2_EasyDCP_Analysis.md

EasyDCP_Analysis

Materials:

Point cloud files (.ply) from EasyDCP_Creation step

GIMP, Photoshop or equivalent software

Known height of container (mm) or ground offset height.

(a) Setup - Creating training data

Materials:

• GIMP or equivalent software.
• Source image files from image acquisition

Use GIMP to create training data:

See example fore.png and back.png in \example\training_data\.

1. Use Free Select Tool to select plant parts only - Use a representative sample. Maybe 3 different plants is enough to represent them all e.g. one light green, one dark green, different light conditions, etc.
   • Deselect "Antialiasing" in Free Select settings pane
   • Select "'1. Pixel (3 x 3)" in Brushes tab
2. Copy the selection and create a new file, PNG with transparency
   • Under Advanced Options in "Create a New Image" dialog, select Fill with: Transparency
3. Save as "fore.png"
4. Repeat steps 1-3 with background samples and save as "back.png"

(b) Running EasyDCP_Analysis

Materials:

• Folder containing .ply point cloud files output by previous step, EasyDCP_Creation
  • See example ply folder in step 5 of EasyDCP_Creation
• training data created in previous step: "(a) Setup"
• Python script to control EasyDCP_Analysis package, example/analysis.py

Control EasyDCP_Analysis via API using python script in your python 3 environment as described in Installation.

We have provided analysis.py in \example\ as a starter script to control EasyDCP_Analysis. Most users should be able to run analysis.py with very little modification. To do: Complete API documentation for details on controlling EasyDCP_Analysis via python.

Key functions:

name	description
easydcp.Classifier()	Create a Classifier object using training data
easydcp.Plot()	Create a Plot object using input .ply file and Classifier object.
easydcp.Plot.remove_noise()	Filter out noise points on Plot object.
easydcp.Plot.dbscan_segment()	Run segmentation using DBSCAN algorithm.
easydcp.Plot.kmeans_split()	Separate plant point clouds and noise point clouds by point number.
easydcp.Plot.rank_split()	Alternative to kmeans_split. Discard noise point clouds by passing known numer of plants as a parameter.
easydcp.Plot.sort_order()	Sort Plant ID in order of increasing x-axis of plant center. Needed when using dbscan_segment or kmeans_split.
easydcp.Plot.xaxis_segment()	Alternative to dbscan_segment. Segment the Plot object into plants by passing the number of segments (plants) as a parameter. This function assumes plants are of approximately equal size and evenly spaced.
easydcp.Plot.get_traits()	Calculate traits for all segments (plants) within the Plot object.

Key Parameters:

name	default	affects	notes
percentile	98	plant height	adjust parameter in line 718 of base.py, __init__() :self.get_percentile_height(container_ht, ground_ht, percentile=95) 98 should give a near-maximum plant height in most cases. For a measurement closer to the true maximum plant height (highest plant point), increase to 99, 99.5, etc. A lower-quality point cloud (e.g., produced from an image set with low overlap or resolution, or low settings used by EasyDCP_Creation) may contain some noise points above the true plant top and could cause overestimation bias. In this case, consider decreasing to 95 or 90.
eps_points	10	segmentation	Set parameter when calling dbscan_segment Higher value may be used if segmentation fails, such as one plant being wrongly divided into two segments. In such case, we recommend trying 12,13,14. Higher eps_points value will increase processing time significantly.

1. Configure EasyDCP_Analysis before launching:

• Ensure your .py script points to correct training data and point clouds. Ensure container_ht is correctly set. - Training data is called as list in Classifier() function, path_list parameter. Vegetation file must be listed before background file. - Point clouds are loaded by plot_path and plot_set variables - Set plot_path to the folder containing .ply files from EasyDCP_Creation step 5. - Important! Use / in your file path, not \ - container_ht parameter is passed within get_traits() function. Default: 0.12 (meters, 12 cm)
  • Ensure to import __init__ before import easydcp. This is already done in analysis.py.

  import __init__
  import easydcp as dcp

2. Execute your .py script using python, with EasyDCP root folder as working directory:

• Using Anaconda, open a terminal and navigate to your EasyDCP folder: `cd D:\Program\EasyDCP``

  (easydcp37) D:\Program\EasyDCP>python example\analysis.py

3. Outputs will be saved in the EasyDCP directory.
   • A folder will be created in the EasyDCP directory for each .ply file processed by EasyDCP. It will contain several intermediary .ply files (output by the classification, remove noise and segmentation steps) and a .png file is for each Plant object measured by the calculation step. The .png file contains an image of the plant point cloud, and visualization of phenotypic traits.
   • A folder called data_out will be created in the EasyDCP directory. It will contain:
     • Segmentation results. Each input .ply file will have a corresponding .png image file, showing an overhead view of the plant points and their segmentation results. (See Figure 1)
     • traits.csv, containing per-plant traits and metadata. This file can be read by a variety of softwares including Excel and R and used for data analysis. (See Figure 2)
   • The output locations may be adjusted using parameters of their respective functions, except the intermediary .ply files which can only be disabled via the write_ply parameter.

We strongly recommend to move your output data out of the EasyDCP directory. Data loss may occur when re-running the script, as folders will be re-created in the working directory, and may overwrite previous output folders.

Figure 1: Segmentation results

Figure 2: Traits