Examples

Example Data

Example data can be found in this link: Example Data

Data includes:

• Drone Images (for Aerial-Based Trait Extraction)
• Rover Images (for Ground-Based Trait Extraction)

Aerial-Based Trait Extraction

Start Docker Desktop and open the GEMINI app. If at any point more specific instructions are needed, navigate to the documentation page for the tab in use.

Upload

• Navigate to the Upload tab. If not already chosen, select Image Data in the Data Type field.
• Populate the following fields with information on the data to be uploaded.
• After all fields have been populated, drag and drop files into the upload box or click in the box to select files via the file explorer.
• Once you have selected all files from the dataset, click Upload to upload the images to the app.
• After the uploading process is finished, click Done.
• If you wish to upload GCP Locations, change the Data Type field to GCP Locations and follow the same upload process as before.

Process

• Navigate to the Process tab. In the data selection menu, select the Year, Experiment, Location, and Population of the previously uploaded data to be processed.
• Click Begin Data Preparation. In the Orthomosaic Generation window, expand the dropdown of the chosen Platform and Sensor. Click Start on the date to perform orthomosaic generation with.
• If GCP Locations were uploaded, use the slider bar and the Previous and Next buttons to step through images and mark the visible GCPs.
• Click Generate Orthophoto. Select the desired quality in the Settings dropdown.

• Click Process Images to begin generation.

• Once generation has completed, click Done on the progress bar. Navigate to the Plot Boundary Preparation window.

• Follow the instructions on importing of field design data. Once data is uploaded, proceed to the Population Boundary step.
• Select the orthomosaic to create the boundary for via the Select an orthomosaic dropdown.
• Use the Draw tool to outline the boundary of the portion of the orthomosaic you wish to analyze.
• Use the Edit, Select, and Translate tools to modify a created boundary.
• Click Save and Proceed when finished to proceed to the Plot Boundary step.
• Select the correct orthomosaic as before. Use the rectangle generation tool in the bottom left to create the number and size of rectangles needed for the scale of analysis needed.
• Use the available tools to edit the placement of the rectangles.

• Click Save when finished.

• After creating the plot boundary, navigate to the Aerial Processing window.

• Expand the dropdown of the chosen Platform and Sensor once more. Click Start to process the traits of the desired date.

Stats

• Navigate to the Stats tab. Expand the dropdown of the chosen Platform and Sensor to see data for the available dates.
• Click Load to view the processed data of choice. Click Download CSV if desired.

Map

• Navigate to the Map tab. Open the data selection menu to select the data to analyze.
• Use the dropdown menu to select the Trait Metric to view.
• The traits can be seen overlaid on the orthomosaic based on the plot boundaries created earlier.

Ground-Based Trait Extraction

Please complete Aerial-Based Trait Extraction before proceeding with Ground-Based Trait Extraction. You will need to utilize the plot the plot boundaries created in the Aerial-Based Trait Extraction process.

Upload

Images

• Navigate to the Upload tab. If not already chosen, select Image Data in the Data Type field.
• Populate the following fields with information on the data to be uploaded.
• After all fields have been populated, drag and drop files into the upload box or click in the box to select files via the file explorer.
• Once you have selected all files from the dataset, click Upload to upload the images to the app.
• After the uploading process is finished, click Done.

Metadata

• Change the Data Type field to Platform Logs.

• Populate the following fields with information on the data to be uploaded and then upload the files.

  • The metadata of these images include camera information, GPS locations, and timestamps.
  • Refer to the the msgs_synced.csv file to format your metadata file for personal use.

Process

Locate Plants

In this section, you will use an early date to locate plants in the field.

• Select the Ground Processing tab. In the data selection menu, select the Year, Experiment, Location, and Population of the previously uploaded data to be processed.
• Navigate to the Locate Plants tab. In this tab, you will annotate individual plants in an image, train a machine learning model to detect individual plants, and then find every plant in the field.
• Select the Platform and Sensor you would like to do this process for.

• Click on the Labels button to label your data. Click on Annotate and wait for the software (CVAT) to open. This could take a while. Note: Please allow pop-ups for this process.

• A new tab should open with the labelling software. Create an account if you do not have one already.

• Perform your annotations and then download the annotations. Use YOLO format during your export!

• Upload the .txt files into the app.

• After annotating your images, you can now train your model. Click on the button under Model for the specific date. Next, press Train Model.

• You can expand the progress bar to track the model performance.

• After training, you should be able to see the generated model.

• Next, click on the button under the Locations column.

• Select the model you would like to use for this process. It is preferred to select the model with the highest performance score.

• Then, you can press the Locate button.

Trait Extraction

In this section, you will extract traits from the located plants. You can select any date for this process.

• Navigate to the Label Traits tab. In the Select Trait dropdown, select the trait you would like to extract.
• Select the Platform and Sensor you would like to do this process for. Similarily to the Locate Plants, annotate the necessary images and upload them.

• Next, go to the Train Traits tab. Select the Trait you would like to train the model for.

• Click on the New Model button.

• Select the Platform, Sensor and Date you would like to train the model for.
• Then, press the Train Model button.

• Again, you can view the resulting model after the process is done.

• Finally, go to the Extract Traits tab. Select the Trait you would like to extract.

• Select the Platform, Sensor, Date, Locations ID, and Model ID you would like to use for this process. It is preferred to select the model with the highest performance score.

• Then, you can press the Extract button.