When analyzing remote sensing images in (very) high spatial resolution, the investigated objects are usually much larger than the pixel size, i.e., a single target object is represented by a lot more than only one pixel in the image. This means that in addition to spectral properties of single pixels we can (and often need to) extract further information from the spatial arrangement of pixel values by analyzing features such as texture, shape or edge intensity. In addition, the spatial context of objects can play an important role during the analysis.
In this course we will explore methods for extracting such features from remote sensing images based on examples from research and practical applications. We will also look at how to leverage these features for the segmentation and classification of remote sensing data through techniques like geographic object-based image analysis (GEOBIA) and deep learning with convolutional neural networks (CNNs).
The course consists of three parts running in parallel: a lecture part, a discussion part (where students read and discuss research papers applying these methods), and a practical part (where students apply the methods themselves).
Details about the organization and grading will be given during the seminar.
- Lehrende/r: Christian Knoth