Christian Knoth - Introduction to Deep Learning in R for analysis of UAV-based remote sensing data

Summary: The aim of this tutorial is to develop a basic understanding of the key practical steps involved in creating and applying a convolutional neural network (CNN) for image analysis – and how to do that in R. These steps are: - Building your model - Preparing your data - Training your model - Predicting with your model Besides the basic workflow, we will discuss two strategies for tackling small data problems, which is specifically important when working with UAV-based data: data augmentation and transfer learning. In addition, we will look at aspects that are important for many remote sensing applications of CNNs: we´ll develop a model for pixel-by-pixel classification (instead of image classification) using an architecture called “U-net”. We will also address the practical question of how to turn a remote sensing image into something that can be processed by our CNN, and how to reassemble the predictions back to a map. Finally, we will briefly touch on the topic of inspecting what a trained model has learned. Installation instructions & material: https://github.com/DaChro/ogh_summer_school_2020 References: Chollet, F., and J.J. Allaire. 2018. Deep Learning with R. Manning Publications. Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. 2015. “U-Net: Convolutional Networks for Biomedical Image Segmentation.” In Medical Image Computing and Computer-Assisted Intervention – Miccai 2015 How to cite this video: http://doi.org/10.5446/49550