Built an end-to-end machine learning system to classify satellite imagery by land use. Upload any satellite image through the web UI and the model returns the predicted land-use class in real time.
Trained ResNet-18 image recognition model on the EuroSAT satellite dataset to identify land-use categories. Uploaded images are preprocessed with OpenCV (CLAHE contrast enhancement) before inference. Containerized the PyTorch / FastAPI with Docker.
github.com/anthonykovari/eurosat_classification ↗Demo
The same model powers a full Chicagoland land-use change tracker — real Sentinel-2 imagery classified across a 13,000 km² AOI with a year slider and split-screen compare mode.
View live map ↗