508 pages | 153 Color Illus.
The text is focused on the development and implementation of statistically motivated, data-driven techniques for digital image analysis of remotely sensed imagery and features a tight interweaving of statistical and machine learning theory with algorithms with computer codes. It develops statistical methods for the analysis of optical/infrared and synthetic aperture radar (SAR) imagery, including wavelet transformations, kernel methods for nonlinear classification, as well as an introduction to deep learning in the context of feed forward neural networks. The material is self-contained and illustrated with many programming examples, all of which can be conveniently run in a web browser.
Each chapter concludes with exercises complementing or extending the material in the text. Numerous examples of programming the Google Earth Engine and TensorFlow APIs are given. New in the fourth edition is an in-depth treatment of a recent sequential change detection algorithm for polarimetric SAR image time series. The accompanying software consists of Python (open source) versions of all of the main image analysis algorithms, thus accessible to all readers with a computer and an Internet connection.
Images, Arrays, and Matrices. Image Statistics. Transformations. Filters, Kernels and Fields. Image Enhancement and Correction. Supervised Classification Part 1. Supervised Classification Part 2. Unsupervised Classification. Change Detection. Mathematical Tools. Efficient Neural Network Training Algorithms. Software