In this liveProject, you’ll fill the shoes of a data scientist at UNESCO (United Nations Educational, Scientific and Cultural Organization). Your job involves assessing long-term changes to freshwater deposits, one of humanity’s most important resources. Recently, two European Space Agency satellites have given you a massive amount of new data in the form of satellite imagery. Your task is to build a deep learning algorithm that can process this data and automatically detect water pixels in the imagery of a region. To accomplish this, you will design, implement, and evaluate a convolutional neural network model for image pixel classification, or image segmentation. Your challenges will include compiling your data, training your model, evaluating its performance, and providing a summary of your findings to your superiors. Throughout, you’ll use the Google Collaboratory (“Colab”) coding environment to access free GPU computer resources and speed up your training times.
This project is designed for learning purposes and is not a complete, production-ready application or solution.
This course is for intermediate Python programmers who know basic data science techniques. You’ll augment your skills with some cutting-edge machine learning methods in demand across industry. To begin this liveProject, you will need to be familiar with:
- Basic Jupyter Notebook
- Intermediate NumPy
- Intermediate Matplotlib
- Basic SciPy
- Basic pandas
- Intermediate Python package installation using conda and pip
- Basics of neural networks or multi-layer perceptrons
- Basic concepts in using digital imagery for environmental monitoring
you will learn
In this liveProject, you’ll learn to build a large-scale deep learning model to classify satellite imagery. You’ll develop techniques for handling satellite imagery data, performing geospatial operations on your data such as coordinating transformations, and optimizing model training algorithms by tuning model architecture, learning rates, loss functions and other model hyperparameters. These skills can easily be applied to a wide variety of deep learning tasks using imagery, and beyond.
- Accessing cloud data servers to download satellite imagery
- Manually creating your own ground truth data from imagery
- Using the VGG-JSON image annotation format
- Using Graphical Processing Unit (GPU) computation on Google Colab
- Merging imagery and performing operations on raster datasets
- Using Keras and TensorFlow for deep learning
- Evaluating model performance by comparing estimated and observed results
- Data augmentation for boosting model training
- Optimizing model performance using experimentation
- Understanding model performance metrics (such as Dice and Jaccard scores)