i-ninte/geoforest

Automated forest cover monitoring for Ghana using Sentinel-2 satellite imagery, RF-DETR object detection, and temporal tracking — from raw data pipeline to near-real-time deforestation alerts. # GeoForest 🌳 **Satellite-based monitoring of deforestation and illegal mining (galamsey) in Ghana using deep learning and multispectral satellite imagery** GeoForest is a geospatial machine learning project that uses satellite imagery and computer vision to detect **forest degradation, illegal logging, and galamsey expansion** in Ghana. The project combines **multispectral satellite data, vegetation indices, object detection models, and temporal analysis** to monitor forest canopy changes over time. --- # Motivation Illegal mining (galamsey) and unsustainable logging contribute significantly to **forest loss and environmental degradation in Ghana**. Monitoring these activities at scale is challenging using traditional ground-based methods. Satellite imagery provides an opportunity to build **automated monitoring systems** capable of detecting environmental change across large geographic regions. GeoForest aims to build an **AI-powered monitoring pipeline** that can: * detect vegetation loss * identify possible mining expansion * track forest degradation over time * support environmental monitoring efforts The project leverages data from **Sentinel-2**, which provides multispectral imagery suitable for vegetation monitoring. --- # Project Architecture The system follows a geospatial ML pipeline: ``` Satellite imagery (Sentinel-2) ↓ Data preprocessing & tiling ↓ Vegetation index computation (NDVI) ↓ Object detection model (RF-DETR) ↓ Temporal change analysis ↓ Forest loss / galamsey detection ``` The system integrates **two primary signals**: 1. **Computer vision detection** 2. **Vegetation health monitoring** Combining these improves detection reliability. --- # Dataset The dataset used for this project is publicly available on Kaggle. **Dataset** Ghana Sentinel-2 Forest Dat… The dataset contains **multispectral Sentinel-2 image tiles covering regions of Ghana**, prepared for machine learning workflows. …