Wetlands Segmentation & Cloud Detection from Satellite Imagery
Upload a satellite image or TIFF file to identify wetland areas and detect cloud cover. Optionally, you can also upload a ground truth mask for evaluation.
Input
Load Examples
Click an example below to load files:
Upload Multi-Band TIFF File | Ground Truth Mask (Optional) |
---|
Results
About these models
This application uses two models:
1. Wetland Segmentation Model:
- Architecture: DeepLabv3+ with ResNet-34
- Input: RGB satellite imagery (extracted from first 3 bands of TIFF if provided)
- Output: Binary segmentation mask (Wetland vs Background)
- Resolution: Processed at 128×128 pixels
2. Cloud Detection Model:
- Architecture: LightGBM Classifier
- Input: Coefficient of Variation (CV) features extracted from up to 10 image bands (from TIFF)
- Output: Binary classification (Cloudy vs Non-Cloudy) with probability
Tips for best results:
- Use the 'Upload TIFF' tab for multi-band satellite data to enable accurate cloud detection.
- The cloud detection model expects up to 10 bands. Performance may vary with fewer bands.
- The example files demonstrate cloudy/non-cloudy scenarios with corresponding ground truth.
- The models work best with images similar in characteristics to those used in training.
- For ground truth masks, both TIFF and standard image formats (PNG, JPG) are supported. Ensure the mask clearly delineates the target class.
Repository: dcrey7/wetland_segmentation_deeplabsv3plus