In this project, I am trying to build a classifier that distinguishes images of Martian terrain with frost. You can find the dataset here This dataset was created to study Mars’ seasonal frost cycle and its role in the planet’s climate and surface evolution over the past 2 billion years. The data helps in identifying low-latitude frosted microclimates and their impact on climate.

Objective

• Train a conventional Convolutional Neural Network combined with a multi-layer perceptron model.

• Construct a Transfer Learning model for the identical scenario.

• Become acquainted with image processing libraries such as OpenCV or Keras.

What I have done

• Implemented a three-layer CNN, incorporating image augmentation techniques with Keras for empirical regularization. Achieved optimal model parameters through early stopping.

• Utilized transfer learning for model evaluation, comparing precision, recall, F1 score with the CNN + MLP model, offering valuable insights into their relative performance.

Result

Click here to view the result in Jupyter Notebook