DeepStomata (stomatal pore quantifier)

Facial Recognition Technology for Automated Stomatal Aperture Measurement

At a glance

A three step image analysis program for quantification of stomatal aperture from bright field images.

1. Identifying the coordinate of the stomata by HOG + SVM.
2. Classifying the status (open, partially open, closed, false positive) by CNN.
3. Pore quantification responsive to the object status.

Author

Yosuke Toda Ph.D (Agriculture) JST PRESTO / ITbM invited researcher Institute of Transformative Bio-Molecule (ITbM) Nagoya University tyosuke@aquaseerser.com

Author's env.

Tested in Mac OSX, anaconda env. python 3.5

Requirements

highly recommended to run under anaconda

python>3 #python 3.5 preferred for easy installation of opencv3 via conda

matplotlib==2.0.0

numpy==1.11.2

scipy==0.18.1

scikit_image==0.14.0

tensorflow==0.12.0rc0

PIL==4.3.0

common==0.1.2

cv2==1.0

dlib==19.1.0

setuptools==32.3.1

Installation

1. Download this repository.

2. Unzip.

3. Open terminal.

4. Move to the Unzipped directory.

pip install .

Note

• Tensorflow must not be ver. 1.0.. Codes are not compatible.

• Several packages such as cv2 and dlib cannot be installed via pip in anaconda environment. In such cases, comment out the requirements.txt like the following

#cv2 ==1.0
#dlib==19.1.0

updated 11/7/2018, commented out by default

and install respectively via conda install

ex.

conda install -c menpo opencv3 dlib

Usage

• In terminal

Analyze a directory containing jpeg images in the example folder

from deepstomata import *
deepstomata("PATH_TO_THE_EXAMPLE_FOLDER/examples")

will generate annotated folder, FOLDER_NAME_all.csv, FOLDER_NAME_clasification_count.csv, vervose folder