EuroSAT: Land Cover Classification with TensorFlow

This project provides a framework for training and evaluating a convolutional neural network (CNN) for land cover classification using the EuroSAT dataset.

README Breakdown

This README will guide you through the functionalities of the project and provide instructions for running it.

1. Introduction

• The EuroSAT dataset is a large-scale Earth observation dataset consisting of satellite images categorized into 10 different land cover classes. This project leverages TensorFlow to build and train a Convolutional Neural Network (CNN) model for accurate land cover classification.

2. Installation

• Required libraries (TensorFlow, TensorFlow Datasets, matplotlib, NumPy) are listed in requirements.txt.
• Installation of required libraries: pip install -r requirements.txt

3. Data Preprocessing (preprocessing.py)

• Data Loading
  • Loads the "eurosat/rgb" dataset from TensorFlow Datasets, focusing on the RGB channels for simplicity. The dataset "eurosat/all" includes 13 bands, but is much larger; modeling requires more computation resources.
  • Splits the dataset into training and testing sets (80% training, 20% testing).
• Data Normalization:
  • Converts image pixel values to float32.
  • Normalizes pixel values to the range [0, 1] by dividing by 255.
• Data transformations applied using tf.data:
  • map: applies normalization and keeps labels.
  • cache: caches the transformed data for faster access.
  • shuffle: shuffles the training data for randomization.
  • batch: groups data points into batches of 32 for training.
  • prefetch: prefetches data for efficient pipeline execution.

4. Data Exploration (data_exploration.py)

• Prints information about the dataset features and structure.
• Displays sample images and their corresponding labels.
• Calculates the distribution of samples across different land cover classes.

5. Model Building and Training (classification.py)

• The architecture of the CNN model was built using tf.keras.Sequential.
  • Convolutional layers (Conv2D) with ReLU activation for feature extraction.
  • Max pooling layers (MaxPooling2D) for downsampling.
  • Flatten layer to convert the image data into a 1D vector.
  • Dense layers with ReLU activation and Dropout for classification.
  • Output layer with 10 units and softmax activation for 10 land cover classes.
• The compilation step used optimizer (Adam), loss function (sparse_categorical_crossentropy), and metrics (accuracy).
• The training process used model.fit with a specified number of epochs (initially 10) and validation data (test set).
• The model is saved as a .keras file with potential error handling using try-except.
• The resulting model is briefly evaluated on test accuracy using model.evaluate.

6. Model Evaluation (evaluation.py)

• The saved model .keras is loaded.
• The model is then evaluated on various metrics, such as using model.evaluate on the test set and printing the test accuracy.
• In its current form, the model has a test accuracy of 78.78%

7. Usage

• Instructions on how to run the different scripts:
  • Preprocessing: python preprocessing.py
  • Data Exploration: python data_exploration.py (will generate output files)
  • Training: python classification.py (trains and saves a model)
  • Evaluation: python evaluation.py (loads and evaluates the saved model)

8. Additional Notes

• Possible further functionalities planned for the project include hyperparameter tuning, visualization of training progress.
• References:
  • EuroSat dataset: https://github.com/phelber/eurosat
  • EuroSat Tensorflow documentation: https://www.tensorflow.org/datasets/catalog/eurosat

9. License

• Placeholder for now.