Abstract—Image classification using Remote Sensing (RS) or satellite images provides data about Land Use and Land Cover (LULC), which is also used in several applications such as environmental monitoring and urban planning. In recent times, Deep Learning (DL)-based approaches have been used for LULC classification and have achieved high effectiveness. However, it remains challenging because of class similarity and mixed-pixel issues, which lead to misclassification and reduced classification performance. In this manuscript, the Deformable Sub-Pixel Convolutional Layer with Convolutional Neural Network (DSPCL with CNN) is developed to effectively classify LULC by mitigating class similarity and mixed-pixel issues. The deformable convolutional layer is incorporated instead of the traditional convolutional layer, which adaptively changes the filters and captures deep features. Then, the sub-pixel convolutional layer is incorporated into the output layer, which separates the pixels into multiple pixels and classifies the classes more effectively. The developed DSPCL with CNN method achieved 98.62% accuracy on the EuroSAT dataset and 99.01% accuracy on the NWPU-RESISC45 dataset compared to conventional techniques.
 
Keywords—convolutional neural network, deformable convolutional layer, land use and land cover, mixed pixels, sub-pixel convolutional layer


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).