and standard maps highlighting segmentation results. The figure shows correct segmentation of the upper forest half and misclassification of the lower half as background, emphasizing the limitations of using only satellite images for segmentation.
• Figure 4: The flow chart of proposed stage 1. Specifically, $S(\cdot)$ denotes the semantic segmentation network, and $C(\cdot)$ corresponds to the classification head that evaluates pixel-wise landing suitability.
• Figure 5: Flow chart of the recommendation and filtering scheme.