Prompt Generation for Enhanced Camouflaged Object Detection in Low-Altitude Economy

To ensure the safety of both aircraft and operators in low-altitude economy (LAE) activities, precise environmental perception capabilities are essential for effective collision prevention. However, achieving accurate perception remains challenging, particularly when obstacles are camouflaged and visually blended into their surroundings, making detection difficult, even with the robust foundation model, the Segment Anything Model (SAM). Although SAM's prompt-based strategy improves its performance in the Camouflaged Object Detection (COD) task, its reliance on limited prompts introduces new challenges. Instead of manually annotating prompts, our work introduces a multimodal learning approach that utilizes a Vision-Language Model (VLM) to automatically generate mask prompts. By integrating visual and textual information, this work generates high-quality prompts that significantly enhance the performance of SAM in identifying camouflaged objects. Experimental results demonstrate that the proposed method achieves an average improvement of 13% over the baseline SAM across three COD benchmark datasets.