This repository focuses on using Artificial Intelligence and Satellite Imagery to monitor global fish farming. By automating the detection of fish cages, we provide tools to help protect coastal ecosystems from unregulated expansion.
- Deep Learning: Utilizes advanced models (UNet, YOLO, DeepLabv3+) to identify aquaculture structures.
- Multi-Sensor Data: Integrates Optical, Radar (SAR), and Hyperspectral imagery for high-accuracy monitoring.
- Sustainability Focus: Designed to help researchers and environmentalists track the impact of aquaculture on marine environments.
We address common remote sensing hurdles such as:
- Environmental Noise: Overcoming interference from clouds, waves, and water turbidity.
- The "Look-Alike" Problem: Distinguishing man-made cages from natural reefs or dikes.
- Data Scarcity: Improving model generalization across different geographic regions.
Train or retrain segmentation models for cage detection:
- 01_image_collector.ipynb - Collect raw satellite imagery
- 02_image_conversion.ipynb - Convert and prepare images for training
- 03_dataset_split.ipynb - Split data into train/validation/test sets
- 04_model_training.ipynb - Train the segmentation model (UNet-based)
Detect fish cages in new satellite imagery using a pre-trained model:
- CageAnalysis.ipynb - Main inference notebook for cage detection and analysis
- Downloads Sentinel-2 imagery from Copernicus Dataspace
- Runs model inference on satellite images
- Generates output shapefiles with detected cage locations
- Compares results with reference data
- Gather satellite images in your area of interest
- Convert images to appropriate format (notebook 02)
- Split data into train/validation/test (notebook 03)
- Train the model with your dataset (notebook 04)
- Prepare Sentinel-2 imagery (or use automatic download in CageAnalysis.ipynb)
- Run
inference/CageAnalysis.ipynbwith your pre-trained model - Get detected cage locations as shapefiles and visualizations
Tech Stack: Python | Computer Vision | Remote Sensing | Deep Learning