AI-Powered Land Classification: Analyzing Deep Learning Models for Urban and Desert Images

Abstract

Land degradation and environmental concerns resulting from rapid urbanization in Oman in recent years have made innovative and advanced strategies and approaches for sustainable urban planning & environmental preservation a necessity. This research presents an AI-based classification and analysis method that utilizes Deep Learning algorithms to analyze satellite and drone images for monitoring urban growth and desertification. A Convolutional Neural Network (CNN)-based binary image classification model was developed to distinguish between urban infrastructure and natural landscapes based on the datasets retrieved from Kaggle and its performance was evaluated by accuracy, precision, recall, and F1 score. The CNN model achieved an accuracy of 53%, but the confusion matrix demonstrated poor performance for classifying urban areas with a recall of 0.00 for the Urban class. To address this issue, a pre-trained model was implemented, achieving 54% accuracy and stronger class-wise recall for both categories compared to the baseline CNN model. The results demonstrated both the potential and the limitations of deep learning models for land classification tasks, delivering valuable insights for urban planning and environmental monitoring, where visual similarities pose major challenges. Despite modest accuracy, the study demonstrates the feasibility of AI-driven land assessment as an additional tool for environmental monitoring, urban expansion tracking and desertification analysis. Furthermore, it aligns with Oman Vision 2040 digital transformation and supports Sustainable Development Goals (SDGs), particularly SDG 13 (Climate Action) and SDG 15 (Life on Land), by emphasizing data-driven approaches for sustainable development and ecological resilience.