Invasive water hyacinth classification using a Machine learning approach

Water Hyacinth (Eichhornia crassipes) has become a major ecological threat to inland water bodies in India, with extensive proliferation across wetlands in Assam. The present study was conducted to classify aquatic vegetation as Water Hyacinth (WH) in two ecologically important wetlands, Deepor Beel and Digholi Beel in Kamrup Metro and Kamrup Rural districts, respectively. Three spectral indices Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI) and Leaf Area Index (LAI), were integrated by adopting a weighted linear combination method to create a Weighted Composite Raster (WCR), which enhanced training datasets for image classification. Statistical classifiers, such as the Maximum Likelihood Classifier (MLC), and machine learning classifiers, such as the Random Tree (RT) and the Support Vector Machine (SVM), were employed over two iterations. First, only ground samples (GT) were used; then, additional class attributes (GT+WCR) were employed. Ground-truthing validated the comparative results, revealing that the machine learning classifier using the novel approach of GT with WCR significantly outperformed the SVM, achieving overall accuracies of 96% and 93% and Kappa coefficients of 95% and 90% for Deepor Beel and Digholi Bil, respectively, as compared to the traditional statistical classifier. The study thus demonstrates a novel approach that enables precise quantification of invasive (WH), offering scalability and reliability for ecological management and control strategies, ensuring sustainable conservation of wetland ecosystems.