AquaSentinel

# AI-Driven River Pollution Monitoring System  ## 🌊 Overview This project proposes an **AI-powered river pollution monitoring system** that combines **remote sensing**, **adaptive IoT sensor networks**, and **deep learning-based image classification** to detect and manage river pollution in real-time. Traditional methods rely on fixed sensors and manual testing—this system improves on that by delivering scalable, intelligent, and adaptive pollution management. --- ## 🧠 Key Features - **Hybrid Deep Learning Framework**: - Uses CNNs for low-level feature extraction. - Combines with Vision Transformers (ViT) for semantic-level classification. - **Adaptive Spatiotemporal Pollution Estimation**: - Integrates sensor feedback, probabilistic modeling, and Gaussian Process Regression. - Dynamically adjusts sensor positions to minimize monitoring uncertainty. - **Reinforcement Learning Optimization**: - Sensor deployment and data acquisition schedules are optimized using a reward-driven policy. - Balances between accuracy, computational cost, and measurement frequency. - **Real-time Data Assimilation**: - Incorporates a Kalman Filter-based mechanism for continuous pollution concentration updates. - **Edge & Federated Learning Support**: - Enables distributed real-time analytics while preserving data privacy. --- ## 🧪 Datasets This project leverages several remote sensing and environmental monitoring datasets: - **Sentinel-2 MSI** (ESA): High-res multispectral imagery.- **Global Mangrove Watch**: Mangrove distribution and coastal pollution tracking.- **MDID**: Hyperspectral & multispectral datasets for segmentation tasks.- **ESA-OC-CCI**: Ocean color data for turbidity and chlorophyll analysis. --- ## 🏗️ Architecture Highlights - **Dual-Branch Network**: Combines CNN and ViT for robust image segmentation.- **Dynamic Sensor Placement**: Informed by pollution gradients and uncertainty variance.- **Reinforcement-Guided Monitoring**: Employs multi-agent RL for coordinated sensor movement.- **Bayesian Data Fusion**: Enhances predictions using probabilistic modeling. See the schematic diagrams on **page 8**, **11**, and **12** of the paper for architectural illustrations. --- ## 📊 Results & Evaluation The model was benchmarked against state-of-the-art methods like **UNet**, **HRNet**, **DeepLabV3+**, **SegFormer**, and more. Across datasets, our system achieved: - **+2.5–3.0% IoU and Dice Score improvement**- **Superior spatial and spectral attention**- **Lower false positives/negatives in segmentation tasks** Full performance tables and ablation studies are available in the paper (see *Tables 1–4* and *Figures 5–8*). --- ## 🛠️ Installation ```bashgit clone https://github.com/your-username/river-pollution-monitor.gitcd river-pollution-monitorpip install -r requirements.txt