Mushroom genera determination using machine learning
Date
2021
Authors
Olesk, Johanna
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Mushroom determination using classification manuals is a tedious and time-consuming task for
mycologists and mushroom hunters. Machine learning provides a tool to automate this process
based on mushroom images using a small dataset. Since mushroom genera level classification
has been understudied, it is important to direct attention to this matter. In this study, advanced
machine learning algorithms were used in order to classify Cantharellus, Coprinus, Pholiota and
Russula mushroom genera that are widely spread in Estonia. The classification was done based
on the image grayscale pixels. To improve the classification accuracy, majority voting and mean
rule methods from the ensemble-based classification were applied to the dataset. The highest
accuracy obtained was 75.38%, with the majority voting method fusing five high performing
classifiers. This study showed that ensemble methods improve the mushroom genera classification
accuracy compared to individual classifiers. In addition to a novel approach of classifying
mushrooms on the level of genera, a new labelled mushroom image dataset was collected that
can be used in the future for similar studies.
Description
Keywords
Machine learning, Ensemble learning, Mushroom genera classification, Image classification