Mushroom Species Classification in Natural Habitats Using Convolutional Neural Networks (CNN)

Abstract

Mushrooms are known for their significant nutritional value and are essential to the human diet. However, the dilemmas associated with ingesting poisonous mushroom species stress the critical need for accurate identification methods. Despite many efforts to identify mushroom species, these methods are often limited in identifying them from their natural habitat. This study addresses this gap by presenting a computer vision approach that uses machine learning for accurate and reliable image-based classification of mushrooms from their natural habitat. The proposed solution aims to enhance the safety of mushroom consumption by precisely classifying mushroom species. The images of mushroom species are taken from their natural habitat to increase their applicability in real-world scenarios. The study proposed Convolutional Neural Network (CNN) models and different image augmentation techniques to accurately identify one hundred and three (103) mushroom species. Evaluation of the model from the 20% of the test dataset showed an accuracy of 96.70% and high precision-recall and F1 score for each mushroom class. The study achieved a 4.4% increase in accuracy from the state-of-the-art approaches in mushroom species identification. This research is significant to mycologists, scientists, and the general public in promoting the safe usage of mushroom species.