We would like to show you our approach for cereal ear counting in field conditions. The objective of this study is to demonstrate a quick and efficient way to count wheat ears in field conditions. Durum wheat and barley are by extension the most cultivated cereals in the south and east areas of the Mediterranean basin.
In those areas, as a consequence of the climate change, the environmental conditions will be changed. We have to work in order to increase the production. In this sense, remote and proximal sensing images have become an important tool in field high-throughput phenotyping using different kids of sensors.
One of the essential points for improving cereal productivity is a more efficient assessment of yield. Determined by the following three yield components:Ear density, the number of grains per ear, and the thousand kernel weight. We will try to develop an automatic counting of ears per unit crop area in other words ear density.
The protocol that we've developed is using a 20 megapixel camera taking pictures downward looking or perfectly xenapho or natur at a distance of approximately 80 cm from the top of the crop canopy. For validation purposes, we do both in the field wheat ear counts, or barley ear counts as well as manual counts of the image in order to validate the technique and fine tune the algorithms. It's also important when taking the field photos, that the pictures be captured within two hours of solar noon.
This is important because it avoids shadowing effects, that complicate the image analysis in the 2nd phase of the protocol. In this part, we would like to show you our approach for cereal ear counting in field conditions. This work was carried out in collaboration with ITACyL, INIA, and Syngenta.
Let's begin. The first step of our protocol is select the appropriate crop growth stage. In our case, we have use the stages between grain filling and near chromatology that correspond in the South East case a number between 60 and 87.
In the Figure A:Wheat and Figure B:Barley we chose an example of satellite images of our data set. The image capture has 3 parameters:sensor width, photo lens and distance between the camera and the canopy. With this information we can calculate the square meters of the image.
Algorithm implementation and adjustments, these are the pa-blan-y steps. As an input we have an RGB image. Laplacian frequency filter we use it to remove part of the soil, leave and unwanted brightness.
The medium filter reviewing the noise, and finally find maxima determine local picks. The output image show the ears detected. Algorithm implementation and adjustments.
If the images were taken with different camera specification or distance between the canopy and the camera we can adjust some algorithm parameters. The Laplacian filter is still the same. In the medium filter and find maxima we can change the diameter and the noise parameter.
Algorithm validation. For the validation step we have marked each ear in the original image and then the number of marks in the image were counted using a simple algorithm. The results were used to calculate the success rate.
We have also included a circle to have a physical reference in the image. Algorithm implementation using images. This is the cereal scanner plug-in.
You will find in the central tab data counting. In options you can select the input images and also the place where you are going to save the results. I would like to show you how the macro works using one image.
We're going to select this image and now we're going to run the first step. This is the frequency formation of the image and here we have the result. The second step is the medium filter.
We're going to run this step, and this is the result. Finally we're going to run the last part of the cut and these are the result, these final numbers is the number of ears detected. The other way to the same is click here, we process, find maxima, and click here.
This is another way to use the ear counting algorithm. You can visit our website integrativecropecophysiologygroup. com and here in software development, you will find the cereal scanner.
To access permission please write directly to this email. Please follow the steps to install the plug in. With the plug in installed please go directly to plug ins, cereal scanner, open cereal scanner.
Now we're going to use two images. Go directly to cereal scanner, ear counting, and in options, select your files. Here you can use the distance between the canopy and the camera.
In our case we use it 80 cm. Here you can select the focal length. Finally in the results file you will find the ear counting results and just process.
These are our results. Here we have the name of the image and in front we have the number of ears detected. These are the results for Wheat and Barley.
For each graphic the X-axis represents the manual counting. Click by click, and the Y-axis represents the algorithm counting. Both axis in a square meter scale.
In the fifth graphic for wheat we have obtained a determination coefficient equal to 0.62 and in the second graphic for barley we have obtained a determination coefficient equal to 0.75. In the final image for wheat we have also obtained a determination coefficient equal to 0.75. In counting is one of the most laborious to work and time consuming during variety evaluation cycle and yield prediction.
For this reason a fast and friendly technique is needed to improve and expand it's use in precise agricultural and crop breeding and for yield forecasting. This use of our cereal scanner is the fruit of collaboration of Public Provides Corporation with the objective to make ear counting with improved efficiency and time and resources. On the other hand to capture data of high quality to end up with more precise product database.
While yet to conclude I would like to stress similar aspects of the mis-a-lory you would have been presented here, one is the local, so this is really a form of technology that doesn't need any kind of complement so people may go to the field with their camera, with a mobile phone, taking images under sunlight and that's all. Second, a very important point, is that this is only an assundun methodology that is really a contrast with the cognancy to an issue of the people counting the number of ears in a very different ways. So the local that has an ex-tres-cent before is important because you may do in few minutes what a couple or three people will need perhaps in several hours.
So the counting, the budget is a really important point. Another aspect is that the methodology is minimal for decreasing resolution. That means that you may mount the RGB camera in a megaplant in a kind of automatic plant-form and then you may do an automatic counting on this facts.
And finally I am saying this is not the end of the way so we have really a percentage methodology that may improve in the future through the combination with all the approaches like for example the use of different RGB color and spaces for example in combination with multi-spectral imagining or even the use of thermal or thermal images. And that's all, thank you.
