A Comparative Study of MobileNetv2 and VGG-16 Convolutional Neural Network Architectures for Identification of Medicinal Mushrooms

Convolutional Neural Networks (CNNs) have considerably enhanced the field of Artificial Intelligence (AI) during the past ten years through deep learning. The efficacy of CNNs has been established in research, particularly in agricultural automation. The researchers in this study assessed the performance of two CNN architectures - MobileNetv2 and VGG-16 - in predicting six medicinal mushrooms - Lions Mane, Oyster, Reishi, Shiitake, Shimeji, and Volva. The aim was to compare the accuracy levels achieved by each architecture. A dataset comprising of 600 image samples of six medicinal mushrooms was employed to train the two architectures, with both training procedures employing an 80 by 20 ratio. 80 percent of the total data was utilized to train the models, while the remaining 20 percent was reserved for the validation set. The results show that MobileNetv2 yielded a testing accuracy rate of 97.3 percent while the VGG-16 achieved testing accuracy of 72.6 percent. This implies that accuracy in predicting medicinal mushroom using MobileNetv2 is higher than VGG-16 by 24.7 percent. Hence, using MobileNetv2 architecture will provide optimal results over VGG-16 in identifying medicinal mushrooms. For future studies, the researchers aim to train these medicinal mushrooms dataset in a continual learning scenario and evaluate the extent of catastrophic forgetting.