Skip to content

This project applies deep learning and transfer learning to classify Martian HiRISE images for frost detection. A custom CNN + MLP model and pre-trained networks (EfficientNetB0, ResNet50, VGG16) were used to improve classification accuracy.

Notifications You must be signed in to change notification settings

rajnandinithopte/Machine-Learning_MarsFrostDetection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

4 Commits
Β 
Β 
Β 
Β 

Repository files navigation

Machine-Learning: Mars Frost Detection

πŸ”· Identification of Frost in Martian HiRISE Images

πŸ”Ά Overview

This project focuses on binary classification of Martian terrain images to detect frost using deep learning models. A custom CNN + MLP model and transfer learning (EfficientNetB0, ResNet50, VGG16) were used for feature extraction and classification. Data augmentation, dropout, batch normalization, and L2 regularization were applied to improve model generalization.

πŸ”· Dataset Used

  • HiRISE Mars Terrain Images (Dataset Link)
  • The dataset contains 119,920 image tiles extracted from 214 HiRISE subframes.
  • Each image tile is labeled as either "frost" or "background".
  • Provided splits for training, validation, and testing.

πŸ”· Libraries Used

  • TensorFlow/Keras - Building deep learning models.
  • OpenCV - Image augmentation (cropping, zooming, rotating, flipping).
  • Matplotlib & Seaborn - Data visualization.
  • scikit-learn - Model evaluation metrics.
  • Pandas & NumPy - Data processing and numerical computations.

πŸ”· Steps Taken to Accomplish the Project

πŸ”Ά 1. Data Preprocessing & Augmentation

  • Resized images to 299x299 pixels.
  • Applied image augmentation: random cropping, zooming, flipping, contrast adjustment, and translation.
  • Normalized pixel values between 0 and 1.

πŸ”Ά 2. Training a CNN + MLP Model

  • Built a 3-layer Convolutional Neural Network (CNN) followed by a Multi-Layer Perceptron (MLP).
  • Used ReLU activation for all layers.
  • Applied softmax function for binary classification.
  • Regularization techniques used:
    • Batch Normalization
    • Dropout (30%)
    • L2 Regularization
  • ADAM optimizer and cross-entropy loss were used.
  • Model trained for at least 20 epochs, with early stopping based on validation loss.
  • Precision, Recall, and F1-score were reported.

πŸ”Ά 3. Transfer Learning with Pre-trained CNNs

  • Utilized EfficientNetB0, ResNet50, and VGG16 for feature extraction.
  • Froze all layers except the final fully connected layer.
  • Extracted features from the penultimate layer and trained a classifier.
  • Used ReLU activation, batch normalization, dropout (30%), and softmax activation.
  • Trained for at least 10 epochs (preferably 20 epochs) with early stopping.
  • Compared results with CNN + MLP model.

πŸ”Ά 4. Model Evaluation & Analysis

  • Reported Precision, Recall, and F1-score for all models.
  • Compared CNN + MLP vs. Transfer Learning performance.
  • Plotted training and validation loss curves to analyze convergence.

πŸ”Ά 5. Findings & Comparison

  • CNN + MLP required more training data but performed well with augmentation.

  • Transfer Learning (EfficientNetB0, ResNet50, VGG16) achieved higher accuracy due to pre-trained feature extraction.

  • EfficientNetB0 provided best performance in terms of validation loss and classification accuracy.


πŸ“Œ Note

This repository contains a Jupyter Notebook detailing each step, along with results and visualizations.

About

This project applies deep learning and transfer learning to classify Martian HiRISE images for frost detection. A custom CNN + MLP model and pre-trained networks (EfficientNetB0, ResNet50, VGG16) were used to improve classification accuracy.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published