Klasifikasi Citra Makanan Indonesia Menggunakan Convolutional Neural Network

Wicaksono, Satrio Hanif (2023) Klasifikasi Citra Makanan Indonesia Menggunakan Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 05111940000103-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 3 October 2025. Download (7MB) | Request a copy

Abstract

Makanan tradisional merupakan kearifan budaya kolektif dari sistem pangan yang telah diturunkan dari generasi ke generasi. Indonesia merupakan salah satu negara yang memiliki warisan budaya yang paling beragam di dunia dengan total 1.340 jenis suku di Indonesia. Indonesia merupakan destinasi kuliner yang kaya dan beragam, dengan warisan budaya dan kekayaan alam yang memberikan pengaruh signifikan terhadap masakannya. Namun, dengan keberagaman jenis makanan yang ada, seringkali sulit bagi wisatawan atau bahkan penduduk lokal untuk memahami dan mengenali makanan-makanan tersebut. Oleh karena itu, penting untuk dibuat sebuah sistem klasifikasi makanan. Food image recognition atau pengenalan citra makanan merupakan topik yang menarik karena digunakan dalam industri kesehatan untuk mendapatkan informasi seperti kalori dan nutrisi dalam makanan. Bagian menantang dari pengenalan citra makanan adalah latar belakang, intensitas, dan perspektif yang berbeda. Riset dalam food recognition mayoritas menggunakan dataset makanan cepat saji Amerika dan makanan Jepang. Dataset yang berbeda memerlukan perlakuan yang berbeda. Convolutional Neural Network (CNN) merupakan pilihan terbaik untuk membuat model yang paling efektif untuk citra makanan secara akurat. CNN telah menunjukkan kelebihan dalam melakukan image classification, recognition, segmentation, dan retrieval. Tugas Akhir ini menggunakan sepuluh jenis makanan Indonesia; rendang, sate, nasi goreng, bakso, soto, rawon, gado-gado, pempek, gudeg, dan bebek betutu menggunakan Convolutional Neural Network dengan arsitektur standar dan model transfer learning Xception, DenseNet, dan EfficientNet dengan optimizer function Adam dan RMSProp untuk melakukan klasifikasi citra. Model transfer learning dibuat dengan tanpa melakukan unfreeze lapisan, 5% lapisan unfreeze, 10% lapisan unfreeze, 20% lapisan unfreeze, dan 30% lapisan unfreeze. Peningkatan unfreeze lapisan dapat meningkatkan performa dari model transfer learning. Model transfer learning EfficientNet-B4 dengan hyperparameter tuning unfreeze 20% lapisan menghasilkan performa terbaik dengan hasil akurasi 0,9488, presisi 0,9482, recall 0,9461, dan F1-score 0,9470
=======================================================================================================================================
Traditional food represents the collective cultural wisdom of a food system that has been passed down from generation to generation. Indonesia is one of the countries with the most diverse cultural heritage in the world, with precisely 1,340 ethnic groups in Indonesia. Indonesia is a rich and diverse culinary destination, with cultural heritage and natural resources that significantly influence its cuisine. However, with the diversity of food types available, it is often challenging for tourists or even local residents to understand and recognize these foods. Therefore, it is important to create a food classification system. Food image recognition is an interesting topic as it is used in the healthcare industry to obtain information such as calories and nutrition in food. The challenging part of food image recognition is the different backgrounds, intensities, and perspectives. Research in food recognition mostly uses American fast food and Japanese food datasets. Different datasets require different treatments. Convolutional Neural Network (CNN) is the best choice to create the most effective model for accurately classifying food images. CNN has shown advantages in image classification, recognition, segmentation, and retrieval. This Final Project uses ten types of Indonesian food: rendang, satay, fried rice, meatballs, soto, rawon, gado-gado, pempek, gudeg, and bebek betutu using a Convolutional Neural Network with standard architecture and transfer learning models Xception, DenseNet, and EfficientNet with the Adam and RMSProp optimizer functions for image classification. Transfer learning models are created without unfreezing layers, 5% unfreezing layers, 10% unfreezing layers, 20% unfreezing layers, and 30% unfreezing layers. Enhanced layer unfreeze can improve the performance of the transfer learning model. EfficientNet-B4 transfer learning model with hyperparameter tuning unfreeze 20% layer produces the best performance with results of accuracy 0.9488, precision 0.9482, recall 0.9461, and F1-score 0.9470.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CNN, Klasifikasi Citra, Makanan tradisional, Transfer Learning; CNN, Traditional Food, Transfer Learning, Image Classification
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. R Medicine > R Medicine (General) > R858 Deep Learning
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Satrio Hanif Wicaksono
Date Deposited:	04 Aug 2023 01:51
Last Modified:	11 Aug 2023 08:42
URI:	http://repository.its.ac.id/id/eprint/101129

Actions (login required)

View Item