Land pollution classification model using improved graph attention network-based optimization framework

Owing to increased urbanization, industrial processes, as well as poor waste management that are harming soil quality and ecosystem health, the categorization of land pollution has become more important. Accurately recognizing areas of polluted soil enables timely response as well as long-term environmental care. However, this endeavor faces significant challenges, including a lack of well-annotated information, wide variety in surface textures, as well as overlapping features among contaminated and uncontaminated areas. Because of these complexities, automatic as well as reliable categorization is a difficult but essential endeavor. Therefore, this research performs the Land Pollution Classification (LPC) model using intelligent deep learning methodology. The dataset is initially accumulated from standard benchmark sources referred as Sentinel-2 Land Use Land Cover (Sen-2 LULC) dataset. The pre-processing is next done with the help of cloud masking and cloud-cover removal approaches. Further, the features are extracted using the Local Binary Pattern (LBP) technique. Additionally, the final classification stage of the proposed LPC model is accomplished with the help of deep learning model referred to as novel Improved Graph Attention Network (IGAT). The parameter optimization in GAT is performed by the nature inspired optimization algorithm called Wombat Optimization Algorithm (WOA). Accuracy maximization is considered as the final objective function behind the entire process of the proposed LPC model. The proposed IGAT-WOA is 14.29% and 12.20% better than the other methods with respect to accuracy and specificity for the developed LPC model respectively.