<p>Remote sensing imagery plays a vital role in numerous societal applications, offering exceptionally high spatial resolution that enables the extraction of fine-grained features from each pixel. A modified Swin (Shifted Window) Transformer model, termed the Swin Transformer for Remote Sensing Image Classification (Swin-RSIC), is proposed for the effective analysis of very high resolution remote sensing imagery. Swin-RSIC incorporates a sparse attention mechanism to efficiently capture long-range dependencies while optimising parameter efficiency and memory usage. The proposed Swin-RSIC model has been validated on four major land use datasets, achieving a peak accuracy of 98.06%. Additionally, it demonstrates strong performance in cross domain classification, with an accuracy of 88.18% without relying on domain adaptation techniques. To enhance model interpretability, Class Activation Mapping (CAM) techniques, Local Interpretable Model-agnostic Explanations (LIME) and insertion deletion methods were employed. These methods offer visual insights into the key regions that influence the predictions of the proposed model, thereby improving transparency in decision-making. Results from the CAM methods indicate that the proposed model produces more precise and comprehensive feature activations across critical land regions. LIME visualisations highlight the ability of the model to focus on relevant spatial features. Insertion and deletion experiments validate the model robustness by demonstrating its ability to effectively retain and utilise essential features while maintaining high classification confidence. The combination of high classification accuracy and enhanced interpretability underscores the effectiveness of the modified Swin transformer model in both land use and cross-domain remote sensing image classification.</p>