Identification of Medicinal Mushrooms using Computer Vision and Convolutional Neural Network

Abstract

Recent studies have demonstrated the predictive capability of deep learning methods in different agriculture fields. In this study, the researcher developed a device through the integration of Convolutional Neural Network (CNN) deep learning models and Raspberry Pi that can classify six medicinal mushrooms including Lion’s Mane, Oyster, Reishi, Shiitake, Shimeji, and Volva. The pre-trained CNN – Inception-V3 architecture was utilized to train 600 sample images of medicinal mushroom. The study employs the 80 by 20 ratio, in which the model was trained using 80 percent of the entire data and its performance was validated using the remaining 20 percent. Overall, the accuracy of the model achieved 92.7 percent. Although a relatively satisfactory performance was obtained, improvement of model performance should be sought using different optimization methods. Furthermore, for future studies, continual learning methods that alleviate catastrophic forgetting can be applied in the developed device to allow robust predictions with other types of mushrooms or other type of tasks.