How to process RGB and multispectral drone imagery in Pix4D (Drones in agriculture series, 2/7) 

Thanks!!

Thanks!

Hi Claire! Thanks for your comment! Realistically, I have used those RGB geotiffs and had very good results. It depends on your objectives. Keep in mind that those RGB geotiffs will often vary (for example) based on white balance settings on the camera. If you were trying to publish in a high-level remote sensing journal and wanted vegetation index values that were comparable between environments/years, it might not be best. In that case, it would be best to use a calibrated multispectral camera and pull the reflectance layers. But for a lot of other applications (including plant/soil classification, and many applications in the private sector without nit-picky peer review), those VARI or ExG values based on the RGB orthomosaics are super useful. We actually did exactly that for one evaluation year and the reviewers didn't have a problem publishing it in Remote Sensing, so again, it often probably wouldn't be problematic. I hope that helps.

Great question. These images were all nadir. I haven't been including as many obliques recently, typically I think the nadir image sets have fewer artifacts in the orthomosaicking and DSM steps. That being said, when I map vertical surfaces or overhangs (like trees/orchards, for example), I absolutely recommend including obliques! I usually fly a normal nadir mission then modify the gimbal pitch to around 20-30 degrees, and fly a second mission over the area with lower overlap

Hi, yes it should all be the same

Great question! There are some example data sets available from Pix4D here, including RGB and multispectral images: support.pix4d.com/hc/en-us/articles/360000235126-Example-projects-real-photogrammetry-data Pix4D also works with a lot of image types, you should even be able to take images of objects with your phone and it should be able to give you decent results. Although full disclosure, all I have processed is drone images haha.

I usually take them before the flights. Usually it is best to fly around solar noon, and the sun angle doesn't change very quickly when the sun is high in the sky. When in doubt, you can always take images before and after!

​@@travisparkerplantscience Thank you. I will remember the solar noon-time for multispec imaging. Your demo has been very helpful. So far, I was successful to replicate your demo with my datasets (RGB). One challenge was to calculate sDSM - as my grids were made on the closed canopy, and I didn't want to include the entire plots. There were many biomass cuts inside the plots, and we wanted to rule out the border areas. So I created another layer of grids covering the soil surface near to each plots, and calculated sDSM. I believe it worked good. I will try extracting the NDVI and NDRE from multispec soon. I assume NDVI Geotiff from Pix4D can be directly used for zonal stat (please let me know if I am wrong)., and I should create a raster for NDRE using the index bands from pix4D. One more question is, by any chance, did you (or do you plan to) make any video on georeferencing using GCPs? Again, thank you for making these demos available! BR, Eusun

@@eusunhan8199 Hi Eusun, It sounds like you did an excellent job troubleshooting the problem and making a second grid to get the data you needed! Nice work! Sometimes the "buffer" tool can be useful for that kind of thing, to get somewhat larger or smaller cells for data extraction. I have been meaning to make a video on that for a while. Yes, that NDVI geotiff is already 100% ready for data extraction. You can calculate NDRE based on the RE and Red bands (from the "Index" folder in the outputs, if I'm not mistaken). This is really easily done using the raster calculator in QGIS. I used GCPs a little bit years ago, but they really aren't very useful for me so I haven't used them since probably 2017-2018. It depends on your particular project, but often they aren't that important. I think there are several good tutorials provided by Pix4D on how to do it in that program, and several more tutorials on how to do it QGIS, although I'm not sure which I used now.

@@travisparkerplantscience Yeah! your suggestion on NDVI and NDRE worked really well, thank you! And I agree with you on GCPs. I can simply adjust grids and other things data by data for now. Also I experienced some distortion after georeferencing - when somebody did it for me. That being said, I am happy with the workflow as I learned from your demo, and will dig out more when necessary. My next target is thermal images - and I assume I will directly use the image with the grids to extract the temperature. Great to get to know you and your work here and other platforms, hope we keep in touch. BR, Eusun

@@eusunhan8199 Hi Eusun, processing thermal data works the same way, the outputs are very similar to those of the multispectral and RGB cameras, just make sure you choose the right processing method in Pix4D at the beginning (for my Zenmuse XT, it is "Thermal Camera"). Keep in mind that the grids can be saved, then duplicated, then shifted slightly between flight dates and different raster layers, if needed. Thanks again for watching! If you publish using these methods, consider dropping us a citation! Parker, T. A., Palkovic, A., & Gepts, P. (2020). Determining the genetic control of common bean early-growth rate using unmanned aerial vehicles. Remote Sensing, 12(11), 1748.

Hi Ranadeep, Pix4D has some excellent datasets here: support.pix4d.com/hc/en-us/articles/360000235126-Example-projects-real-photogrammetry-data

Hi Benaja, thanks for asking! I have used several devices for the processing, and hardware requirements can be found here: support.pix4d.com/hc/en-us/articles/115002439383-Computer-requirements

@@travisparkerplantscience Thank you!

Hi Tejo, I've never run into this problem. My only suggestions would be to 1) buy a radiometric calibration panel, or if absolutely necessary, 2) use images of other peoples calibration panels taken with the same type of camera under as similar of lighting conditions as possible. I hope this helps!

Well, thank for the advise. If we dont have reflectant images are we can skip the radiometry calibration step or choose no camera on pix4d mapper?

@@tejosuryo5400 I'm also wondering the same question as I'm also using DJI P4 multispectral with no calibration panel provided. Glad if anyone can respond

@@jusnainimuslim1895 drone has the sunshine sensor thus radiometric correction, just skip on the step radiometry calibration on the pix4d mapper software. I have had processing some area and then do ground check there is good acceptance on kappa statistic analysis.

What do you mean? Are you thinking of ODM?

Hi Jean Pierre, here is a paper where we did this kind of thing: www.mdpi.com/2072-4292/12/11/1748

Hi Abdul, I would recommend using a mission planning application for your flights. They are free, easy to use, and will ensure that you have consistent image capture. They honestly make the flights much faster and easier. Pix4Dcapture is free and supports the Parrot BlueGrass, I would recommend trying it out!! www.pix4d.com/product/pix4dcapture

5m vertically? Or horizontally? You can improve this using ground control points in Pix4D or subsequently in QGIS, although it isn't always perfect. For a lot of projects you can calculate differences relative to known positions in the field anyway and don't need the correction between sampling time points though

Hi Nercia, thanks, I'm glad you liked the video! In the Pix4D outputs there is a folder called "4_index" and in that there is a folder called "reflectance" with those data, if you told Pix4D to generate them. To make calculations with these outputs, check out this video: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-U2qJn7jHYDw.html

@@travisparkerplantscience Thank you very much for another tutorial, very rich in knowledge. Thanks again. Please help me with this, I am working with an underwater image and I would like to know, how can I do the correction water column using the invariant depth index in Pix4D? Or, where can I find the metadata values?

@@travisparkerplantscience Hi. Thanks again, the video helped me a lot. Now I would like to know how I can do the water column correction for an image taken in seagrass using the drone here on Pix4D. What parameters should I consider? Thank you.

@@nerciajoaquimnhanice7965 Hi, unfortunately I am not familiar with this approach. I would recommend putting the outputs into QGIS, ArcGIS, or similar programs and try generating and then pulling out the data you need with the raster calculator and zonal statistics tool, or similar functions. Sorry I can't be of more help!

Unfortunately it isn't free. I taught a class on using drones so the license I am using was from that. At some point here I am interested in making a video on Open Drone Map, which is a free program to do photogrammetry. A lot of the outputs are very similar, to the point that it seems like the people making Open Drone Map are probably experienced Pix4D users and are reverse-engineering it. More info here: docs.opendronemap.org/installation.html

You can add GCPs if needed, but the cameras and aircraft already have GPS units, and that almost always works well enough for my applications. In the early days I used GCPs more, but I haven't for a long time because they aren't that necessary for my applications. Pix4D has some videos on how to do that though, if you need to

@@travisparkerplantscience I'm getting a large percentage of photos left out of calibration in my end product. For crop fields larger than a few acres, do you think GCPs would help calibrate the more repetitive imagery?

@@MikeDeGreef Hi Mike, to be clear, by "left out of calibration" are you thinking of the "Quality Check" section of the Quality Report? One important consideration is to have fairly high overlap between images, a minimum 70-75% frontlap and sidelap. In general it is also best for images to be taken from a single altitude, when possible. If you are mapping a highly uniform area (especially water with few features, grasslands, etc) it can be hard for the software to find quality tie points. In my experience, GCPs become slightly less necessary overall as the mapped area area gets bigger, since a 1-5m imprecision in a camera GPS (for example) has a greater relative effect when mapping an area of 100 x 100 meters relative to an area of 1000 x 1000 meters, for example. I hope this helps.

@@travisparkerplantscience Thanks for the quick reply, very helpful. To be more clear, I am using ArcGIS SiteScan, though I think it uses Pix4D on the backend for photogrammetry

@@MikeDeGreef I hadn't heard of SiteScan, but it looks like you are right that it does use Pix4D for the photogrammetry!

Hi Ramandeep, this is exactly the video you are looking for, from 17:49 and onwards! At 20:11 you will notice I click on the "Index calculator" tab. If you want the index values, check the boxes for the five bands in the "Indices" field (see around 21:15) and make sure that the "Merge tiles" box is checked immediately above that. The results will be in the 4_indices folder after processing.

@@travisparkerplantscience ok I will try following your vedio stepwise. As I am totally new or very beginner to this, so I was searching something that can teach me every step. But yeah I will do as guided by you in your last text. Thank you so much

@@ramandeepkumar7401 this video is for sure designed to show you every step! Hopefully everything makes sense

@@travisparkerplantscienceso last question now....I just want to make sure that we are on same page.....I am graduate student in agriculture ....I have flew drone and collected data using mica sense rededge. Now I want to extract that data into 5 vegetation Indices NDVI, NDRE, SCCCI and FENDVI. So this vedio will give me all these? I. Sorry if my question will not make any sense since I am actually struggling to find any find that could help me.

@@ramandeepkumar7401 sorry I misunderstood your original question! I thought you were looking for the five bands from the RedEdge. To calculate those vegetation indices, start with the methods I mentioned previously to get the index values for each wavelength. You can then use the "raster calculator" in QGIS (for free) to then do calculations with each of them. See more here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-U2qJn7jHYDw.html Hope that helps!

Hi Southone, there is some awesome example data here: support.pix4d.com/hc/en-us/articles/360000235126-Example-projects-real-photogrammetry-data I hope this helps!

@@travisparkerplantscience Thank you for sharing ! If possible I would like to get a sample in a VDO ,

Hi, I would recommend starting out with this playlist: ru-vid.com/group/PLifbxiLZAb-QHX2QjBqCNKysSHM3gPuTX I mostly work with QGIS. The shapefiles and raster graphics discussed in these videos can be designed in QGIS for free. I very rarely need to do anything with them that can't be done in QGIS, but if so, they can be read into R and data can be extracted using the raster and rgdal packages. Also, you're English is great, but I understand the predicament in general haha, you'll see I have videos in Spanish and other languages haha

@@travisparkerplantscience i didn't expect such a fast reply. Thank you very much!

Hi Francis, thanks for asking! I have a Facebook for personal stuff, but I have a professional Twitter here: twitter.com/travisparker91 and researchgate here: www.researchgate.net/profile/Travis-Parker-2