3D Robotics

Farm drones are hot!

3689532145?profile=originalThis magazine isn't online, but you can get it on your iPad, Kindle or Android tablet via NextIssue. The article is a pretty basic overview, with examples of using a Rite Wing Zephyr and a hexacopter along with a Canon s100 modified with the IR filter removed. But it's notable that drones are getting this kind of attention in agriculture. 

E-mail me when people leave their comments –

You need to be a member of diydrones to add comments!

Join diydrones


  • All,

    I have got Canon SX260 converted by MaxMax + their Remote Sense Explorer software. I would like to calculate NDVI based on their image while using Matlab.
    Can you provide any brief directions on the matter?
    I am interested in the data format of the BIN files RSE writes


  • Thank you very much for your answer and congratulations with your new position. You've got a new customer at Roboflight!

  • Cristo -  I cannot answer all your questions right now because some are so camera altitude dependent that it is difficult to answer some questions without referring to some of our tables, and I am in the process of packing my office.  I am retiring from Kansas State University and have accepted the position of Executive Vice President of RoboFlight (www.roboflight.com).  I can tell you when we flew an area approximately 1 x 1 mile at 1 inch resolution, we collected around 400 pictures.  Buy as big of a computer as you can afford with at least 32 gb of memory and better with 64 gb.  We flew an area of about 12,000 acres using a Cessna 172 at 4000 feet and generated about 1100 images and it took a week to mosaic these images on a high end PC Desktop with 32 gb of memory.  If processing of the imagery is not something you want to mess with, contact RoboFlight and we can help you get this done and back in a timely manner. Many want to fly the imagery but lack the image processing skills to deal with the imagery, calibration, etc.  I have been teaching digital image processing at the Ph.D. level for 33 years and so this is something I will be helping with at RoboFlight.

  • Kevin -- As a novice, this forum proved to be invaluable to me, for which I am very grateful! I take the liberty of extending your invitation to Grant to myself...

    Your comment that the bottleneck lies in the calibration and processing of the  images, is the one area on which I am still unclear.

    I take the following senario (If I am off the mark, please correct me):

    CHDK software in a Canon S100 as an intervalometer
    take pictures every 4 seconds during flight
    at 1 inch resolution
    flying at 130m
    with an overlap of 60%
    cover 640 acres
    in 18-20 minutes

    1. How many images will there be?

    2. What will size the individual images be?

    If I use Agisoft Photoscan Standard to process the images

    3. What will the requirements for a workstation to process these images be?

    4. How long will it take?

    5. What will the size of the processed image be?

    If I then use Agpixel to do the analysis,

    3. What will the requirements for a workstation to process these images be?

    4. How long will it take?

    Thanks again to all for the priceless information on the forum.

  • On the series of wavelength images: if you look closely at the detail, as opposed to the overall brightness, you will notice that where you have the densest chlorophyll (the trees), the brightness is actually best in the 720nm image. Some of the dark contrasting areas in the field at higher nm values, that appear bright in the lower nm range, are due to the strong NIR absorption by moisture at the higher NIR wavelengths. This effect may result in an apparent low plant biomass in those areas that is not related to the actual biomass. When interpreting an NDVI derived from the higher wavelengths for areas like the example where there are strong differences in soil moisture, there is a danger of misinterpretation as far as biomass is concerned.
  • James, thanks for sharing the Event38 link. I was not aware of them.
    The graph they provide can be misleading. The graph shows % transmission, and not the sensor response. The MaxMax conversion has a very similar transmission graph, but the important parameter is the sensor response. I'm not sure why the colors in their example do not appear "correct", but it may just be related to how the image was processed.
    I'd like to see their camera compared to the MaxMax. If they are willing to let me do a head to head comparison, using the same methods and equipment on both, I'd be happy to share the results with the community here.
  • I hadn't realised you had done so much testing, i just wanted to ask some questions to ensure i am buying the right camera before i shell out a tonne of money! Thanks for the help though, i am relatively new to these cameras, i am used to operating UAVs so it may take me a little bit up time to get up to speed. Only a few weeks ago i was looking at tetracam and now i wouldn't touch it from what some people have said. This maxmax conversion is definitely the most promising camera i have seen so far, especially when combined with the Agpixel software. 

    Does anyone know how much Agpixel costs? I couldnt find a price anywhere :/

  • James, as they say, "the proof is in the pudding."  We are getting r-square values using the MaxMax NDVI cameras of 0.91 for corn yields and 0.94 for tallgrass biomass estimates.  The NIR band is located along the red edge, which is where a lot of the action is taking place for many plant components.  We are getting amazing correlations with our MaxMax NDVI cameras. I have a 2,151 band $100,000 spectroradiometer that we used to collect spectral readings on 10,000 soybean plots and when we ran statistical analyses on the data, the NIR bands on the red edge were selected among the 2,151 bands as the best predictors of soybean yield.  I think the NIR band on the MaxMax camera is placed right in the right spot.

  • Anyone had a play with this? Looks like it might be a slightly better solution than the maxmax conversion...


  • Comparing grayscales of a field at different wavelengths you might see what i mean (see below)

    Is the conversion going to be good for NDVI, because there isn't a massive amount of reflection in this scale. Not as much as there is at 800nm anyway.


This reply was deleted.