PlantRootAnalysis

Spring 2022 SULI Internship Program with Dr. Daniela Ushizima from Berkeley Laboratory.

How to start the analysis

1. Download and open FINAL_ALGORITHMS folder on Google Colaboratory (two files in folder)
2. Run all code on a Python notebook

Project Overview

• Importance: Characterizing root system architecture furthers our understanding of the development of plants, which can lead to increased crop production and environmental protection.
• Through computer vision, we developed four semantic segmentation algorithms for Spring Wheat plant roots. Segmented binary images are essential to the quantification and phenotyping of roots.
• Automatic Thresholding Algorithms: Otsu and Li Method
• Machine Learning Algorithms: Random Forest and MLP Classifiers

Automatic Thresholding Algorithm

The thresholding segmentation algorithms performs binary classification consisting of 4 steps:

• image denoising
• applying automatic threshold (Otsu v. Li)
• background cleaning
• implementing morphological operations

Machine Learning Classifiers

The ML segmentation algorithms performs binary classification consisting of 4 steps:

• Background Cleaning
• Feature Selection
• Determine optimal classifier parameters
• Train and Test Models (Random Forest v. MLP)

Measuring Plant Roots

Recorded properties for scientific purposes within Berkeley Lab through skimage.measure.regionprops()

• area
• perimeter
• feret diameter max
• solidity

Results Summarized

The automatic thresholding algorithms outperformed the machine learning models. Ostu's method provided the best dice coefficient score above 75%.