Deep Learning-Based Mushroom Species Classification: Analyzing the Performance of CNN on Diverse Fungi

This paper proposes the classification of mushrooms using the EfficientNetB7 deep learning architecture. Correct classification of mushrooms is important because most species are considered either edible or toxic due to ecological studies, food safety, and medicinal purposes. The data involve several species of mushrooms, including Agaricus, Amanita, Boletus, Cortinarius, Entoloma, Hygrocybe, Lactarius, Russula, and Suillus. In this data set, there are a total of 676 samples. Each species in the dataset would therefore fall under a classification based on an EfficientNetB7 model, actually designed for scalability and efficiency. This model reached an accuracy of 80% overall. The classification report has very promising precision, recall, and F1-scores for the nine species. The maximum F1-score achieved was by Boletus with 0.93, while Hygrocybe achieved the maximum precision of 0.96. However, certain species like Agaricus and Suillus achieved only a recall of 0.56 and 0.59, respectively, which could be reasons for further refinement. Separately, the macro-average F1-score was 0.78, while the weighted average came out to be 0.80; both are values of pretty consistent performance in most of the species. This work demonstrates great performance in mushroom classification by EfficientNetB7 and seamlessly provides meaningful insights into its use for species differentiation. Its performance might be further tested on other larger and more diverse datasets while studying, at the same time, techniques such as data augmentation and transfer learning in order to leverage accuracy in the classification of difficult-to- distinguish species. The results are useful in the development of efficient automatic identification systems for mushrooms, where the importance of correct classification is solicited not only by ecology but also by commerce.