Remote sensing image classification is a critical task for analyzing geographical and environmental data. This project proposes a novel hybrid framework that combines Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) to classify remote sensing images into 45 distinct classes a challenging benchmark due to high inter-class similarity and diverse scene complexity. To tackle this, the architecture integrates Inception V3 and EfficientNet, two state-of-the-art CNN models known for their efficiency and strong feature extraction capabilities, to extract rich spatial features from input images. These features are further refined using a Hierarchical Vision Transformer framework, specifically the Swin Transformer, which effectively captures long-range dependencies and hierarchical representations through shifted windows. To enhance adaptability across varying image resolutions and scene types, a Dynamic Window Shift Unit is incorporated, improving the model’s robustness and spatial flexibility. This synergistic integration of CNN-based and Transformer-based modules results in a highly accurate, scalable, and robust solution for remote sensing image classification, with promising applications in environmental monitoring, urban planning, and disaster response.

Keywords: Remote Sensing, Image Classification, Swin Transformer, EfficientNet, Vision Transformers, Dynamic Window Shift, Hierarchical Model, Deep Learning, Scene Classification, Environmental Monitoring.