I have been quite successful to perform general aerial mapping, producing othomosaic and digital surface model for our local plantation company. Our Skywalker base UAV power by APM is a very reliable tools for acquire aerial images. Currently for a typical mission we fly at 300 meter altitude and resulting ground sample distance is about 9 to 12 cm per pixel using Canon S100 GPS. 


I am very excited to learn that Agpixel is providing a way to transform Near Infra Red images to NDVI, which may become a very powerful tools for the agronomist or the plantation management personnel to zoom in the problematic area as soon as possible before more severe damage or losses occur. This type of information will be crucial for the agronomist in their recommendation of fertilizer input, which account for 60% of the overall cost of production for oil palm plantation.      

Normal RGB Othomosaic image


To acquire NIR image, I am using Maxmax modify NDVI Canon S100 GPS camera. The images quality is very good. Agpixel has maxmax modify camera setting ready to use so the user no need to tweak parameter like imageJ. A few click later, some beautiful false color image can be generated for viewing or export to various format.


Here come the tricky part, while the false color image is very attractive but how to adjust the parameter so the information presented in the false color image will be accurate and useful to the user has yet to finalize. More research will be needed. It will not serve the purpose to supply an eye appealing NDVI false color images and only to found out that the data is not accurate or misleading.


According to Mark Lanning, the CEO of Agpixel, the program is under massive update and I hope the final product can deal with cloud shadow issue, support large ( Gigabyte ) file and output geo-reference images. For the mean time, this program may be targeted for researcher and the professional from remote sensing field. I can not find a manual or tutorial on how to use it. In my opinion, a comprehensive and detail manual plus video tutorial is crucial if this product is target for general mass market. Most of the farmer or even agronomist may not well train in remote sensing subject.   


With the hard work of Mark and his team, I hope Agpixel will become a leading software for precision agriculture and ultimately help the world to produce more food in the most efficient way.

Near Infra Red Othomosaic image

Agpixel Green NDVI

False Color Composite image

Views: 5679

Comment by Deon van der Merwe on October 16, 2013 at 5:14am
Keeyen, thanks for sharing those images. They clearly demonstrate the scale of operations we can achieve with low cost, efficient, small unmanned systems that are within reach of many individuals within this community. You identified the key question, which is how to generate useful information from the data. Just generating images and stitching them together has value for mapping, but you need reliable, quantitative data in precision agriculture. Much of our efforts have gone into developing methods that turn images into useful data, and the good news is that it can be done using fairly straight forward methods, including image calibration, an appropriate level of image redundancy that allows you to overcome the artifacts generated by wide angle lenses, and using software such as Agisoft and AgPixel. The benefits of precision agriculture applications of unmanned systems, at least those aspects that involve vegetation status measurements, do not have to remain in the hands of the few.
Comment by ECODRONES on October 16, 2013 at 6:08am

Amazing job, Keeyen. Well done.

Congrats!

Comment by Dries Raymaekers on October 16, 2013 at 1:12pm

Hi Keeyen, thank you for your post. I always like to read about your mapping experiences, good job! I also want to try out a NIR converted camera to see what are the benefits/limitations, preferably based on the Canon S100, which I am using at the moment. However, asking around to several companies who do the conversion resulted in a bad recommondation for this camera. Apperantly the NIR converted camera is terrible related to hotspots (different performance of the lens related to viewing angle). Analysis of many cameras in this link. What is your experience with this hotspot issue with the S100?

Comment by Cala on October 16, 2013 at 5:44pm

Great Job, how did you mount your camera in the Sw? can you show a photo? thank's

Comment by LanMark on October 16, 2013 at 9:35pm

I am here :).. just got back from the Kansas UAV conference which was a very nice event and every single person there seemed to bring something unique to the event..  I really didn't see much overlap.

Just fo clarify, it technically is Goldfinch Technologies where AgPixel is a product of.

Keeyen, I am always amazed at the sorts of data sets you are generating utilizing the low cost, efficient sUAS like Deon mentioned. Very nice.   


Things are changing so rapidly that I just haven't gotten to a guide/manual yet.. and the current publicly released version of AgPixel is in a lot of ways intended to generate feedback from a large group of end users.. from the nursery guys, grape vines, high value crops like apple trees, turf, corn, soybeans, rice and all sorts of different users.


The updates coming down the pipe will start to really address the producer focused features like ShapeFiles, geo referencing, custom formulas, custom color pallets, different classification techniques and improved atmospheric adjustment functionality..  lots of stuff.. and also now that crops are being harvested there is a lot of data to get to and figure out.  I am also working on some web focused products and services.

One thing that I think for me was made pretty clear from the Kansas UAV conference is that we all have a role to play..  which is why people participate on here and all of that.   UAVs is a big thing and just getting bigger and we all really benefit from what unique things each person brings to the table.   I sure wouldn't have been able to create my own quadcopter without reading on here what others have said worked and didn't work as well.    Same is really true with AgPixel.. your feedback Keeyen (and crazy large files I am testing with) along with Deon and Kevin Price and their amazing KSU researchers.. and the other contacts in 36 other countries is much needed to build something of usefulness..  you really need that feedback loop direct to the builders.

There is a lot more to precision agriculture remote sensing than slapping a NDVI image over top of it.   We are noticing interesting problems caused by the high resolution imagery which wouldn't exist if pixels were bigger and things averaged out..   but all new sets of issues to learn from and resolve.   All things that are being researched and figured out by very talented people that all bring some unique thing to the table.

Comment by LanMark on October 16, 2013 at 9:36pm

Keeyen,  how many acres does your image represent?

Comment by Nuno on October 17, 2013 at 1:18am

This is very interesting. I'm studding also about this.

LanMark you put  a interesting problem. What's the problem with NDVI  high resolution images?

Comment by Cala on October 17, 2013 at 6:24am

Lan I agree with you that high resolution images is a problem to process, now I'm prefering low resolution cameras for processing the images.

Comment by Nuno on October 17, 2013 at 7:21am

Can you explain why High resolution images are a problem.

Comment by LanMark on October 17, 2013 at 7:40am

@Nuno..  with high resolution imagery you can get higher NVDI values in the shadows between rows of corn for example.. and clearly shadows don't yield more corn than the actual plants... but in the NIR-VIS/NIR+VIS its entirely possible and does happen at the more extremes of the image sensor, so in this case low brightness values in the bands.   Obviously the size of the files generated, the low bandwidth in areas you are capturing imagery is just to name a few others.   Keeyen I believe has little to no internet connection in the areas that they are flying.   So a web based AgPixel didn't make as much sense as a installed one.. even though web based would have been cross platform...  moving Gigs of data to a central processing center is problematic with the larger datasets being captured.   Some of the military sensors can capture pedabytes of information while in flight.

There is a happy middle ground to find based on your application.   If you are doing turf management you need a lot higher resolution (pixel to distance).. where row crops like corn doesn't need that.. and the farm equipment can't consume it even if you did generate it that way.. granted as a newer post on here demonstrates there is some nice things you can do with the higher resolution such as better object classification techniques.

If your limiting factor is the 400ft (120m) ceiling the FAA has right now which hopefully will switch to 700ft at some point..  it may make more sense to lower the resolution of your camera and thus utilize more of the physical photon detection surface per pixel.. and thus higher quality pixels.. then use the 16 bit color depth instead of 8bit.  But that is one of those items I am exploring.   Understanding your inputs (and outputs) is critical in this..  all hardware has limits and knowing them and working around them to utilize their strengths is very much what is needed, these cameras were not made to be doing this sort of work. This is something KSU has done a great job on figuring out and a lot more being done.

Camera manufactures keep managing to pack in more pixels into the same sensor size.. that doesn't mean the sensors are getting that much better over time.   You have amazing algorithms being run on the data being captured which some of the new cameras are dual core even.   So I guess my point is to not be fooled in thinking that a 36MP camera is twice as good as 18.. much like computer processors.   Its about what is being done and the quality it is being done given your limitations.

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service