PlantAI Tutorial for Basic Plant Counting

Setting up the Data


Download the sample imagery for this tutorial @ [insert link here]


Navigate to the Solvi home page (https://solvi.ag/) and create a new farm providing a suitable name


Click on ‘Upload Images’, choose ‘Stitched Map’ and navigate to the sample image location on your PC. Populate the ‘Field’ & ‘Crop’ values, and choose ‘Upload 1 Image’.



This process will take some time (approximately 15 minutes in this example). After completion, you should receive a confirmation email.


Open the project within the ‘Imagery’ tab then define the Field Boundary. (This will make further processing more efficient)



Adjust the corner points to match the field perimeter.



NOTE If you are working over a particularly large area, then the processing stages are likely to take a long time. This being the case, we can recommend you reduce the field perimeter to a smaller area at first to test your AI model. Once successful, you can reset the boundaries.


We are now ready to begin using the Plant AI tool!


Running the Simple Plant Count PlantAI Tool


Navigate to the ‘Analysis’ tab, select ‘PlantAI’ and choose to create a new detection. Select ‘Basic Counts’. The ‘Provide Examples’ setup will appear


Zoom into the orthomosaic image and using the box tool, draw a perimeter around 3 plant specimens (cabbages). A perimeter box will then appear automatically afterward.


Note that it is recommended to choose an area of the image where there is both plants and non-plants present to help train the AI more effectively.



At this point, you can either keep using the box tool, drawing a box around each plant shape OR you can use the select tool (arrow icon) to select the blue perimeter box (it’s colour will change to red) then select the magic wand tool , which will try and find more plant specimens automatically.


Keep going until all the plants within the bounding box have been selected. Then when you are ready, select ‘Continue’. Then select ‘Run Test Detection’. The processing window will appear and should take around 15 minutes to complete (for this specific example).


The Review Test Results screen will be displayed where a manual inspection of the detection accuracy is required.



If there are any plants that have not been properly detected, zoom into this area, select ‘Provide more examples’ then select ‘add examples’.


To speed up this process, we can use results from the previous iteration. Once the bounding box has been created, use the select tool (arrow icon) to select the bounding box (its colour will change from blue to red), then click the “Get from Latest Results” button.

This will try to automatically find plants. Afterward, you can manually add additional plants using the box tool. When finished, select ‘continue’, then ‘Run Test Detection’ to test this new iteration of the training data.




NOTE The red boxes from the previous detection are only for reference, all new examples require that all plants within the bounding box to be selected again from new.


Continue this process improving the detection until the test area detects all plants correctly. Once you are satisfied with the results, then click on ‘Continue with the whole field’.


Review field boundaries and price summary, adjusting the boundaries if necessary to avoid processing of irrelevant areas. When ready, click ‘Run Detection’. The processing will now begin for the whole area and will take some time to complete.



The final results of the PlantAI detection are now displayed. If there are still plants not detected or false positives, select ‘Add more examples’ and repeat. Otherwise, select ‘Proceed to Analysis’



Analysis

Now that the plants have been properly detected, we can begin analysis. Select ‘Plant Count’ to display the number of plants detected.



Congratulations! You have successfully reached the end of the basic plant counting workflow for PlantAI with Solvi. You can continue to explore the functionality of this tool by looking at the ‘Plant Health’ option or creating and analysing areas of interest using the ‘Zonal Statistics’ option.

Still need help? Contact Us Contact Us