Pixel‐based Classification Techniques for Satellite Image Time Series
Charlotte Pelletier, Silvia Valero · 2021
This chapter begins by introducing some fundamentals about supervised classification in the context of remote sensing applications. It includes a description of some key data preprocessing steps and the classical evaluation procedure based on the confusion matrix. The chapter overviews three well-established classification algorithms massively used for satellite image time series (SITS) classification: support vector machines, random forests and k-nearest neighbors. The SITS data offers a natural time-based representation of the data, which is used in most classification tasks. Remote sensing applications usually involve complex classification problems requiring the use of a large amount of training samples. The chapter presents several classification approaches proposing low-dimensional temporal feature representations of Earth Observation (EO) time series. It describes the phenological temporal features extracted from vegetation indices, bag-of-words approaches and shapelet methods. The chapter details the main deep learning networks used for the classification of EO time series, that is, temporal convolutional neural networks and recurrent neural networks.