Leaf Disease Coverage



I’m trying to find an easy way to asses leaf disease coverage. The disease is powdery mildew and the images of the leaves were taken on a light blue background. The tutorials that I have found suggest using the color threshold tool to select the total leaf area, then select measurement. The problem is that no matter how I adjust the thresholds, the leaf is never fully selected. This seems to be because powdery mildew is very light in color. Does anyone have a better method for accomplishing this task? I just need to know the percentage of the leaf covered by disease. I have a sample photo if anyone would like to try or can give pointers.

Autonomize count and identification of particles in RGB image


Go ahead and post your example image. That would be helpful for us to see - to help you better.

To start - take a look at the Trainable Weka Segmentation (TWS) plugin - a great tool for segmentation that comes directly with Fiji. NOTE: Fiji is Just ImageJ - it is simply a distribution of ImageJ that comes with a bunch of plugins bundled - ready for you to use out-of-the-box. If you are just getting started, we recommend downloading/using Fiji.

eta :slight_smile:


Here is one of the images I’m using. As you can see, the diseased area is very lightly colored and not clearly delineated.



Using TWS (as I linked above) and the default settings - I got some pretty good results using a cropped section of your image:

I would say it’s worth learning more about this plugin and improving the segmentation yourself - looks promising. Hope this gets you a good start.

eta :slight_smile:


This is actually a great start. Thanks for the help! I’ve been playing with it for a couple hours and have gotten great results. Is there a way to save the training so that I don’t have to retrain on each image? I tried saving the classifier but that didn’t seem to do what I wanted. I read the FAQ for the plug-in, but I’m still struggling a bit.



That’s great! I’m glad it’s helping. You need to save the classifier (that saves your training) - then you can apply that same classifier to multiple images acquired in the same manner.

What is it that you want??

You can generate a probability map that you can use to auto-threshold and then analyze particles… this will set up your segmentation of your leaf and then mildew… to then measure their areas, etc.

There is a helpful workshop on Segmentation in Fiji (and corresponding slides) - that goes over the basics of this plugin, but also other ways to develop segmentation workflows in ImageJ/Fiji that will help you in your analysis.

I hope this helps! If you have more questions… feel free to ask.

eta :slight_smile:


Here is what I came up with. Tell me if you see any potential issues with this approach. I was able to get a classified image. I then saved the classified image as a PNG and converted it to RGB, I then used the color counter plugin to get the number of pixels in each color. I added the red and green together to get total leaf size and used the number of red pixels to calculate the percentage of coverage. It still needs some tweaking, but this allowed me to avoid the auto-threshold and analyzing particles portion. Any thoughts on doing things this way?



That sounds fine to me. As long as you treat each image the same - applying the same classifier and subsequent analysis workflow - you should be good.

eta :slight_smile: