Skip to content

Latest commit

 

History

History
136 lines (92 loc) · 3.72 KB

length_documentation.md

File metadata and controls

136 lines (92 loc) · 3.72 KB

Leaf Length and Leaf Width

Authors

Zeyu Zhang, Robert Pless

Overview

Description

This pipeline uses 3D scanner point cloud and raw data to find leaves and compute the leaf length.

  1. Using the connected component algorithm to find the candidate leaf regions

  2. Heuristic search leaves based on the component shape and area

  3. For each leaf, build a weighted graph G: (v, e) where v is

    the pixels inside the leaf, and the e is the distance between pixel and its neighbor (two points are neighbors if their distance within the image coordinate is less than a threshold).

Fig 3. Illustration of the leaf length and width. The pink contour is the outline of the leaf. The blue line is the path of leaf length and the orange line is the path of leaf width. Red dots shows the vertices of the approximated polygon. Green dots shows the quinquesection points on the path of leaf length.

  1. Find leaf length:

    a. Find the approximated polygon of the leaf contour using > Douglas-Peucker algorithm.

    b. For all pairs of points on the approximated polygon, find the > longest shortest path. The length of this path is considered > as the leaf length.

  2. Find leaf width:

    c. Find quinquesection points on the leaf length path.

    d. For each quinquesection point i,

    i.  Find the point **a** on the edge that has the shortest path
    > to the point **i**

ii. From the points that on the opposite of the **a** with
    > respect to **i**, find the point **b** has the shortest
    > path to the point **i**.

iii. The shortest path from **a** to **b** on the graph **G** is
    > the candidate leaf width.

    e. Choose the longest path from the candidate leaf width path as > the leaf width.

  3. After getting the whole season result, average the measurement per

    day per plot

  4. Using kalman filter to filter the result along the time for each

    plot

  5. Find the peak point and only keep the measurement before the peak.

Inputs

3D scanner raw data and point cloud data.

Outputs

CSV file that contains the leaf length of each plot per day

Implementation

Libraries used

Pandas

Numpy

Scipy

Sklearn-image

Networkx

References

Connected component: [https://en.wikipedia.org/wiki/Connected_component_(graph_theory)]{.underline}

Weighted graph: [http://mathworld.wolfram.com/WeightedGraph.html]{.underline}

Douglas, David H., and Thomas K. Peucker. "Algorithms for the reduction of the number of points required to represent a digitized line or its caricature." Cartographica: The International Journal for Geographic Information and Geovisualization 10.2 (1973): 112-122.

Rationale (e.g. why method x over y)

QA/QC

Failure conditions

Known issues

  1. When the plants grown, overlaps among leaves breaks leaves into

    pieces. Which result the measured leaf length goes down in the latter season.

  2. There is some noise (missing points in the point cloud) that can

    break the leaves apart. This may caused by the over/under exposure in the scanner reflection data.

{width="4.942708880139983in" height="3.941030183727034in"}