Land Use-Cover Change Modelling Using Deep Learning Techniques with a focusing on an Agrarian Community

Land use change detection modelling is critical to understanding the dynamics of natural and human-induced environmental transformations. It embodies profound implications for agricultural productivity, food security, and sustainable land management, particularly in developing agrarian communities. While existing studies provide some insight, it is important to investigate modern techniques for further improvement. This study, therefore, proposes a Convolutional Neural Network (CNN)-Channel-wise Attention (CNN-CwA) deep learning model for land-use and land-cover change (LULCC) classification. It integrates a CwA mechanism into a CNN backbone to enable dynamic spectral channel recalibration and improve land cover discrimination in heterogeneous agrarian landscapes. The model is trained and validated on the EuroSAT benchmark dataset, comprising 27,000 geo-referenced Sentinel-2 satellite image patches spanning ten land cover classes. A five-fold cross-validation strategy was employed alongside t-Distributed Stochastic Neighbour Embedding (t-SNE) visualisation to assess model generalisation and feature discriminability. The CNN-CwA model achieved a mean Intersection over Union (mIoU) of 0.8852, representing a 2.86% improvement over the baseline CNN (mIoU = 0.8606), with notable per-class gains for Sea or Lake of +0.0811, Pasture of +0.0365, River of +0.0292, and Highway of +0.0288. Vegetation-related classes including Forest, Herbaceous Vegetation, and Permanent Crop all exceeded the 0.90 IoU threshold. The findings affirm the model's reliability for agronomically relevant land cover discrimination. In addition, it demonstrates the practical utility of the CNN-CwA model as a robust and scalable tool for LULCC classification in agrarian landscapes. The model also serves as a computational foundation for evidence-based agricultural land use planning and food security policy formulation in Nigeria and similar developing-country contexts.