Detection and Classification Of Harmful Algae Using Machine Learning

In this study, we devised a robust predictive model for estimating cyanobacteria density, a critical aspect of environmental monitoring, by employing satellite images alongside sophisticated machine learning methods. Leveraging Landsat and Sentinel-2 imagery as foundational data sources, we conducted comprehensive feature extraction to capture intricate ecosystem nuances. This involved extracting essential parameters such as RGB values and elevation data, augmented by over 23,000 manually sampled data points to ensure diverse representation. Our dataset underwent rigorous partitioning for training and testing, facilitating model training and evaluation. We employed three gradient boosting models - LightGBM, XGBoost, and CatBoost - renowned for their ability to handle complex datasets and provide accurate predictions. Through optimization of model hyperparameters, we achieved enhanced performance and generalization capability. Integration of Gradio, an open-source UI library, enabled visualization of prediction results, offering stakeholders an interactive platform to assess cyanobacteria density across geographical areas. The implementation of this predictive pipeline serves as a crucial bridge between satellite imagery and actionable insights for environmental management. Future directions include refining feature engineering techniques, exploring additional data sources, and fostering collaborations with environmental agencies for broader applications in ecosystem conservation.