Transforming remote sensing images to textual descriptions

Remote sensing data is growing enormously by virtue of the advancements in satellite and drone technology. Generating description of these remote sensing images have gained much attention in recent past owing to its applications in remote sensing image retrieval (RSIR) and image analysis. Remote sensing images capture huge diversity in different perspectives and levels. Due to overhead perspective and significantly larger scale of the scene, extraction of visual features from the remote sensing images to generate descriptions has become a challenging task. In this work, a model is proposed to generate novel captions by considering multi-scale features processed through adaptive attention based decoder using topic sensitive word embedding. The proposed model has been quantitatively evaluated using the benchmark remote sensing image captioning datasets and ablation study has been conducted to investigate the reason behind its effectiveness. Experimental evaluation depicts that the proposed model shows promising results compared to the existing state of the art remote sensing image description models.