A Comparison Analysis Between ResNET50 and XCeption for Handwritten Hangeul Character using Transfer Learning

Dede Kurniadi; Nabila Putri Nurhaliza; Benedicto B. Balilo Jr; Hilmi Aulawi; Asri Mulyani

doi:10.15575/join.v10i2.1606

Authors

Dede Kurniadi Department of Computer Science, Institut Teknologi Garut, Indonesia https://orcid.org/0000-0002-1455-2821
Nabila Putri Nurhaliza Department of Computer Science, Institut Teknologi Garut, Indonesia
Benedicto B. Balilo Jr CS/IT Department, College of Science, Bicol University, Philippines
Hilmi Aulawi Department of Industrial Engineering, Institut Teknologi Garut, Indonesia
Asri Mulyani Department of Computer Science, Institut Teknologi Garut, Indonesia https://orcid.org/0000-0001-8496-7726

DOI:

https://doi.org/10.15575/join.v10i2.1606

Keywords:

Adam Optimizer, Grad-CAM, Hangeul Character, K-Fold, ResNet50, Transfer Learning, Xception

Abstract

The enthusiasm for Korean pop culture in Indonesia has contributed to a growing interest in learning the Korean language, including its writing system, Hangeul, which currently ranks as the 6th most studied language. Hangeul has a unique structure, where each character is arranged in syllabic blocks of consonants and vowel combinations. The main challenge in Korean character classification lies in the similarity between characters and the complex structure, making it more difficult for models to recognize. This study aims to compare two deep convolutional neural networks are ResNet50 and Xception, using transfer learning for handwritten Hangeul character classification. While previous studies have examined CNN-based character recognition, this study highlights the effectiveness of deeper architectures with limited yet augmented data. Unlike earlier works, it incorporates Grad-CAM visualizations, transfer learning with partial fine-tuning, and multiple train-test ratios to analyze model behavior. A total of 1,920 images across 24 classes were evaluated using 5-fold cross-validation, with extensive augmentation and preprocessing to simulate variation. The Machine Learning Life Cycle (MLLC) framework assessed model performance through accuracy, precision, recall, F1-score, and AUC. Both models achieved high performance, with ResNet50 consistently outperforming Xception in most folds, especially in precision and F1-score. ResNet50 achieved perfect scores (100%) across all metrics, while Xception also performed strongly with up to 99.74% accuracy. These results indicate that ResNet50 is more effective in classifying Korean letters on the dataset used in this study. For future research, a robustness evaluation can be applied using data that was not included in previous training or testing.

References

[1] S. Jung-Eun and S. Yoonhee, “Hallyu Story with Statistic Indonesia,” Korean Foundation for International Cultural Exchange (KOFICE), Korea, pp. 1–23, Feb. 2022.

[2] T. P. Pramadya and J. Oktaviani, “Korean Wave (Hallyu) dan Persepsi Kaum Muda di Indonesia: Peran Media dan Diplomasi Publik Korea Selatan,” Insignia: Journal of International Relations, vol. 8, no. 1, p. 87, Apr. 2021, doi: 10.20884/1.ins.2021.8.1.3857.

[3] Hestianingsih, “Pengaruh K-Wave, Korea Masuk Daftar 10 Bahasa Terpopuler 2023,” Wolipop.

[4] A. A. Rosyadi, “Karakteristik Surel Bisnis Berbahasa Korea,” JLA (Jurnal Lingua Applicata), vol. 4, no. 1, p. 13, Oct. 2020, doi: 10.22146/jla.57448.

[5] C. Young-mee, S. Ho-Min, S. Sung-Ock, and C. Schulz, Integrated Korean, Third. University of Hawaii Press, 2020.

[6] Y. Rizki, R. Medikawati Taufiq, H. Mukhtar, and D. Putri, “Klasifikasi Pola Kain Tenun Melayu Menggunakan Faster R-CNN,” IT Journal Research and Development, vol. 5, no. 2, pp. 215–225, Jan. 2021, doi: 10.25299/itjrd.2021.vol5(2).5831.

[7] W. S. Eka Putra, “Klasifikasi Citra Menggunakan Convolutional Neural Network (CNN) pada Caltech 101,” Jurnal Teknik ITS, vol. 5, no. 1, Mar. 2016, doi: 10.12962/j23373539.v5i1.15696.

[8] Y. Septiana, A. Mulyani, D. Kurniadi, and H. Hasanudin, “Handwritten Recognition of Hiragana and Katakana Characters Based on Template Matching Algorithm,” IOP Conf Ser Mater Sci Eng, vol. 1098, no. 3, p. 032093, Mar. 2021, doi: 10.1088/1757-899X/1098/3/032093.

[9] R. E. Wiratna, A. Sahputro, B. D. Hadi, and E. Y. Puspaningrum, “Pengenalan Karakter Hijaiyah Menggunakan Metode Convolutional Neural Network (CNN),” Jurnal Teknologi Informasi dan Komunikasi, vol. 17, no. 1, Jul. 2022, doi: 10.33005/scan.v17i1.2937.

[10] S. Oktaviani, C. A. Sari, E. Hari Rachmawanto, and D. R. Ignatius Moses Setiadi, “Optical Character Recognition for Hangul Character using Artificial Neural Network,” Institute of Electrical and Electronics Engineers Inc., Sep. 2020, pp. 34–39. doi: 10.1109/iSemantic50169.2020.9234215.

[11] A. Aris Widodo, M. Y. Izza Mahendra, and M. Z. Sarwani, “Recognition of Korean Alphabet (Hangul) Handwriting into Latin Characters Using Backpropagation Method,” International Journal of Artificial Intelligence & Robotics (IJAIR), vol. 3, no. 2, pp. 50–57, Nov. 2021, doi: 10.25139/ijair.v3i2.4210.

[12] Radikto and Rasiban, “Pengenalan Pola Huruf Hangeul Korea Menggunakan Jaringan Syaraf Tiruan Metode Backpropagation dan Deteksi Tepi Canny,” 2022. doi: https://doi.org/10.31004/jpdk.v4i5.6722.

[13] E. Hartati, D. Alamsyah, and Nataliatara, “Korean Letter Handwriting Recognition Using Convolutional Neural Network Method Vgg-16 Arsitektur Architecture,” International Journal of Artificial Intelligence and Robotic Technology, vol. 1, no. 3, Mar. 2021, Accessed: May 27, 2024. [Online]. Available: https://journal.sracademicpublishing.org/index.php/IJAIRTec/article/view/33/pdf

[14] A. D. Snowberger and C. H. Lee, “Handwritten Hangul Graphemes Classification Using Three Artificial Neural Networks,” Journal of Information and Communication Convergence Engineering, vol. 21, no. 2, pp. 167–173, Jun. 2023, doi: 10.56977/jicce.2023.21.2.167.

[15] A. Hussain and A. Aslam, “Ensemble-based Approach Using Inception V2, VGG-16, and Xception Convolutional Neural Networks for Surface Cracks Detection,” Journal of Applied Research and Technology, vol. 22, no. 4, pp. 586–598, Aug. 2024, doi: 10.22201/icat.24486736e.2024.22.4.2431.

[16] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” Dec. 2015, [Online]. Available: http://arxiv.org/abs/1512.03385

[17] F. Chollet, “Xception: Deep Learning with Depthwise Separable Convolutions,” Oct. 2016, [Online]. Available: http://arxiv.org/abs/1610.02357

[18] N. Duklan, S. Kumar, H. Maheshwari, R. Singh, S. D. Sharma, and S. Swami, “CNN Architectures for Image Classification: A Comparative Study Using ResNet50V2, ResNet152V2, InceptionV3, Xception, and MobileNetV2,” SSRG International Journal of Electronics and Communication Engineering, vol. 11, no. 9, pp. 11–21, Sep. 2024, doi: 10.14445/23488549/IJECE-V11I9P102.

[19] Y. Mulyani, R. Kurniawan, P. B. Wintoro, M. Komarudin, and W. M. Al-Rahmi, “International Journal of Aviation Science and Engineering AVIA Systematic Comparison of Machine Learning Model Accuracy Value Between MobileNetV2 and XCeption Architecture in Waste Classification System,” vol. 4, no. 2, pp. 75–82, 2022, doi: 10.47355/AVIA.V4I2.70.

[20] T. J. Bradshaw, Z. Huemann, J. Hu, and A. Rahmim, “A Guide to Cross-Validation for Artificial Intelligence in Medical Imaging,” Jul. 01, 2023, Radiological Society of North America Inc. doi: 10.1148/ryai.220232.

[21] I. K. Nti, O. Nyarko-Boateng, and J. Aning, “Performance of Machine Learning Algorithms with Different K Values in K-fold CrossValidation,” International Journal of Information Technology and Computer Science, vol. 13, no. 6, pp. 61–71, Dec. 2021, doi: 10.5815/ijitcs.2021.06.05.

[22] M. G. M. Abdolrasol et al., “Artificial Neural Networks Based Optimization Techniques: A review,” Nov. 01, 2021, MDPI. doi: 10.3390/electronics10212689.

[23] W. Jia, M. Sun, J. Lian, and S. Hou, “Feature dimensionality reduction: a review,” Complex and Intelligent Systems, vol. 8, no. 3, pp. 2663–2693, Jun. 2022, doi: 10.1007/s40747-021-00637-x.

[24] S. Agha, S. Nazir, M. Kaleem, F. Najeeb, and R. Talat, “Performance evaluation of reduced complexity deep neural networks,” PLoS One, vol. 20, no. 3 MARCH, Mar. 2025, doi: 10.1371/journal.pone.0319859.

[25] Z. Zhou, X. Yang, H. Ji, and Z. Zhu, “Improving the classification accuracy of fishes and invertebrates using residual convolutional neural networks,” ICES Journal of Marine Science, vol. 80, no. 5, pp. 1256–1266, Jul. 2023, doi: 10.1093/icesjms/fsad041.

[26] M. Abdullah Al Alamin and G. Uddin, “Challenges and Barriers of Using Low Code Software for Machine Learning,” arXiv e-prints, p. arXiv:2211.04661, Nov. 2022, doi: 10.48550/arXiv.2211.04661.

[27] Jamescasia, “Handwritten Hangul Characters,” Kaggle. Accessed: Jan. 22, 2025. [Online]. Available: https://www.kaggle.com/datasets/wayperwayp/hangulkorean-characters/data

[28] B. O. Soufiene and C. Chakraborty, Machine Learning and Deep Learning Techniques for Medical Image Recognition. in Advances in Smart Healthcare Technologies. CRC Press, 2023. [Online]. Available: https://books.google.co.id/books?id=BSfdEAAAQBAJ

[29] Y. Zhang, D. Hong, D. McClement, O. Oladosu, G. Pridham, and G. Slaney, “Grad-CAM helps interpret the deep learning models trained to classify multiple sclerosis types using clinical brain magnetic resonance imaging,” J Neurosci Methods, vol. 353, p. 109098, 2021, doi: https://doi.org/10.1016/j.jneumeth.2021.109098.

[30] S. Purnamawati, D. Rachmawati, G. Lumanauw, R. F. Rahmat, and R. Taqyuddin, “Korean Letter Handwritten Recognition Using Deep Convolutional Neural Network on Android Platform,” in Journal of Physics: Conference Series, Institute of Physics Publishing, Mar. 2018. doi: 10.1088/1742-6596/978/1/012112.

[31] Y. Son and J. W. Park, “Detecting Forged Audio Files Using ‘Mixed Paste’ Command: A Deep Learning Approach Based on Korean Phonemic Features,” Sensors, vol. 24, no. 6, Mar. 2024, doi: 10.3390/s24061872.

[32] K. Maharana, S. Mondal, and B. Nemade, “A review: Data Pre-processing and Data Augmentation Techniques,” Global Transitions Proceedings, vol. 3, no. 1, pp. 91–99, 2022, doi: https://doi.org/10.1016/j.gltp.2022.04.020.

[33] R. Sarki, K. Ahmed, H. Wang, Y. Zhang, J. Ma, and K. Wang, “Image Preprocessing in Classification and Identification of Diabetic Eye Diseases,” Data Sci Eng, vol. 6, no. 4, pp. 455–471, 2021.

[34] I. Muraina, “Ideal Dataset Splitting Ratios in Machine Learning Algorithms: General Concerns for Data Scientists and Data Analysts,” in 7th international Scientific Research Conference, 2022, pp. 496–504.

[35] M. Shafiq and Z. Gu, “Deep Residual Learning for Image Recognition: A Survey,” Applied Sciences, vol. 12, no. 18, p. 8972, Sep. 2022, doi: 10.3390/app12188972.

[36] D. A. Alkurdi, M. Cevik, and A. Akgundogdu, “Advancing Deepfake Detection Using Xception Architecture: A Robust Approach for Safeguarding against Fabricated News on Social Media,” Computers, Materials & Continua, vol. 81, no. 3, pp. 4285–4305, 2024, doi: 10.32604/cmc.2024.057029.

[37] I. H. Kartowisastro and J. Latupapua, “A Comparison of Adaptive Moment Estimation (Adam) and RMSProp Optimisation Techniques for Wildlife Animal Classification Using Convolutional Neural Networks,” Revue d’Intelligence Artificielle, vol. 37, no. 4, pp. 1123–1130, Aug. 2023, doi: 10.18280/ria.370424.

[38] A. D. Hutson and H. Yu, “The Sign Test, Paired Data, and Asymmetric Dependence: A Cautionary Tale,” American Statistician, vol. 77, no. 1, pp. 35–40, 2023, doi: 10.1080/00031305.2022.2110938.