mfdeenn/CNN-Based-Illegal-Mining-Detection

Deep learning model (U-Net + EfficientNet) for detecting illegal mining sites from satellite imagery in Ghana 🌍 GalamseyNet: Deep Learning Detection of Illegal Mining Sites Illegal mining, locally known as “galamsey,” is one of the most serious environmental challenges in Ghana. It leads to deforestation, land degradation, and pollution of major rivers, threatening ecosystems and livelihoods. GalamseyNet is a deep learning–based satellite image segmentation system that detects illegal mining sites in the Amansie Central District (Ashanti Region, Ghana) using Convolutional Neural Networks (CNNs). This project demonstrates how AI + Remote Sensing can support faster, scalable, and data-driven environmental monitoring. 📌 Project Objectives Collect and preprocess satellite imagery of mining regions Build a CNN-based segmentation model to detect illegal mining areas Evaluate model performance using standard computer vision metrics Provide a framework for AI-assisted environmental monitoring 🛰️ Data Collection & Processing Satellite imagery sourced using Google Earth Engine Images processed and prepared in QGIS Manual annotation done in ArcMap to create binary masks: Class 1: Illegal mining areas Class 0: Forest, rivers, roads, settlements Images split into 256 × 256 patches for model training Dataset split: 70% Training 15% Validation 15% Testing 🧠 Model Architecture We implemented a U-Net image segmentation model with an EfficientNet-B3 encoder. Why this architecture? EfficientNet-B3 (Encoder) Pretrained on ImageNet for strong feature extraction from satellite images. U-Net (Decoder) Enables precise pixel-level segmentation and boundary detection. This combination allows the model to effectively distinguish disturbed mining land from natural vegetation and other land cover types. ⚙️ Technologies Used Category Tools Programming Python 3.11 Deep Learning TensorFlow, Keras Training Platform Google Colab (GPU) Geospatial Processing QGIS, ArcMap Data Handling NumPy, Pandas Visualization Matplotlib 📊 Model Performance The model was evaluated using segmentati …