Komparasi EfficientNetV2-S dan ConvNeXt-Tiny untuk Klasifikasi Napas Normal-Abnormal pada ICBHI 2017
DOI:
https://doi.org/10.33005/santika.v6i1.1081Keywords:
suara pernapasan, ICBHI 2017, log mel spectrogram, EfficientNetV2-S, ConvNeXt-TinyAbstract
ifikasi suara pernapasan berbasis kecerdasan buatan berpotensi mendukung deteksi dini gangguan pernapasan secara lebih objektif dan konsisten. Penelitian ini bertujuan membandingkan kinerja EfficientNetV2-S dan ConvNeXt-Tiny untuk klasifikasi siklus napas pada dataset ICBHI 2017 menggunakan representasi log-mel spectrogram. Data diproses pada tingkat respiratory cycle dan dibagi menggunakan subject-independent split 60:40, dengan 20% dari train pool digunakan sebagai data validasi. Model dilatih pada skema empat kelas, yaitu normal, crackle, wheeze, dan both, kemudian hasil prediksi dipetakan ke skema dua kelas, yaitu normal dan abnormal. Tahap praproses meliputi segmentasi audio, resampling, normalisasi, augmentasi audio, pembentukan log-mel spectrogram, dan augmentasi citra. Pada evaluasi empat kelas, ConvNeXt-Tiny menunjukkan performa lebih baik dengan Average Score 0,5190, accuracy 0,5427, dan AUC OVR Macro 0,7176, sedangkan EfficientNetV2-S memperoleh Average Score 0,4915, accuracy 0,5050, dan AUC OVR Macro 0,7063. Pada evaluasi dua kelas, kedua model menunjukkan peningkatan performa; ConvNeXt-Tiny memperoleh Average Score 0,6309 dan Harmonic Score 0,6295, sedangkan EfficientNetV2-S memperoleh Average Score 0,6269 dan Harmonic Score 0,6199. Hasil penelitian menunjukkan bahwa kedua model lebih efektif untuk klasifikasi normal-abnormal dibandingkan klasifikasi rinci empat kelas, dengan ConvNeXt- Tiny unggul secara keseluruhan.
References
B. M. Rocha et al., “An open access database for the evaluation of respiratory sound classification algorithms,” Physiological Measurement, vol. 40, no. 3, p. 035001, 2019, doi: 10.1088/1361-6579/ab03ea.
D. Bardou, K. Zhang, and S. M. Ahmad, “Lung sounds classification using convolutional neural networks,” Artificial Intelligence in Medicine, vol. 88, pp. 58–69, 2018, doi: 10.1016/j.artmed.2018.04.008.
N. Asatani, T. Kamiya, S. Mabu, and S. Kido, “Classification of Respiratory Sounds Using Two Resolution Spectrograms and TF-CRNN,” in Proceedings of the 33rd Annual Meeting of Biomedical Fuzzy Systems Association, 2020, pp. 64–67. https://www.jstage.jst.go.jp/article/pacbfsa/33/0/33_64/_pdf/-char/en.
N. Asatani, T. Kamiya, S. Mabu, and S. Kido, “Classification of respiratory sounds using improved convolutional recurrent neural network,” Computers & Electrical Engineering, vol. 94, art. no. 107367, 2021, doi: 10.1016/j.compeleceng.2021.107367.
Z. Neili and K. Sundaraj, “A comparative study of the spectrogram, scalogram, melspectrogram and gammatonegram time-frequency representations for the classification of lung sounds using the ICBHI database based on CNNs,” Biomedical Engineering / Biomedizinische Technik, vol. 67, no. 5, pp. 367–390, 2022, doi: 10.1515/bmt-2022-0180.
W. Ariyanti, K.-C. Liu, K.-Y. Chen, and Y. Tsao, “Abnormal Respiratory Sound Identification Using Audio-Spectrogram Vision Transformer,” arXiv preprint, 2024, doi: 10.48550/arXiv.2405.08342.
Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A ConvNet for the 2020s,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 11976–11986, doi: 10.1109/CVPR52688.2022.01167.
M. Tan and Q. V. Le, “EfficientNetV2: Smaller Models and Faster Training,” arXiv preprint, 2021, doi: 10.48550/arXiv.2104.00298.
K. M. Lim, C. P. Lee, Z. Y. Lee, and A. Alqahtani, “EnViTSA: Ensemble of Vision Transformer with SpecAugment for Acoustic Event Classification,” Sensors, vol. 23, no. 22, p. 9084, 2023, doi: 10.3390/s23229084.
S. Hamdi, M. Oussalah, A. Moussaoui, and M. Saidi, “Attention-based hybrid CNN-LSTM and spectral data augmentation for COVID-19 diagnosis from cough sound,” Journal of Intelligent Information Systems, vol. 59, pp. 367–389, 2022, doi: 10.1007/s10844-022-00707-7.
T. Truong, M. Lenga, A. Serrurier, and S. Mohammadi, “Fused Audio Instance and Representation for Respiratory Disease Detection,” Sensors, vol. 24, no. 19, p. 6176, 2024, doi: 10.3390/s24196176.
I. Loshchilov and F. Hutter, “Decoupled Weight Decay Regularization,” arXiv preprint, 2019, doi: 10.48550/arXiv.1711.05101
P. Goyal, P. Dollár, R. Girshick, P. Noordhuis, L. Wesolowski, A. Kyrola, A. Tulloch, Y. Jia, and K. He, “Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour,” arXiv preprint, 2017, doi: 10.48550/arXiv.1706.02677.
I. Loshchilov and F. Hutter, “SGDR: Stochastic Gradient Descent with Warm Restarts,” arXiv preprint, 2016, doi: 10.48550/arXiv.1608.03983.
T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, “Focal Loss for Dense Object Detection,” in Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2980–2988, doi: 10.1109/ICCV.2017.324.
