Background-Aware Prompt Learning for Remote Sensing Scene Classification

Shi, Wenxu and Meng, Qingyan and Zhang, Linlin and Zhou, Tingyuan and Atkinson, Peter M. (2025) Background-Aware Prompt Learning for Remote Sensing Scene Classification. In: IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium :. IEEE, pp. 6016-6020. ISBN 9798331508111

Abstract

Remote sensing scene classification (RSSC) plays a pivotal role in applications such as environmental monitoring and urban planning. Traditional deep learning models for RSSC require extensive labeled datasets for training, which is costly and time-consuming, and they often struggle to generalize to new datasets or varying environmental conditions. Recently, VisionLanguage Models (VLMs), like CLIP, have shown significant potential in visual recognition tasks by leveraging contrastive learning on large-scale image-text data. However, when applying VLMs to remote sensing imagery, we observe an inherent limitation where models exhibit background response patterns regardless of the target category. To address this challenge, we propose a background-aware prompt learning (BAPL) framework that effectively guides the model to focus on discriminative features while suppressing background interference. Extensive experiments demonstrate that BAPL significantly outperforms existing methods across few-shot learning, cross-dataset generalization, and domain adaptation scenarios, showing robust performance in real-world remote sensing applications.