Workflow: Automatic Detection

Processing with the GRYFN Plot Extraction Tool involves three to five steps, as illustrated in Figure 1.

Figure 1. Automatic detection workflow.

New/Open Project

The user creates a project by specifying a working directory, datasets, and field extent. Additionally, the user can input prior knowledge regarding trial dimensional information.

Initial Plot Extraction

Plot extraction with the GRYFN Plot Extraction Tool involves three core stages, as shown in Figure 2.

1. Angle estimation: Estimating the angle between the North (for vertical rows) or East (for horizontal rows) and the planting direction.

2. Plant row and alley detection: Identifying the plant row centers (or gaps) and alleys, a task critical for plot extraction.

3. Plot generation: Creating segment boundaries based on the detections.

Figure 2. Steps for initial plot extraction.

Field-Based Refinement (Optional)

Systematic errors like planter angle drift, multiple planting dates, etc. can lead to variable differences across the field. To address these variables, the tool divides the field into smaller tiles and performs plot extraction for each tile for precise angle estimation. While this phase is optional, it is recommended when dealing with a large field, especially those with dimensions surpassing 100 meters along the rows. More tiles lead to more user interaction, so it is best to limit the tiles based on subjective recognition of systematic errors such as converging rows, etc.

Plot-Based Refinement (Optional)

For highly variable errors such as hand planted plots, the GRYFN Plot Extraction Tool offers two features for refining plots based on local field conditions, providing a more accurate representation of real-world dynamics in the field.

1. Refine by Plots: This feature employs an algorithm similar to plot extraction, except it operates on individual plots rather than the entire field. It can refine the orientation and length of each plot, meaning that segments within a plot are locked together in adjustment.

2. Refine by Segments: This feature identifies an oriented bounding box that encloses the above-ground points (pertaining to point cloud data) or vegetation pixels (pertaining to image data) for each segment and allows each individual segment to adjust freely.

Labeling

The last phase assigns labels, including range, row, and segment IDs, to the plots according to the trial information provided by the user. Currently, the user can establish an origin, and a start number for the Row-Range. The plots are then labeled sequentially.