Real-Time Satellite Image Analyzer for Flood Detection
Budget: ₹400 – ₹750 INR
Developed an end-to-end deep learning system that performs pixel-level flood segmentation on satellite imagery in real time. The model accurately identifies flooded areas from Sentinel-2 or similar multispectral satellite data and generates instant flood maps — ideal for disaster response, emergency management, agriculture monitoring, and urban planning.
Key Highlights & Technical Achievements:
Built a UNet-ResNet34 architecture that delivers high-precision binary and multi-class segmentation on satellite images.
Designed a complete AI pipeline including data preprocessing, image augmentation, model training, real-time inference, and visualization of flood masks.
Deployed the model as a production-ready REST API using FastAPI, enabling instant predictions via a single API call (supports image upload and returns segmented flood map in seconds).
Implemented efficient post-processing and overlay visualization so users can immediately see flooded regions overlaid on the original satellite image.
Optimized the pipeline for speed and scalability, making it suitable for real-world deployment on cloud infrastructure (AWS/Azure ready).
What This Project Demonstrates:
Strong expertise in Computer Vision and Semantic Segmentation using modern architectures (UNet with ResNet34 backbone).
Hands-on experience with satellite/remote sensing data and real-time ML deployment.
Ability to take a complex AI problem from raw data → trained model → live production API.
Practical skills in building solutions that solve actual environmental and disaster-related challenges.
Technologies Used:
PyTorch, UNet-ResNet34, FastAPI, NumPy, Matplotlib, OpenCV, Albumentations, AWS (deployment-ready)
Live Demo / GitHub:
https://github.com/santhanpusala/FloodSense?tab=readme-ov-file
Key Highlights & Technical Achievements:
Built a UNet-ResNet34 architecture that delivers high-precision binary and multi-class segmentation on satellite images.
Designed a complete AI pipeline including data preprocessing, image augmentation, model training, real-time inference, and visualization of flood masks.
Deployed the model as a production-ready REST API using FastAPI, enabling instant predictions via a single API call (supports image upload and returns segmented flood map in seconds).
Implemented efficient post-processing and overlay visualization so users can immediately see flooded regions overlaid on the original satellite image.
Optimized the pipeline for speed and scalability, making it suitable for real-world deployment on cloud infrastructure (AWS/Azure ready).
What This Project Demonstrates:
Strong expertise in Computer Vision and Semantic Segmentation using modern architectures (UNet with ResNet34 backbone).
Hands-on experience with satellite/remote sensing data and real-time ML deployment.
Ability to take a complex AI problem from raw data → trained model → live production API.
Practical skills in building solutions that solve actual environmental and disaster-related challenges.
Technologies Used:
PyTorch, UNet-ResNet34, FastAPI, NumPy, Matplotlib, OpenCV, Albumentations, AWS (deployment-ready)
Live Demo / GitHub:
https://github.com/santhanpusala/FloodSense?tab=readme-ov-file