I am trying to classify multi-class land cover and consider the effectiveness of spectral indices such as NDVI. I, however, do not know all the useful indices. Can anyone help me, please?
Useful indices may be NDVI, NDWI, EVI, ...
SVI, NDVI, NDWI, RVI, EVI, PVI, SAVI, ARVI, SARVI, DVI etc.
Hello Phan,
There are several vegetation indices which are used nowadays ( http://www.ccpo.odu.edu/SEES/veget/class/Chap_4/4_5.htm ). However, the comparison and contrast of different vegetation indices depend upon the priority or major goal of any particular research. The following links can be helpful in deciding which one is better for your use:
https://eos.com/blog/6-spectral-indexes-on-top-of-ndvi-to-make-your-vegetation-analysis-complete/
https://www.vineview.com/evi-vs-ndvi-whats-difference/
http://www.un-spider.org/links-and-resources/data-sources/daotm/daotm-vegetation
Cheers
I suggest you follow http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.458.4992&rep=rep1&type=pdf
https://d1wqtxts1xzle7.cloudfront.net/55865759/Coulter_et_al_RSE_Ghana_Landsat_Paper.pdf?1519246306=&response-content-disposition=inline%3B+filename%3DClassification_and_assessment_of_land_co.pdf&Expires=1591782591&Signature=LAbYa088gFUIg~6tvZALqTudgmce89yYkJJKJK6cD1ZnSjavs22vkhnpf1h2vyuweBJi97U7Itpepl-Niu2mLGpcQzaDuXoFmfprMkjPPhhUDe9mgvAvyd7lGf~0mRuscwNEVt7BgouTGLrCMNU9vRLxBCROk3nfskNfCGRzWZgwDOEXGi5lKZENVCLXnSZ1GcCxSBmYTczvupX5BhlsI~alosjLPtFC2pKB~sHDS0AGDWJ50iQPVnkJO3y4OF56lvzuZI6Xm~yn0XYxyNvYOKzS87ClcgzbxZh0JjVffdZvmO31fh7aZaI0PpeV4YUjuaeQj2a-l~9JdytcwJzrqA__&Key-Pair-Id=APKAJLOHF5GGSLRBV4ZA
You can compute several indices (considere those listed in https://www.indexdatabase.de/ for your specific case), if you end with many, you can apply a feature selection algorithm. Here you can find some common VI ( Article A Cloud-Based Multi-Temporal Ensemble Classifier to Map Smal...
)
Hello Duong Cao Phan, If you are interested in automatic multi-class land cover classification, why don't you use unsupervised classification algorithms like k-means or ISODATA for your job?
It classifies the vegetation as one of the class if a large portion of vegetation is included in your data/image. Indeed, it depends upon how good the spectral radiance from vegetation is varying compared to the other land cover features covered in the same image. Moreover, you can select the number of classes you are interested in while performing unsupervised classification.
Note: As already mentioned, the accuracy depends purely on the data. More the variability in spectral radiance between different land cover features, more accurate will be the result. Besides, the spatial resolution will also have an impact on the accuracy of classified results
You can also use those indexes such as: RVI and DVI. The simple ratio vegetation index (termed SR or RVI) is calculated using the following formula.
RVI = NIR/RED.
While the difference vegetation index (DVI), is calculated using this formula:
DVI = NIR-RED.
Duong Cao Phan, if I understand your question correctly and you seek universal multipurpose index for any occasion, I guess its hardly possible. But there are hips of indexes for special purposes, of course....
There are various ways to obtain "better" spectral image classifications beyond using indexes of 2-3 classical spectral bands (alone):
- maximum likelihood classification; (un)supervised (e.g. Gils et al. 2010),
- indexes of narrow spectral bands (e.g. Fauzi et al. 2013) and
- hypertemporal NDVI imagery/ISODATA (Westinga et al. 2020).
You can use RVI, TCI, ARVI, DVI, TCI, SARVI, PVI etc for multi-class land cover classification
- Badges
- Science topic