Performance Evaluation of Pre-Trained Convolutional Neural Network and Transfer Learning for Classification of Spices and Herbal Medicines in Madura

Spices are one of Indonesia's wealth. Based on the data owned. The Negara Rempah Foundation has around 400 to 500 species of herbs worldwide, and 275 types of spices are found in Southeast Asia, especially in Indonesia. The types of spices are diverse and have similarities to one another, so it isn't easy to distinguish. Therefore, to maintain the knowledge about herbs that Indonesia has, especially in Madura, applying an accurate classification of spices is necessary to preserve people's knowledge about spices. In addition, in the industrial sector, it can increase efficiency in the spice industry. The use of technology in the classification of herbs can increase efficiency in the spice industry. With the right technology, the time needed to identify the types of spices can be accelerated, minimizing the risk of human error. The limited image of herbs is also a problem in classifying types of sauces. Convolutional Neural Network (CNN) with transfer learning architecture is an image classification method that performs well on a limited number of datasets. Experiments were carried out using 6 CNN architectures: Xception, MobileNetV2, DenseNet201, VGG16, VGG19, and ResNet50. Ten types of spices are classified, namely ginger, kencur, turmeric, galangal and temulawak. Based on the experiments' results, Xception is the best architecture, with an F1 score of 98.23%.