Analyzing the Effectiveness of VGG Deep Learning Architecture for Mushroom Type Classification

Indonesia is a center of biodiversity, one of the diversity of plant species is mushrooms. Fungi are very simple plants, have nuclei, spores, without chlorophyll, in the form of cells or branching filaments with walls of cellulose or chitin or both. Identification of mushrooms is still difficult due to the large number of types of mushrooms, lack of knowledge about mushrooms, and lack of experts in the mushroom field. Additionally, most mushrooms have a high degree of similarity in certain characteristics, making it difficult for unskilled humans to visually identify the type of mushroom. Therefore, it is important to be able to classify the types of mushrooms so that the public will better understand the types of each mushroom. This research uses a transfer learning approach with the VGG-19 architecture to provide an accurate method for classifying fungal species. There are 4 model scenarios that are trained and the best model performance is obtained using hyperparameter number of epochs 50, batch size 64, and the SGD optimizer gets an accuracy of 77.3% in the training process. After testing using data testing and evaluating using a confusion matrix and classification report, an accuracy score of 70% was obtained