Plant Phenotyping Blog
Recent Ideas from Industries and Science

Digital Disease Quantification in plants, a tailored method

Posted on 21 October 2019 in by Alexander den Ouden

A few months ago I received a request of one of our clients to see if we could help and improve their manual disease quantification. They reached their limit of the amount of plants they could manually score and are looking for solutions to automate this process. They already invested resources in detecting the disease symptoms on their plants but they still had problems detecting the disease in an early stage. At this point they are still scoring the plants manually.

In the past years we have learned a lot about quantifying diseases in plants using our PlantEye sensor, but we still strive to tailor our solution to the specific needs of each client (e.g. different methodologies, species and diseases). So we decided to jump on this, do the tests and develop a disease quantification process with our client and here is how we did it.

Experiment

For the question at hand, we had Cabbage and Cauliflower at our disposal supplied by the client and infected with a disease (we keep this confidential). The plants were infected to different degrees and the infection grade was scored on paper by an expert using a score of 1-10. When the manual scoring was finished, we used our PlantEye to also “score” the plants using the 3D + multiSpectral capabilities of PlantEye.

To have the PlantEye “score” just as the expert we first need to explore the data provided by the PlantEye. Prior knowledge tells us that the spectral parameters we provide are interesting candidates. We choose to evaluate the following three spectral parameters:

  • NDVI (Normalized Difference Vegetation Index, indicator of plant damage)
  • Hue (measure of color)
  • greenness (ratio of green color versus red and blue color)

After evaluation we decided to use the NDVI parameter because it showed promising results. Let me show you why. The following 3D models, with NDVI values overlaid, shows the disease spreading on the plants.

Figure 1 - 3D Scan with low NDVI coloring

Figure 13D Scan with low NDVI coloring

 

The top row of plants is healthy in comparison to the lower row of plants which we can see by the amount of red leaf tissue. Red means a NDVI value of 0,15 or lower which indicates plant damage. Nonetheless, if a human would score the above image the data still would not be objective and tangible. Therefore we need to digitally quantify the amount of leaf tissue with low NDVI values.

Quantify unhealthy plant tissue

If you display the plants in the image above as histograms the healthy plants in the top row have the following distribution:

Figure 2 - Healthy plants NDVI signature

Figure 2Healthy plants NDVI signature

For the lower row of unhealthy plants, the histogram has a different distribution:

Figure 3 - Unhealthy plants NDVI signature

Figure 3Unhealthy plants NDVI signature

 

Infected plants show distinct different patterns

We can already see a more apparent difference. And with the numbers provided by the histogram, we can also create statistics or detectors of disease infection grade. To do that we introduced a binning system which essentially tells how much percent of the leaf tissue has a specific range of NDVI values. If you visualize it on a histogram, it will look like the following:

digital disease quantification - Binning on NDVI histogram signature

Figure 4 – digital disease quantification – Binning on NDVI histogram signature

 

Our software HortControl will then tell you that Bin 1 with NDVI ranges from 0 to 0,1 holds 5,41% of the total. If you multiply the value of 5,41% with the total 3D leaf area, you get the absolute number of leaf tissue affected by the disease.

Correlation between Manual vs Digital scoring  

Results of the correlation between manual expert scoring and automated PlantEye scoring reached up to 0,9 with a p-value of < 0,01.

Scatterplot of NDVI with regression line

Figure 5 – Scatterplot of NDVI with regression line

Results

The results show that we are able to detect this specific disease in an early stage and correlate to the manual measurements. Our client is convinced that the results are reliable and reproducible and the method solves the problems they have.

It is not easy to automate the disease scoring of breeders. There are so many factors to consider like the plant species, the different traits methods and setup. This means we need close collaboration with our clients and adapt our products to the very specific needs and questions. So what is your digital phenotyping question?

Kind regards,

Alexander den Ouden
Technical sales manager

Alexander den Ouden, Digital phenotyping expert
High-tech phenotyping equipment is increasingly adopted to improve the phenotyping process. That means you have access to multiple objective parameters in a very high frequency, 24/7.
These parameters give you the characteristics on how the different genotypes of your crops behave, develop and adjust to the environment. The wealth of data comes with challenges to interpret them and unexpected insights. Because of that, new ways of evaluating crop data have to be adopted and we assist in that process.

Leave a Reply

Your email address will not be published. Required fields are marked *