Overview

The primary functions of ParticleTrieur (PT) are:

  • Organise images according to their sample, resolution and other metadata.

  • Label images by their taxonomic class, and tag them according to their properties.

  • Use in-built AI to help predict the label of an image based on the most similar already labeled images.

  • Load a external trained convolutional neural network (CNN) to help label images.

  • Process particle images, such as by removing borders, normalising intensity, or centering the particle in the image.

  • Calculate morphology information such as circularity or solidity.

  • Export images, by processing them and sorting them into directories by their label.

  • See graphs of statistics such as label counts.

  • Configure and launch CNN training with the MISO particle classification library.

Glossary

PT allows the organisation of particle images according to their metadata. PT projects are saved in human-readable XML format.

A project consists of settings data and a list of particle metadata. The particle metadata is:

  • Filename: The path to the image. This will be relative to the project file if saved on the same drive as the project file, otherwise it will be absolute. Path (wikipedia)

  • Sample: The name of the sample from which the image was taken.

  • Index 1: An index value used to sort images and generate statistics. For example, if index 1 may be set to the depth at which a foraminifera sample was taken.

  • Index 2: A secondary index.

  • Resolution: The resolution of the image in pixels per millimetre.

  • GUID:: A globally unique identifier for the image.

  • Classifications: Labels and their confidence scores for this image, along with the id of the classifier. When manually labelling and image there will be only a single classification with the score set to 1.0. The classification items consist of two values:

    • code: The classification label code of the class.

    • score: The confidence score of the classification [0-1].

  • Tags: Tags to help sort images, see below.

  • Validator: The name of the person who validated the image label.

  • Morprhology: The calculated morphology of the particle.

  • Parameters: All other metadata for the particle.

The project metadata also contains a list of labels used for classification and tags:

  • Labels: Labels are the names of the various classes to which an image can be assigned. There are also non-taxonomic labels such as unlabeled and unsure which are used for images that have not been labelled or cannot be identified confidently. Taxonomic labels are used to train the CNN model.

  • Tags: Tags are used to help sort the images, but are not used in CNN training. There are some build in tags such as auto, which is given to images that are automatically classified by a CNN or other, and duplicate which is given to images identified as duplicates by the Find duplicates… command.

Launching PT

If not already installed, follow the instructions in Installation

  1. Open command prompt (Windows) or terminal (macOS / linux)

  2. Change directory to the one containing ParticleTrieur.jar

cd /PATH/TO/PARTICLETRIEUR

  1. Execute the jar file

java -jar ParticleTrieur.jar

The startup window will show:

../_images/startup.png

On the left:

  • New project will start a new project with the default settings

  • New project from template will prompt to open an existing project, from which a new project will be created with no images but using the same settings.

  • Open project will prompt to open and existing project.

On the right:

Recent projects show as list of up to the last 5 opened projects.

Main Window

Choosing and option will start the program after a moment to load the internal AI algorithms. On the left is a list of particles and their metadata, on the right are the labelling and processing tools and common functions are on a toolbar at the top.

../_images/main.png