Using NDVI along with regular RGB and NIR bands for image classification

Raza_Muhammad · ‎04-10-2019

Is it plausible to use NDVI along with other regular bands for image classification related data processing? Recently, I came across a comment that RED and NIR band might interfere with NDVI or vice versa during the processing. I am working on a project where I want to classify healthy and stressed crop patches using satellite imagery, and I am using all four bands (RGB and NIR) along with NDVI to train the model and classify the satellite images. For the classification, I am using the Forest-based Classification and Regression tool.

GuenterDoerffel · ‎04-11-2019

Hi Muhammad,

not sure if I get this ... but NDVI is a 'derivative' of the red and the NIR band, calculated as : NDVI = (NIR - Red) / (NIR + Red)
This results in values between 0.0 and 1.0 ... and is then often used together with colormaps/colorramps to visualize.

So, as you write above, 'the RED and NIR band might interfere with NDVI and vice versa' ... they are essentially 'dependant variables'.

Regards
Guenter

Raza_Muhammad · ‎04-11-2019

Hi Guenter

Yes, they are mainly 'dependent variables'. And my question was that if we use those together, i.e., Blue+Green+Red+NIR+NDVI for classification, does that pose any problem in our methodology or interfere with the analysis?

I have used Blue+Green+Red+NIR+NDVI as a combination of training rasters for healthy and diseased plants in the crop field and obtained a high and consistent accuracy (high kappa as well) of the classification. However, when I used NDVI alone and the RGB NIR separately, classification accuracy decreased.

Thanks

GuenterDoerffel · ‎04-14-2019

Hi again Muhammad,

to reply to this with a sound statistical/mathematical answer is beyond what I remember from those disciplines. If the results you get are consistent and verified by Ground truth ... I'd consider it proof enough

Regards
Guenter

Raza_Muhammad · ‎04-15-2019

Yes, you are right. Also, as I am using Random Forest, Multicollinearity does not affect prediction accuracy. Thanks, Guenter.