Jupyter Notebooks
Statt nur theoretische Grundlagen zu lehren, bevorzuge ich es, auch praktische Aspekte in meinen Kursen abzudecken. Hierzu verwende ich Jupyter Notebooks, die sich als sehr nützlich in der Lehre erwiesen haben, insbesondere wenn sie in cloudbasierten Rechenumgebungen wie Binder oder Google Colab eingesetzt werden.
Es folgt eine Liste derzeit verfügbarer Jupyter Notebooks, die ich in meinem Kursen verwende und die all auf meinem GitHub-Lehrrepository zu finden sind. Alle diese Notebooks sind unter der MIT-Lizenz frei verfügbar, sodass sie von jedem verwendet werden können.
Die Notebooks sind alle in englischer Sprache und sind grob nach den folgenden Themen sortiert.
Python
Introductory Notebooks for learning the Python programming language.
Basics: A general introduction into the basics of the the Python programming language. Interactivity: low. Prequisites: None
Numpy and Matplotlib: A more detailed introduction into the Numpy and Matplotlib modules. Interactivity: low. Prequisites: basic Python
Pandas: An introduction into Pandas module for data processing and analysis. Interactivity: low. Prequisites: basic Python
Data Processing
Noteboks related to the processing of different data modalities.
- Feature Extraction: An interactive Notebook introducing different feature extraction techniques for different data modalities. Interactivity: high, Prerequisites: some experience with Python
Clustering
Notebooks related to different clustering applications.
Clustering and PCA with sklearn: A brief introduction into unsupervised learning methods such as clustering and Principal Component Analysis. Interactivity: medium. Prerequisites: Numpy experience.
Clustering and Segmentation for Sentinel-2 Satellite Imagery: This example contains segmentation examples based on k-Means clustering and SLIC for a small Sentinel-2 multispectral dataset. Interactivity: medium. Prerequisites: some experience with Python.
Machine Learning Concepts
Introductions to different supervised learning concepts.
Preparing Data for Supervised Learning: This Notebook serves as an interactive worksheet for preparing a dataset for use in a machine learning model. It will guide you through the steps required to use a dataset in a supervised learning setup. Interactivity: high. Prerequisites: Numpy experience.
Full Supervised Learning Pipeline: This Notebook combines all the pieces together to perform a classification task with scikit-learn. Interactivity: low. Prerequisites: some experience with Python.
Data Labeling: This tutorial introduces the data labeling process for multiband imaging data. The data are used for a pixel-wise classification of Sentinel-2 data. Interactivity: medium. Prerequisites: some experience with Python and familiarity with the “Pixel-wise classification with Machine Learning”-Notebook.
Data Augmentations: Building on the “Semantic Segmentation of Sentinel-2 Images with a UNet”-Notebook, this Notebook introduces data augmentations and experiments with a few simple transformations. Interactivity: medium. Prerequisites: experience with Python and semantic segmentation using a UNet; running this Notebook requires access to a GPU!
Transfer Learning: Building on the “Image Classification with Convolutional Neural Networks”-Notebook, this Notebook introduces the concept of transfer learning. Interactivity: low. Prerequisites: experience with Python and semantic segmentation using a UNet; running this Notebook requires access to a GPU!
Regression
Notebooks related to regression tasks.
- Regression on the California Housing Dataset: This is an introductory notebook, featuring different traditional machine learning methods applied to a tabular dataset. Interactivity: medium. Prerequisites: some experience with Python.
Pixel-wise Classification
Notebooks related to pixel-wise classification tasks.
Pixel-wise Classification with Machine Learning: This example showcases the use of traditional machine learning methods for pixel-wise land-use/land-cover classification. Simple data annotation by hand, as well as Maximum likelihood estimation and k-nearest neighbors methods are introduced for this purpose. Interactivity: medium. Prerequisites: some experience with Python.
Pixel-wise Classification with a Multilayer Perceptron: This tutorial introduces the use of Multilayer Perceptrons for pixel-wise classification for land-use/land-cover classification. Interactivity: low. Prerequisites: some experience with Python; running this Notebook requires access to a GPU!
Image-wise Classification
Notebooks related to image classification tasks.
Image Classification with a Multilayer Perceptron: In this Notebook we build a Multilayer Perceptron from scratch using Pytorch and train it on a simple image classification task based on the FashionMNIST dataset. Interactivity: low. Prerequisites: some experience with Python; running this Notebook requires access to a GPU!
Image Classification with a Convolutional Neural Network: This example introduces the use of Convolutional Neural Networks for the task of image classification. Interactivity: low. Prerequisites: some experience with Python, familiarity with the “Pixel-wise classification using machine learning”-Notebook would be useful; running this Notebook requires access to a GPU!
Semantic Segmentation
Notebooks related to semantic segmentation tasks.
- Semantic Segmentation of Sentinel-2 Images with a UNet: This Notebook introduces the workflow for supervised learning with a UNet architecture available in Pytorch. We will us dense labels available in the ben-ge-800 dataset for this task. Interactivity: low. Prerequisites: some experience with Python and familiarity with the “Pixel-wise Classification with a Multilayer Perceptron for Sentinel-2 Satellite Imagery”-Notebook; running this Notebook requires access to a GPU!
Object Detection
Notebooks related to object detection tasks.
- Object Detection with YOLO for Aerial Imagery: This Notebook introduces object detection using YOLO for aerial imagery. We will detect cars from aerial imagery of the city of Stuttgart. Interactivity: low. Prerequisites: some experience with Python; running this Notebook requires access to a GPU!
Remote Sensing
Notebooks related to general remote sensing tasks.
- Thermal Imaging with flyr: Flyr is a library for extracting thermal data from FLIR images written fully in Python. We use this library to read in, modify and analyse thermograms. Interactivity: medium. Prerequisites: some experience with Python.
Eine aktuelle Liste der vorhandenen Notebooks gibt es auf github.