Planktool is an automatic plankton image classification software. Simply put, you feed it an image and it draws a rectangle on identified plankton, with its corresponding class.

The following commands are avaiable:

$ python planktool.py gui
$ python planktool.py web

$ python planktool.py build-dataset
$ python planktool.py build-models
$ python planktool.py build

This repository has both web and graphical user interfaces, available through:

$ python planktool.py gui
$ python planktool.py web

The main difference is that the web interface can choose between different classifiers, while the GUI uses a Random Forest with Generic classes.

For convenience, some trained models are already provided with Planktool. However, to best suit your applications, you may wish to train classifiers yourself.

The input_images directory is used to train different classifiers. Planktool can extract features from your images and generated trained models for two different sets of classes derived from the directory structure:

• Generic classes are the root-level folders;
• Specific classes are the lowest-level folders.

For instance, suppose you have the following directory structure:

input_images
    |- A
    |  |- AA
    |  |  |- AAA (10 images)
    |  |  |- AAB (10 images)
    |  |
    |  |- AB (10 images)
    |
    |- B (10 images)

The generic classes would be composed by 30 As and 10 Bs, while specific would have 10 AAAs, 10 AABs, 10 ABs and 10 Bs.

You can then build the csv dataset using $ python planktool.py build-dataset, and build the classifiers with $ python planktool.py build-models. For convenience, as both commands are frequently used in sequence, you can just call:

$ python planktool.py build