The…interesting results from my first mapping flight indicate that I need to have even more overlap between images. For my second flight, I dialed up the percentages on the Mission Planner to 95% overlap and 80% sidelap. I also set up the Raspberry Pis to run the image capture scripts as a service (systemd) on boot.

Here is a link to the github repository with the updated python scripts I run on the pis: https://github.com/Bobcati/BLD_NDVI_Scripts

More images and overlap

After the flight, I put the photos into photoshop image merge again to see if I had enough overlap. I think the results speak for themselves:

This image has some blurriness on the top right
This photomerge came out very well

Here are the links to the full resolution images:

IR Photo: https://drive.google.com/file/d/1OZyCyW7UsAowfsIsTLQ5eUnsoYk_OpFJ/view?usp=sharing

RGB Photo: https://drive.google.com/file/d/1K13roW3xFs-y1rMmT7aaTy99zUDu_ExO/view?usp=sharing

Issues

Unfortunately, the Pi Zero W took one fewer image than the Pi 4B did (Zero took 77, 4B took 78). I dialed down the resolution of the pi zero Additionally, in order to put the images into GIS software, I have to geotag them with the GPS coordinates stored in the flash memory on the drone. Hopefully the Pi 4B took as many images as the drone triggered, otherwise I will not be able to geotag.

Next Steps

  1. I will attempt to run the NDVI script on the infrared merged image. Hopefully I don’t run out of RAM.
  2. Geotag infrared images.
  3. (longer term) Getting telemetry working over 4G using Tailscale instead of my prior plan of using port forwarding and reverse ssh.

Actually, why wait. I’m writing this post and I can run an NDVI analysis right now. The code I use is an altered version of the formula:

Well, I’ve run the analysis and it all worked out. For some reason, when I run an NDVI analysis using the blue filter, the results often make the shadows seem like they are reflecting the most IR light. The red filter method seems to produce better results. I need to do more testing with a camera that has a proper red filter (stay tuned…).

Method for blue filter NIR camera
Method for red filter NIR camera

Here are the results:

NDVI with the red channel as the NIR channel
NDVI with the blue channel as the NIR channel

Red as NIR channel full res: https://drive.google.com/file/d/1W7WYLnQ18g7xPWomkYj4CvAvjNDVCz5C/view?usp=sharing

Blue as NIR channel full res: https://drive.google.com/file/d/1IsNUrNv0EhQxVk-itpbDc_DQoc5xgdeC/view?usp=sharing

Although some of the beech trees are infected with BLD, the disease has not progressed enough to create clear gaps in the tree canopy. Additionally, many of the beech trees have put out another batch of leaves, making the canopy appear healthy again.