Analisis Pengaruh Data Augmentasi Pada Klasifikasi Bumbu Dapur Menggunakan Convolutional Neural Network

Abstract

Indonesia is a country that produces the best spices and herbs. The types of kitchen spices and spices are indeed very diverse and almost similar, so many people cannot distinguish between the types. So it is necessary to create a digital image processing model to sort out the types of spices and herbs. Due to the minimal data in making the classification model, the use of augmented data is used to make the data more diverse. This study aims to compare architecture by applying augmentation and not applying augmentation from photo data of kitchen spices. From the two models, it will be known which level of accuracy of each model is higher than the two models. This study uses the Convolutional Neural Networks method using a mapped architecture. Then use the confusion matrix method for the results after testing is carried out on data testing. The test was carried out with 80 testing data images consisting of own photo images and internet search images. The result of this research is obtained a model for the classification of kitchen spices. The test results show that CNNs that do not apply augmentation from the model only get an accuracy of 54%, while CNNs that apply augmentations from the model get an accuracy of 80%. The application of augmentation to the model gets higher accuracy because the process of using augmentation is to increase the amount of data by creating new data from existing data so as to make more image data. For internet search data and own photos, higher accuracy is data from internet searches. The model that has the highest accuracy is then implemented into a web-based application.