Evaluasi Kinerja Arsitektur Lightweight GhostNet pada Klasifikasi Tulisan Tangan Aksara Batak

Egi Dio Bagus Sudewo; Hainur Rasyid; Muhammad Kunta Biddinika; Kariyamin Kariyamin; Reni Yunita

doi:10.33506/insect.v12i2.5799

Authors

Egi Dio Bagus Sudewo Program Studi Sistem Informasi, Universitas Royal https://orcid.org/0009-0003-0239-3012
Hainur Rasyid Program Studi Sistem Informasi, Universitas Royal
Muhammad Kunta Biddinika Program Studi Magister Informatika, Universitas Ahmad Dahlan https://orcid.org/0000-0003-4104-3755
Kariyamin Kariyamin Program Studi Teknologi Informasi, Institut Teknologi dan Bisnis Muhammadiyah Wakatobi
Reni Yunita Program Studi Sistem Informasi, Universitas Royal

DOI:

https://doi.org/10.33506/insect.v12i2.5799

Keywords:

Aksara Batak, klasifikasi citra, deep learning, convolutional neural network, GhostNetv2

Abstract

Aksara Batak merupakan salah satu warisan budaya tertulis Indonesia yang keberadaannya semakin terancam akibat menurunnya penggunaan serta terbatasnya sumber daya digital untuk pembelajaran dan pelestariannya. Digitalisasi melalui teknologi pengenalan karakter otomatis menjadi salah satu solusi yang dapat mendukung upaya preservasi aksara tersebut. Namun, klasifikasi tulisan tangan Aksara Batak menghadapi tantangan berupa tingginya variasi bentuk karakter, kemiripan visual antaraksara, serta jumlah kelas yang besar. Penelitian ini bertujuan untuk mengevaluasi kinerja arsitektur lightweight GhostNetv2_100 dalam klasifikasi tulisan tangan Aksara Batak. Dataset yang digunakan terdiri atas 76.050 citra yang terbagi ke dalam 117 kelas, dengan skema pembagian data 70% data latih, 20% data validasi, dan 10% data uji. Tahap pra-pemrosesan meliputi konversi grayscale, thresholding adaptif, resize citra menjadi 224×224 piksel, normalisasi piksel, dan one-hot encoding label. Model dilatih menggunakan optimizer Adam, batch size 64, dan maksimum 30 epoch. Hasil eksperimen menunjukkan bahwa GhostNetv2_100 mampu mencapai akurasi sebesar 99,34%, precision sebesar 99,36%, recall sebesar 99,34%, dan F1-score sebesar 99,34% pada data uji. Dengan total parameter sebesar 5,02 juta (19,17 MB), beban komputasi hanya 0,02 GFLOPs, serta latensi inferensi rata-rata 33,74 ms per citra pada CPU, temuan ini membuktikan efisiensi GhostNetv2_100 untuk implementasi sistem klasifikasi Aksara Batak pada perangkat mobile.

References

[1] Y. Pratama, S. Try, N. Nainggolan, D. I. Nadya, and N. Y. Naipospos, “One-shot learning Batak Toba character recognition using siamese neural network,” Telkomnika, vol. 21, no. 3, pp. 600–612, 2023, doi: 10.12928/TELKOMNIKA.v21i3.24927.

[2] D. A. Dharmawan, B. M. Akbar, R. A. C. Leuveano, M. P. Tahya, and A. R. Yehudha, “Deep Learning Approaches for Batak Script Recognition : A Literature Review,” RSF Conf. Ser. Eng. Technol., vol. 4, no. 1, 2025, doi: https://doi.org/10.31098/cset.v4i1.1069.

[3] E. M. Zamzami and E. Nababan, “An AI-Based Approach to Evaluate Students’ Old Script Writings,” Int. J. Comput. Sci. Cult. Herit., vol. 12, no. 2, pp. 117–127, 2024.

[4] M. M. Sihotang and T. Sitanggang, “Cultural Heritage Preservation and Manuscript Conservation: Safeguarding the Laklak Batak Manuscript of Batak Culture in Indonesia,” J. Ilmu Pendidik. dan Hum., vol. 11, no. 2, pp. 85–101, 2022, doi: 10.35335/jiph.v11i2.17.

[5] R. D. Sianipar, “Implementation of YOLOv8 in Real-Time Batak Script Detection,” J. Comput. Vis. Pattern Recognit., vol. 8, no. 2, pp. 101–110, 2023.

[6] O. Ediansa and I. Riadi, “Klasifikasi Jenis Penyakit Pada Buah Nanas Menggunakan Pendekatan Model VGG16 dan EfficiennetB7,” vol. 12, no. 01, pp. 42–50, 2026.

[7] M. Yaqub et al., “State-of-the-art CNN optimizer for brain tumor segmentation in magnetic resonance images,” Brain Sci., vol. 10, no. 7, pp. 1–19, 2020, doi: 10.3390/brainsci10070427.

[8] K. F. Habie, “Impact of Optimizer Selection on MobileNetV1 Performance for Skin Disease Detection Using Digital Images,” JUTIF, vol. 6, no. 3, pp. 1589–1604, 2025, doi: https://doi.org/10.52436/1.jutif.2025.6.3.4685.

[9] X. Zhou, B. Huang, H. Fan, and P. Zhao, “A Lightweight Convolutional Neural Network for Face Recognition,” in 2024 International Conference on Decision Aid Sciences and Applications (DASA), 2024, pp. 1–5. doi: 10.1109/DASA63652.2024.10836448.

[10] O. S. Parulian, “Efficient Design and Compression of CNN Models for Rapid Character Recognition,” J. Ilmu Komput. dan Inf., vol. 18, no. 1, pp. 127–140, 2025, doi: 10.21609/jiki.v18i1.1443.

[11] I. G. S. Mas Diyasa, P. A. Wijaya, and Y. V Via, “Balinese Script Handwriting Recognition Using CNN and ELM Hybrid Algorithms,” J. Nas. Pendidik. Tek. Inform., vol. 14, no. 1, pp. 49–59, 2025, doi: https://doi.org/10.23887/janapati.v14i1.87524.

[12] A. Mulyanto, E. Susanti, F. Rossi, W. Wajiran, and R. I. Borman, “Penerapan Convolutional Neural Network (CNN) pada Pengenalan Aksara Lampung Berbasis Optical Character Recognition (OCR),” JEPIN (Jurnal Edukasi dan Penelit. Inform., vol. 7, no. 1, pp. 52–57, 2021, doi: https://doi.org/10.26418/jp.v7i1.44133.

[13] A. Rahim, F. Sthevanie, and K. N. Ramadhani, “Klasifikasi Aksara Lontara dari Sulawesi Selatan Menggunakan CNN,” Log. J. Penelit. Inform., vol. 2, no. 2, pp. 35–42, 2024, doi: https://doi.org/10.25124/logic.v2i2.8812.

[14] H. Guo, D. Yang, Y. Liu, and J. Zhao, “Script identification of ancient books by Chinese ethnic minorities using multi-branch DCNN and SPP,” Pattern Anal. Appl., vol. 26, no. 2, pp. 809–821, 2023, doi: 10.1007/s10044-023-01146-y.

[15] K. Han, Y. Wang, Q. Tian, J. Guo, C. Xu, and C. Xu, “GhostNet: More Features from Cheap Operations,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 1580–1589. doi: 10.1109/CVPR42600.2020.00165.

[16] E. D. B. Sudewo, M. K. Biddinika, and A. Fadlil, “DenseNet Architecture for Efficient and Accurate Recognition of Javanese Script Hanacaraka Character,” MATRIK J. Manajemen, Tek. Inform. dan Rekayasa Komput., vol. 23, no. 2, pp. 453–464, 2024, doi: 10.30812/matrik.v23i2.xxx.

[17] A. Prasetiadi, J. Saputra, I. Ramadhanti, A. D. Sripamuji, and R. R. Amalia, “Minimalist DCT-based Depthwise Separable Convolutional Neural Network Approach for Tangut Script,” J. Dinda Data Sci. Inf. Technol. Data Anal., vol. 3, no. 2, pp. 59–64, Jul. 2023, doi: 10.20895/dinda.v3i2.1106.

[18] M. A. Prameswari, M. D. Sulistiyo, and A. F. Ihsan, “Classification of Handwritten Sundanese Script via Transfer Learning on CNN-Based Architectures,” 2023, pp. 401–406. doi: 10.1109/ice3is59323.2023.10335382.

[19] J. Ouyang, H. He, Y. He, and H. Tang, “Dog recognition in public places based on convolutional neural network,” Int. J. Distrib. Sens. Networks, vol. 15, no. 5, 2019, doi: 10.1177/1550147719829675.

[20] Y. Zheng and Q. He, “Research and comparison of lightweight U-Net based on GhostNets for medical and remote sensing images,” in 2023 8th International Conference on Intelligent Computing and Signal Processing (ICSP), 2023, pp. 2082–2086. doi: 10.1109/ICSP58490.2023.10248608.

[21] D. R. K. Babu, P. R. Sowjanya, K. S. Trisha, and G. Kalyan, “TL-GhostNet: Transfer Learning Enhanced GhostNet for Attack Classification while Handling Big Data with Map-Reduce Framework,” in 2026 7th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI), 2026, pp. 1444–1450. doi: 10.1109/ICMCSI67283.2026.11412844.

[22] S. Kumar, A. Kumar, and K. Kumar, “GhostNet-YOLO Algorithm for Object Detection in UAV Image,” in 2023 Seventh International Conference on Image Information Processing (ICIIP), 2023, pp. 293–299. doi: 10.1109/ICIIP61524.2023.10537737.

[23] Z. Wang et al., “Micro-expression recognition method based on improved GhostNet combined with attention mechanism,” in 2025 5th International Conference on Big Data Engineering and Education (BDEE), 2025, pp. 119–123. doi: 10.1109/BDEE67464.2025.00025.