Object-oriented image analysis

Lecturer

Michael Wurm

ECTS

5 ECTS

 

Aim

In this course we will learn the alternative image analysis paradigm of object-based image classification. Image objects are areas in images which consist of pixels from the same land-cover class (e.g. buildings, water surfaces, vegetation) and allow the application of additional image classification methods in comparison to pure pixel-based methods and the integration of multi-modal date (e.g. vector data).

 

Content

In the course we will use satellite images and high resolution aerial images in combination with vector data and analyze various image classification methods (nearest neighbor -> random forests -> deep learning) to extract relevant information from the images.
We will use the software “eCognition”.
The course will be held on 3 days in Würzburg plus an introductory lecture which is being held online.
Session 1: getting to know to image objects and eConition software, basic classification methods
Session 2: advanced classification (Machine Learning)
Session 3: Deep Learning

 

Coding

Coding examples and individual work will be covered

Software

Various software programs will be used, but mainly OpenSource software such as R.

Techniques

Different techniques will be introduced and practically applied.
w

Content

The content of scientific with regard to the audience will be discussed.

General Course News and Updates