Comparative Study of Transfer Learning Models for Mushroom Classification

Mushrooms are spore-bearing fruiting body with 90% of water and packed with fiber, vitamins, and minerals. It is one of the most commonly grown agriculture crop adopted by many of Asian farmers because of supportive climatic conditions and economic and health benefits. Much of the work carried out for binary classification of edible and nonedible mushroom. It is quite difficult to manually identify the different types of mushroom because of its varying shape and color at different stages of its growth. In this paper, authors have explored deep learning models convolutional neural network (CNN) and nine transfer learning models: Xception, InceptionResNetV2, ResNet50, ResNet50V2, InceptionV3, MobileNetV2, MobileNet, VGG16 (Visual Geometry Group), and VGG19 for classification of five types of mushrooms from Kaggle dataset. The dataset is quite challenging since the mushroom images are of different stages with varying background. The performance of the proposed work is measured in terms of precision, recall, overall accuracy, and F1-score. The VGG19 models resulted in the overall best accuracy of 96.28% and 98.09%, respectively, for training and validation data.