Automatic Detection of Invasive Aquatic Weeds Using Remotely Sensed Multispectral Imagery

Recent rise in the population of the invasive aquatic vegetation in the Bhima River and its tributaries near the Pune city of India, is one of the major alarming concerns for the local environment and corresponding ecosystem. The Invasive Aquatic Weeds have been proliferating highly due to the presence of untreated sewage and the high levels of water contamination arising from the hazardous liquids directly flowing in the Bhima River from the industries in the neighborhood of Pune city. An attempt has been made in this research to incorporate the Multispectral Imagery acquired from the Sentinel-2 Satellite belonging to the European Space Agency. An Optimum Index Factor is employed to calculate the subset of three bands with the highest amount of the spectral information from several other spectral bands under consideration. A unique band composite of the selected Optimum Index Factor Bands of Green, Red and Near-Infrared with the Normalized Difference Vegetation Index, has been incorporated to differentiate the Invasive Aquatic Weeds from other materials having unique spectral signatures in the regions of interest. The Machine Learning Architectures of the Artificial Neural Network and the XGBoost have been employed in the Automatic Detection of the Invasive Aquatic Weeds in the Bhima River and one of its tributaries, the Mula-Mutha River. The results of the experimentation concludes that the machine learning approach of the XGBoost outperforms the Artificial Neural Network, both approaches have produced high accuracies due to richness of the data obtained from the meticulous data extraction pipeline.