ABSTRACT
This chapter begins with something easy: a small convolutional neural network. It looks at the data that will be employed to train and use our network. The document shows how to run the Orfeo Toolbox (OTB) applications from the Command Line Interface. However, it is possible to switch to the Graphical User Interface: to do so, just type “otbgui_ApplicationName” instead of “otbcli_ApplicationName”. The PatchesExtraction application integrates seamlessly in the existing sampling framework of OTB. The chapter provides overview of the basic concepts for the training of deep learning models. Model training usually involves a gradient descent of a loss function that typically expresses the gap between estimated and reference data. The goal of the training is to minimize the cross entropy between the data distribution and the model distribution. The chapter utilizes the cross entropy of the softmax function of the 6 neurons as the loss function.
