A Study on the Classification of Local Varieties of Chili Pepper Using Convolutional Neural Networks

Abstract

This study focuses on the classification of local varieties of chili pepper using a convolutional neural network (CNN). The four classes of local varieties of chili pepper that will be included are Siling Labuyo (Capsicum frutescens), Siling Haba (Capsicum annuum), Siling Tingala (a hybrid of Capsicum frutescens and Capsicum annuum), and Siling Demonyo (Naga Viper pepper). The ResNet50 model will be employed as the convolutional neural network, and it will be compared against different architectures such as EfficientNet and VGG16. The configuration of hyperparameters, including learning rate and epoch, will also be explored to enhance the performance of the ResNet50 architecture. A newly created chili pepper image dataset will undergo a review and be processed through data augmentation techniques to increase the number of training data. The comparative results show that the ResNet50 model outperformed the EfficientNet and VGG16 models throughout all tests in terms of accuracy, sensitivity, specificity, and precision.