I would like to add a couple of broad comments:

1. Keep in mind that 'DN' stands for 'Digital Number' and has nothing to do with an image being calibrated or not calibrated or raw spectral band or ratio or PCA results or anything else.  EVERY image assigns a DN value to the pixels.  What is important is what the DN values represent !  If it is raw Landsat TM or drone image data it is the radiance received at the sensor and adjusted / converted to a digital number range based on the specifications of the sensor typically using gains and offset (or bais) values; if an image has radiometric calibration procedures applied then it can represent TOA radiance, TOA reflectance, surface reflectance, or what ever (even for DEM images the pixel values are DNs that in this case represent elevation).  Again, every image (no matter what they represent) has DN values and I would suggest that 'DN' not be used to imply raw image data only as it is being used in this conversion.

2. Since the early days of Landsat MSS 'broad' spectral bands have been used for digital remote sensing analyses and mapping, so I do not agree that your drone aerial images can only be used for photo interpretation.  I worked with some of the very first Landsat MSS images in early 1972 and have since then used many 'broad' band spectral image data sets for various application.

Pat

LANDSAT is a pushbroom sensor calibrated in an integrating sphere to relate digital numbers (DN) to a value of real physical meaning, ie radiance values. After modelling the scene illumination conditions and atmospheric contributions they are able to derive reflection values, which are object specific and not related to external factors. The INFRAGRAM (frame) camera gives you plain DN. They indicate to “calibrate” the camera before you take-off, but illumination conditions are surely not identical on the ground vs in the air.

That leads us to the “NDVI” calculations: NIR-RED/NIR+RED. With NIR the spectral reflection value (not DN around 720!) around 800nm and RED the spectral reflection value around 650nm (so not a DN derived from a BLUE band in case of the infragram camera). As such I would not call the infragram vegetation index NDVI, maybe DIYVI?

One important aspect is to try and make sure that you are working with RAW images. If yes, there are many approaches to process a RAW image file, including the one described in the following paper. Good luck.

Conference Paper Estimation of the spectral sensitivity functions of un-modif...

One thing to keep in mind is that radiometric calibration and absolute NDVI values are NOT required for every application.  If you are not using multi-temporal images or using a library of spectral reflectance to match DN values then more than likely you do not need to worry about converting DNs to reflectance values.  If you are using multi-temporal images then for some (perhaps many) applications a relative radiometric calibration is all that is required (that is, make one of your images the radiometric master and slave the other images onto the master --- several ways to do this). 

If you think that for your application you do require absolute radiometric calibration then perhaps slave the drone images onto radiometric calibrated Landsat TM image (you'll have to digitally compress the drone image spatial resolution to match the TM resolution).

Good luck with your work.

Pat

The above comments are covering one side of the story - I tend to agree that for one time analyses of indices it is OK to use DN. People can check indices values calculated by DN in comparison to those calculated after radiometric correction and transforming to relative reflectance and see if they can live with the difference....

Another issue might be an outcome of using two different images - RGB and NIR. Are these two images georectified?

I also agree with the use of DN for individual scenes to compute radiometric indices. However, if you are analyzing time-series the issue has an additional dimension.

Could you provide more details about the drone used (whether it has it own processing software, and its capabilities) and more importantly details about the camera used and its output format and proprietary processing software if any ... to see if I can help more.

Drone images are tricky to preprocess before using for any application.

Have you tried Agisoft-Photoscan-PRO. A colleague of mine used it to preprocess her drone images.

Hi,

your images are composite (colour) RGB images or three separate layers (R,G,B) + NIR layer? 

Hi,

As Kazimierz mentioned, you should first indicate that you have 3 bands or it is a composite band, if you have 3 separate bands then you have different ways to obtain reflectance; for example you can use a calibrated data as a reference to normalize your data regarding to it; another solution is based on your drone characteristics and other parameters related to the surveying time and conditions.

regards,

Morteza

Hello!

There is a R package named 'Phenopix' which may help your work.

Conference Paper Phenopix: a R package to process digital images of a vegetation cover

Article Phenopix: A R package for image-based vegetation phenology

Hello all, Thank you so much for all the responses!

my images are RGB (true colour), as one raster layer, and the RG -NIR (800 - 900nm) in another raster layer. I will look into all the methods that some of you suggested. 

And yes, I am stitching my drone images with Agisoft photoscan pro. 

Since we did not take any reflectance values for calibrating the DN, I will probably be using DN for analysis. I am not working with a time-series; I just have two time periods - before flowering and after flowering. We are interested to look at how multispectral images can help in the detection and quantification of these flowers in grasslands and meadows. 

Any suggestions of what indices other than NDVI, or any quantification methods are suitable, are most welcome! we also have another camera, taking false colour images of GB-NIR (680 - 800nm). 

Once again thank you for the inputs!

regards, 

Jessica

I think that the method suggested by Chavez, is good enough to achieve your objectives,

good luck,

Hi Jessica,

So, you do not have multispectral images (narrow bands). You have just RGB and quite broad NIR band. This is not material that is used to do Remote Sensing. These data you can use to perform photointerpretation only.

I would like to add a couple of broad comments:

1. Keep in mind that 'DN' stands for 'Digital Number' and has nothing to do with an image being calibrated or not calibrated or raw spectral band or ratio or PCA results or anything else.  EVERY image assigns a DN value to the pixels.  What is important is what the DN values represent !  If it is raw Landsat TM or drone image data it is the radiance received at the sensor and adjusted / converted to a digital number range based on the specifications of the sensor typically using gains and offset (or bais) values; if an image has radiometric calibration procedures applied then it can represent TOA radiance, TOA reflectance, surface reflectance, or what ever (even for DEM images the pixel values are DNs that in this case represent elevation).  Again, every image (no matter what they represent) has DN values and I would suggest that 'DN' not be used to imply raw image data only as it is being used in this conversion.

2. Since the early days of Landsat MSS 'broad' spectral bands have been used for digital remote sensing analyses and mapping, so I do not agree that your drone aerial images can only be used for photo interpretation.  I worked with some of the very first Landsat MSS images in early 1972 and have since then used many 'broad' band spectral image data sets for various application.

Pat

please see the attached doc. This may serve your purpose for understanding pl.

ansar

To convert DN values to reflectance,

First convert DN values to Radiance and then Radiance to Reflectance.

there are techniques available:

1. FLAASH

2. QUAC

3. ATREM

4. ATCOR

for more details about atmospheric correction and other algorithms for hyperspectral data processing and analysis please read this paper:

https://www.researchgate.net/publication/321227500_A_Research_Review_on_Hyperspectral_Data_Processing_and_Analysis_Algorithms

Dear briefly, you need to convert DN values to Radiance then to Reflectance.

Regards

@ Jessica Tay Ying Ling what approch to you use for converting spectral bands to reflectance (composite RGB)? Works? I have the same problem. Thanks in advanced!

@Alexandre @ Jessica

A Rough Answer: Basically there are four sensors that measure energy in the red-, green-, blue-, and infrared- portion of the spectrum. If the maximum wavelength of each sensor is known, energy in “DN” units may be assigned to it. People initially sensed their environment based on the detectors they were given. Broadly speaking if the red green and blue sensors corresponded to those in the eye, we could see the “color” of the object viewed by the drone. We’d expect something green to have more vegetation than something blue. Only two numbers are needed to quantify the color we see. Here we (the drones) also have an infrared detector. . Suppose we were critters starting out with all four detectors. ... How would the world look to us?

The attached paper provides a way of answering that question; a way of seeing the world with new detectors. A robust formula for comparing the shapes of finitely sampled spectra is presented. Only three numbers would be needed to quantify the “color” we’d see. These can be easily graphed.

Good Luck

(To be clear, I am not talking about false color, but of a way of digesting the full spectrum of a target using many detectors.)

Preprint A Model of Color Vision for Remote Sensing

@ Jack Kornfield thank you very much!!!! Best wishes!!!

@Alexandre

You are certainly welcome. I would suggest the ideas in the paper enable a first order approximation to how insects may see "color" in the world. And, one could, for example, compare the “colors” as seen by bees to those to those seen by humans. What “colors” do bees respond to?

Good luck

@ Jack Kornfield I will study our paper and I will make a lot of questions, this is sure!! Thanks again!!

From DN to Delightful Graph

A measure of the shape of the upwelling radiation, albeit “distorted,” can be determined from digital numbers, DN.

For Landsat’s four sensors, the shape of the upwelling radiation may be described by three derivatives. A way of calculating the derivatives is shown in:

https://www.researchgate.net/publication/324455514_A_Model_of_Color_Vision_for_Remote_Sensing

For the four Landsat channels, three derivatives can be calculated for Landsat data if we consider the four Landsat sensors modeled as “top-hats” centered at their average wavelength (see attached) and each with a measured radiation is assigned a “DN” value.

Using data from Borden, et alia, pages 145 and 151 (attached) and assuming a band-pass filter with the same shape as the eye centered at 8.25 micrometer the derivatives may be calculated and plotted on an x,y,z coordinate system. After converting the x,y,z coordinate system on a r theta phi coordinate system the following graphs are obtained.