Book Review: ‘Hyperspectral Remote Sensing of Vegetation’
- Published on Wednesday, 21 November 2012 04:12
- 17 Comments
By Dr. Pieter Kempeneers, VITO, Belgium
I have read the book, “Hyperspectral Remote Sensing of Vegetation,” with great pleasure. It provides a comprehensive overview with plenty of useful references to the literature.
On one hand, there is an extensive overview of existing techniques and state-of-the-art methods. In a single reference, the reader is offered an overview of hyperspectral vegetation indices (HVI) in clear tables, without the need to search and browse through a large number of papers to find the appropriate index for his or her particular problem. The authors also provide sufficient detail on the reflectance behavior of vegetation, which gives the reader insight into the physical background of hyperspectral spectroscopy.
On the other hand, there are many practical examples, presented as case studies. The main part of each section is mostly set up in a typical paper style, focusing on the successes of imaging spectroscopy. Nevertheless, most end with conclusions that are more nuanced and show, next to its potential, the problems and future steps to overcome if a true operational procedure is required.
The book consists of contributions from many authors that are experts in their fields. This results in interesting views on the various topics. A potential drawback is that each author provides his or her own introduction, which results in considerable repetition throughout the book. I found very little cross references that went beyond the level of a chapter.
However, whether this is a weak point or rather an asset can be disputed and depends on how the book is read. If used as a reference book (which in my opinion is where its true potential lies), the repetition is rather an advantage for the reader, who is not forced to read through the preceding chapters in order to understand the context. One exception of the lack of cross references is Chapter 28, where the editors provide a unique synoptic view on the material presented in the book. I might have put this chapter at the beginning of the book. I suspect that many readers will not read the book in its entirety and could potentially miss this interesting overview (in my own experience, first chapters are more frequently read than final chapters).
I would highly recommend the “Hyperspectral Remote Sensing of Vegetation” to anyone dealing with the subject. Although minor, I have found a few issues that could be improved in my opinion. In general, I had the impression that there is a lack of coherence. One example that might improve this is adding cross references beyond the level of sections within a chapter. Harmonizing the references (instead of using names and numbers for literature interchangeably) is another. The same is true for numbering equations (no numbers in Chapter 3). Occasionally, there are different acronyms used for the same thing, e.g., narrowband vegetation index (NVIs, on page 397) instead of hyperspectral vegetation index (elsewhere).
The book is very well written, both in style and content. Bearing in mind that the audience of the book might be nonexperts, I would suggest to rephrase some parts in the overview on supervised classifiers (section 4.4.3 on hyperspectral data mining).
Section 22.214.171.124, dealing with maximum likelihood classifiers (MLC), concludes with: “A comparison of MLC, SAM, artificial neural network (ANN), and decision tree classifiers found that MLC had the highest accuracy .” By putting this in an overview of supervised classifiers, non-expert readers could be mislead and get the false impression that the authors intend to raise a general statement, rather than describing a particular case study where one classifier outperformed the other.
Since we know from Duda and Hart there is no free lunch theorem, it should be made clear to the user that each problem is different and there is no such thing as a best classifier. Rather than an illustration of this, some readers might get the impression that the authors contradict themselves only two pages further (page 113), dealing with ANN, which concludes that: “ANN-based classification of land cover types have shown to have higher accuracy than MLC, SAM, and minimum distance classifier .”
The following anecdote illustrates of the usefulness of the book. When I received my copy at work, colleagues were quickly interested and browsing through it. Soon after I took the book home for review, my colleagues kept on asking me when I was returning it to work, so they could start using it.
Pieter Kempeneers is an engineer in electronics and obtained his Ph.D. in physics on information extraction from hyperspectral imagery, applied to vegetation, from the University of Antwerp. Since 1999, he has been working as a researcher in remote sensing with the Flemish Institute for Technological Research (VITO), Belgium. From 2008 to 2011, he was with the Joint Research Centre of the European Commission at Ispra, Italy.