A Novel Wetland Detection and Classification Using Invasive Weed Optimization Algorithm with Deep Learning Model

Abstract

This research shows some of the worthy ecosystem services provided by wetlands are biodiversity support, water regulation, and carbon sequestration. These ecosystems are under threat due to human activities such as urbanization and industrialization, so their identification and classification are very important for their conservation. Synthesis aperture radar has been successful in detecting wetlands since it can penetrate through cloud covers and is sensitive to moisture content. Although promising Support Vector Machine (SVM) and Random Forest (RF) approaches, this still often requires high preprocessing and exhaustive handcrafted feature extraction. Deep learning models, especially Convolutional Neural Networks (CNNs), manage to improve classification performance but have problems with slow convergence rates and serious overfitting in big data. Based on these gaps, the paper presents a new approach proposing the integration of the modified Invasive Weed Optimization (IWO) algorithm with CNN-based detection for Synthetic Aperture Radar (SAR) images, focusing on wetland classification. Addressing the slow convergence problem and possible entrapment into a local optimum, enhanced IWO parameters in the CNN include learning rate and batch size. Also, to test the applied model, SAR data demonstrates a classification accuracy of about 90%, with key metrics such as precision at 88% and recall at 91% showing superior performance compared to that obtained using traditional techniques. Conclusively, the study states that the combination of CNN and the improved IWO algorithm highly enhances the accuracy of wetland detection, thereby providing a more robust solution for conducting environmental monitoring and conservation efforts.