Visual Question Answering on Remote Sensing Data

This research aims to improve Visual Question Answering systems for remote sensing data by introducing the Remote Sensing Visual Question Answering framework. Remote Sensing Visual Question Answering addresses the challenges of interpreting complex, multi-dimensional remote sensing images by providing precise answers to natural language questions. In this study, the BERT encoder is used for text embeddings, while the ResNet-152 and Swin Transformer architectures are used for feature extraction from pictures. Additionally, a bilinear attention mechanism is also incorporated to effectively fuse these embeddings, enabling the model to capture intricate relationships between visual and textual information. The proposed approaches underwent thorough evaluation on low-resolution datasets and showed significant improvements in accuracy relative to other baseline models, especially in tasks like scene categorization, object identification, and spatial reasoning. The results underscore the potential of integrating advanced deep learning architectures for remote sensing applications, paving the way for more intuitive and accessible data analysis in this domain, and highlighting future research directions in Remote Sensing Visual Question Answering systems.