eurosat

Satellite images classification

ANN to SNN conversion on land cover and land use classification problem for increased energy efficiency.

A fast and easy-to-use Remote sensing Image format COnverter for High-throughput Deep-Learning (rico-hdl).

Federated Learning in Satellite Constellations using Flower

Interactive web app for land use classification from Sentinel-2 satellite imagery using deep learning.

WildSAT is an ecological AI system that lets users search satellite imagery using natural language queries like "dense tropical forest" or "wildfire-prone vegetation" via a CLIP-style contrastive pipeline, using a ResNet18 + ViT-B/16 hybrid encoder and DistilBERT as text encoder, visualized on an interactive Mapbox satellite map.

Multispectral satellite image classification on EuroSAT using Swin Transformer, Res2Net, and CSWin with SHAP explainability — Apache-2.0

Trained a ResNet50 model on the EuroSAT satellite imagery dataset w/ PyTorch. Analyzed the model's encoder by visualizing linear interpolations within the embedding space to illustrate the semantic separation in the learned feature representations.

A Geo-AI engine for automated ESG and supply chain monitoring. This project uses a ResNet-101 model, fine-tuned on the EuroSAT satellite dataset, to classify land use and detect environmental risks like deforestation, helping enterprises meet regulatory requirements and enhance transparency.

Residual Network implementation for classifying satellite images from EuroSAT dataset

Trabalho da disciplina de Sensoriamento Remoto (GEO05038) de 2022/1

Drift monitoring and automated retraining for a EuroSAT land-use classifier in production

Satellite image land-use classification using PyTorch and ResNet18 trained on the EuroSAT dataset.

Satellite image restoration benchmark with synthetic corruptions, EuroSAT data, residual U-Net baseline, experimental conditional DDPM, metrics, reports, API, and Streamlit demo.

Satellite image land-use classifier & temporal change detector using transfer learning (ResNet-18), embedding-based change detection with heatmaps, and an interactive Streamlit dashboard.