This repository contains a satellite image classifier that identifies signs of deforestation and pollution using transfer learning with a pre-trained ResNet50 Convolutional Neural Network (CNN) model. The model is fine-tuned for binary classification tasks to detect 'habitation' as a proxy for deforestation and 'slash_burn' as an indicator of pollution in Amazon rainforest satellite images.
The dataset used for this project is sourced from the Amazon Deforestation from Space Kaggle dataset. The dataset contains multi-label satellite images taken from the Amazon rainforest, with each image associated with one or more tags.
Python 3.7+
TensorFlow 2.0+
pandas
scikit-learn
Keras
Matplotlib
-
Download the dataset from Amazon Deforestation from Space and extract the contents to a directory named data.
-
Run the satellite_image_classifier.py script to train the model:
python satellite_image_classifier.py
-
After training, the model will be saved as 'model.h5'. You can use the predict_deforestation_pollution() function in the satellite_image_classifier.py script to make predictions on new images.
This project is licensed under the MIT License. See the LICENSE file for details.