Deep Learning Based Approach for Classification of Mushrooms

Abstract

Deep learning algorithms have produced amazing results in recent years when used to identify items in digital photographs. A deep learning technique is suggested in this work to classify mushrooms in their natural habitat. The study's objective is to identify the most effective method for categorizing mushroom images produced by well-known CNN models. This study will be helpful for the field of pharmacology, mushroom hunters who gather mushrooms in the wild, and it will help to lower the number of people who are at risk of becoming ill from poisonous mushrooms. Images are taken from data labelled by INaturalist specialist. The photographs show mushrooms in their natural environment and feature a variety of backgrounds. The "Mobilenetv2_GAP_flatten_fc" model, which was the study's top performer, had a training data set accuracy of 99.99%. It was 97.20% accurate in the categorization that was done using the validation data. Using the test data set, the classification accuracy was 97.89%. This paper presents the results of a performance comparison between the best-performing model and a multitude of state-of-the-art models that have undergone prior training. Mobilenetv2_GAP_flatten_fc model greatly outperformed the trained models, according to the precision, recall, F1 Score. This illustrates how the basic training process of the suggested model can be applied to enhance feature extraction and learning.