Skip to content
← Projects

HydroVexel — AI Oil Spill Detection

A deep learning system that detects oil spills from satellite imagery with 94.57% accuracy using U-Net with Attention Gates, deployed on Streamlit.

Built by

GitHubLive demo
PythonTensorFlowU-NetAttention GatesStreamlitOpenCVDeep Learning

Overview

HydroVexel is an AI-powered oil spill detection system that analyzes satellite imagery and instantly identifies oil spills in the ocean — at 94.57% accuracy. Built for the GDG Noida Build-a-thon, it's aimed at protecting marine environments through rapid detection and faster response.

Why It Matters

Oil spills devastate marine life, pollute coastlines, damage fishing and tourism economies, and cause long-term ecological harm. Traditional detection is slow and manual — HydroVexel makes it instant and accurate, so response teams can act faster.

What I Built

  • Real-time detection — upload a satellite image, get an instant result
  • 94.57% accuracy from a U-Net segmentation model with Attention Gates
  • Visual analysis — heatmaps, segmentation overlays, and confidence scores
  • Cloud storage to save and track detections over time
  • Zero-install web app with an ocean-themed, animated UI

How It Works

A modular segmentation pipeline:

  1. Preprocessing — image resizing, normalization, and augmentation
  2. Model — U-Net with Attention Gates focuses the network on spill regions
  3. Inference — pixel-level segmentation produces the spill mask
  4. Visualization — overlays, heatmaps, and a confidence score are rendered back to the user

Tech Stack

ML / CV: Python, TensorFlow, U-Net + Attention Gates, OpenCV

App & Deploy: Streamlit (live), cloud storage for detection history

Use Case

Environmental agencies and maritime responders who need rapid, automated screening of satellite imagery to spot oil spills early — replacing slow manual review with instant, explainable detection.

Outcome

Built for the GDG Noida Build-a-thon; deployed and publicly usable on Streamlit with a measured 94.57% detection accuracy.

Related projects

RAG Document Assistant

A production-style Retrieval-Augmented Generation system — documents are chunked, embedded with SentenceTransformers, indexed in Endee's HNSW vector database, and retrieved via cosine-similarity k-NN search to ground Groq's Llama 3.3 70B in a user-provided knowledge base with full source attribution.

→

Noloop

A glass-box medical claims adjudication and fraud-defense platform — an AI agent pipeline turns unstructured claim documents into explainable, role-specific decisions for patients, hospitals, and insurers.

→

Echo Echo

An AI music generation platform — multi-agent LLM orchestration (CrewAI + Groq Llama 3.1) composes 30-second instrumental sketches via the Suno v4 API, generates thematically aligned lyrics, and scores every output through an automated copyright-similarity detection pipeline.

→