jeongwhanchoi / MLND-Capstone-Project

Capstone Project for Udacity Machine Learning Nanodegree

MLND-Capstone-Project

Iceberg Classifier

This is a repository for the capstone project going on in Udacity. The project received data from Statoil/C-CORE Iceberg Classifier Challenge at the Kaggle Competitoin. The project goal is to distinguish between iceberg and ship from SAR image data from satellites. For this, deep learning will be conducted using CNN model.

Software and Libraries

• Python 3.54
• Jupyter Notebook
• Tensorflow
• Keras
• Numpy
• Matplotlib

Datasets

The datasets are hosted in a Kaggle competition. Please download data from the link provided.

This following link contains both train(train.json.7z) and test(test.json.7z) data. https://www.kaggle.com/c/statoil-iceberg-classifier-challenge/data

3D image samples for the training data

Iceberg

Ship

iPython Notebooks

• Exploratory Visualization
  • 3D Images
    • Model_00
  • Global Image Stats
    • Explore_Image_Stats
  • Correlations Between Features
    • Explore_Image_Stats
• Image Processing
  • Sample_Images
• Benchmark CNN Model
  • Model_00
• Improved CNN Model
  • Model_01
• VGG16 CNN Model
  • K=3
    • Model_03-SGD-k=3
  • K=5
    • Model_03-SGD-k=5
  • K=7
    • Model_03-SGD-k=7
  • K=10 (Final CNN Model)
    • Model_03-SGD-k=10

Final Model I proposed

• VGG16 CNN Model
  • Final Classifier Model
    • Model_03-Adam-k=10_amsgrad-0001-15p

Transfer Learning Model

Example Results

Model Evaluation