簡易檢索 / 詳目顯示

回結果列表
研究生： 巫嘉德
Leon Ugalde, Daniel Alonso
論文名稱： 基於可解釋深度學習之 QCM 電子鼻訊號動態分析於病原菌與濃度分類之研究
Explainable Deep Learning Analysis of Signal Dynamics in QCM-Based Electronic Nose for Bacterial Pathogen and Concentration Classification
指導教授： 林哲偉
Lin, Che-Wei
學位類別： 碩士
Master
系所名稱： 工學院 - 生物醫學工程學系
Department of BioMedical Engineering
論文出版年： 2026
畢業學年度： 114
語文別： 英文
論文頁數： 119
中文關鍵詞： 電子鼻 、石英晶體微天平 、深度學習 、可解釋人工智 、Integrated Gradients 、病原菌分類 、濃度分類 、揮發性有機化合物 、訊號動態分
外文關鍵詞： Electronic nose, Quartz crystal microbalance (QCM), Deep learning, Explainable artificial intelligence (XAI), Integrated Gradients, Pathogen classification, Concentration classification, Volatile organic compounds (VOCs), Signal dynamics analysis
相關次數： 點閱：19  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 近年來，深度學習已廣泛應用於電子鼻（Electronic nose, E-nose）訊號分析並展現優異的分類效能，然而其黑箱特性限制了對模型決策機制及感測訊號動態的理解。目前，可解釋人工智慧（Explainable Artificial Intelligence, XAI）應用於時序性電子鼻訊號分析與病原微生物辨識之相關研究仍相當有限。
    本研究首次系統性地引入 XAI 技術，針對基於石英晶體微天平（Quartz Crys-tal Microbalance, QCM）之電子鼻訊號進行動態分析，探討不同量測階段（吸附、穩態、脫附）對模型決策之貢獻。研究提出一套整合一維卷積神經網路（1D-CNN）與 Integrated Gradients 歸因分析的框架，用以解釋模型在病原菌分類與濃度辨識任務中的決策行為。實驗資料包含五種細菌病原菌（Escherichia coli、Klebsiella pneumoniae、Pseudomonas aeruginosa、Streptococcus agalactiae 及 Staphylococcus aureus），於六種濃度層級（10²–10⁷ CFU/mL）下製備，共計 1,200 筆樣本，並採用留一交叉驗證（Leave-One-Out Cross-Validation, LOOCV）進行模型評估。
    實驗結果顯示，所提出之模型在病原菌分類任務中達到 99.58% 之準確率；在多類別濃度分類任務中，準確率介於 75.71% 至 98.21%；在高低濃度二元分類任務中，準確率介於 82.91% 至 100%。可解釋性分析結果顯示，病原菌辨識主要依賴穩態階段與脫附階段之訊號特徵，而濃度辨識則呈現系統性的歸因極性轉換現象低濃度樣本以脫附階段之正向歸因為主，高濃度樣本則以穩態階段之正向歸因為主此一現象與 QCM 質量負載感測原理相符。上述結果證實模型學習到具有物理意義的吸附脫附動態特徵，而非僅依賴訊號振幅進行分類。
    本研究透過訊號動態分析建立了深度學習模型決策與感測器物理行為之間的可解釋連結，不僅深化了對電子鼻系統運作機制的科學理解，亦為任務導向的感測器選擇與系統最佳化提供理論依據。

    Deep learning has been widely applied to electronic nose (E-nose) signal analysis, achieving excellent classification performance. However, its black-box nature limits the interpretability and understanding of model decision-making processes and underlying sensor–signal dynamics. To date, research on explainable artificial intelligence (XAI) applied to time-series E-nose signal analysis and pathogenic microorganism identification remains limited.
    This study presents the first systematic application of XAI for signal dynamics analysis in quartz crystal microbalance (QCM)-based E-nose systems, examining the contributions of different measurement phases (adsorption, steady-state, and desorption) to model decision-making. An integrated framework combining a one-dimensional convolutional neural net-work (1D-CNN) with Integrated Gradients attribution analysis was developed to interpret model decisions across multiple classification tasks, including pathogen discrimination and concentration differentiation. The experimental dataset comprised five bacterial pathogens (Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Streptococcus agalactiae, and Staphylococcus aureus) prepared at six concentration levels (10²–10⁷ CFU/mL), totaling 1,200 samples. Model performance was evaluated using leave-one-out cross-validation (LOOCV).
    The proposed model achieved 99.58% accuracy for pathogen classification, 75.71%–98.21% for multi-class concentration classification, and 82.91%–100% for binary (low vs. high) concentration classification. Explainability analysis revealed that pathogen discrimination primarily relies on steady-state and desorption phase features, whereas concentration differentiation exhibits systematic attribution polarity transitions—low-concentration samples are characterized by positive attributions during desorption, while high-concentration samples show dominant positive attributions during the steady-state phase—consistent with QCM mass-loading sensing principles. These findings demonstrate that the model learns biophysically meaningful adsorption–desorption dynamics rather than relying solely on signal amplitude.
    Through signal dynamics analysis, this work establishes an interpretable bridge between deep learning model predictions and sensor physical behavior, providing both scientific in-sight into E-nose system operation and a theoretical foundation for task-specific sensor selection and system optimization.

    摘要 I Abstract III List of Abbreviations V Table of Contents VI List of Figures XI List of Tables XIV Chapter 1 – Introduction 1 1.1 Motivation 1 1.2 Challenges in E-nose Signal Analysis 1 1.3 Need for Explainability in Sensor-Based Deep Learning 2 1.4 Problem Statement and Research Gap 3 1.5 Objectives of This Study 3 Chapter 2 – Literature Review 5 2.1 Electronic Nose Systems for Pathogen Detection 5 2.1.1 Fundamentals of E-nose Technology 5 2.1.2 VOC Generation by Pathogenic Microorganisms 6 2.1.3 Applications and Advantages of E-nose in Pathogen Detection 7 2.1.4 Challenges and Limitations 7 2.2 QCM Sensors in E-nose Applications 8 2.2.1 Operating Principles of QCM Sensors 8 2.2.2 QCM Signal Characteristics and Temporal Dynamics 9 2.2.3 Advantages and Limitations of QCM Sensors for Pathogen Detection 9 2.3 Signal Processing and Feature Extraction in E-nose 10 2.3.1 Signal Processing 10 2.3.2 Handcrafted Feature Extraction 10 2.3.3 Classical Machine Learning Classifiers 10 2.3.4 Limitations of Traditional Approaches 11 2.4 Deep Learning for E-nose Signal Analysis 11 2.4.1 Overview of Deep Learning Architecture in E-nose Applications 12 2.4.2 1D-CNN 12 2.4.3 Limitations and the Need for Explainability 13 2.5 Concentration Classification and QCM Adsorption Dynamics 13 2.5.1 Concentration Effects on Signal Dynamics 13 2.6 Explainable Artificial Intelligence for Sensor-Based Models 14 2.6.1 Overview of XAI Methods 14 2.6.2 Integrated Gradients for Time-Series Analysis 15 2.6.3 XAI Applications in E-nose Research 15 2.7 Research Gap and Motivation for This Study 16 Chapter 3 – Materials and Methods 19 3.1 E-Nose System and Dataset 19 3.1.1 QCM 19 3.1.2 Bacteria Dataset 20 3.1.3 Signal Acquisition Protocol 21 3.2 Data Preprocessing 22 3.2.1 Baseline Removal 22 3.2.2 Z-score Normalization 23 3.3 Deep Learning Model 23 3.3.1 1D-CNN Architecture 23 3.3.2 Training Configuration 24 3.3.3 Cross-Validation Strategy 24 3.4 Performance Evaluation Metrics 27 3.4.1 Confusion Matrix 27 3.4.2 Accuracy, Precision, Recall, and F1-score 28 3.5 Explainable Artificial Intelligence 28 3.5.1 Integrated Gradients Method 29 3.5.2 Baseline and Interpolation Steps 29 3.5.3 Attribution Visualization 30 Chapter 4 – Results 31 4.1 Overview of Experiments 31 4.1.1 Dataset Overview 31 4.1.2 Signal Phase Definitions 31 4.1.3 Experimental Design 32 4.1.4 Evaluation Metrics and Validation Strategy 33 4.1.5 Explainability Analysis Using Integrated Gradients 34 4.1.6 Implementation Environment 34 4.2 Pathogen Classification 34 4.2.1 Classification Results 35 4.2.2 Explainable Signal Dynamics and Sensor Contribution 36 4.3 Concentration Classification 44 4.3.1 Classification Results 44 4.3.2 Explainable Analysis of Concentration-Dependent Attribution Patterns 45 4.4 Binary Concentration Classification 56 4.4.1 Classification Results 56 4.4.2 Explainable Analysis of Binary Concentration Attribution Patterns 58 Chapter 5 – Discussion 63 5.1 Pathogen Classification Performance and Model Behavior 63 5.2 Concentration Classification Performance 64 5.3 Baseline Selection for Integrated Gradients Interpretation 66 5.4 Sensor-Level Interpretation Using Integrated Gradients 67 5.5 Signal Phase Interpretation Based on QCM Sensor Dynamics 69 5.6 Concentration-Dependent Attribution Patterns 70 5.7 Comparison with Existing Studies 72 5.8 Limitations and Future Work 74 Chapter 6 – Conclusion 76 6.1 Summary of Findings 76 6.2 Key Contributions 78 Appendix A – Confusion Matrix for Pathogen Concentration Classification 80 Appendix B – Integrated Gradients Signed Plot and Sensor Contribution for Pathogen Concentration Classification 83 Appendix C – Confusion Matrix for Binary Classification Between Low Concentration and High Concentration Across Different Pathogens 95 Appendix D – Integrated Gradients Signed Plot and Sensor Contribution for Low and High Concentration Classification 97 References 100

    [1] P. Rajapaksha, A. Elbourne, S. Gangadoo, R. Brown, D. Cozzolino, and J. Chapman, “A review of methods for the detection of pathogenic microorganisms, ” Analyst, vol. 144, no. 2, pp. 396–411, 2019.
    [2] O. Lazcka, F. J. Del Campo, and F. X. Munoz, “Pathogen detection: A perspective of traditional methods and biosensors,” Biosensors and bioelectronics, vol. 22, no. 7, pp. 1205–1217, 2007.
    [3] Y. Yang and L. Wei, “Application of E-nose technology combined with artificial neural network to predict total bacterial count in milk, ” Journal of Dairy Science, vol. 104, no. 10, pp. 10558–10565, 2021.
    [4] Y. Li, Z. Wang, T. Zhao, H. Li, J. Jiang, and J. Ye, “Electronic nose for the detection and discrimination of volatile organic compounds: Application, challenges, and perspectives, ” TrAC Trends in Analytical Chemistry, vol. 180, p. 117958, 2024.
    [5] B. Liu et al., “Lung cancer detection via breath by electronic nose enhanced with a sparse group feature selection approach, ” Sensors and Actuators B: Chemical, vol. 339, p. 129896, 2021.
    [6] Z. Ye, Y. Li, R. Jin, and Q. Li, “Toward accurate odor identification and effective feature learning with an AI-empowered electronic nose, ” IEEE Internet of Things Journal, vol. 11, no. 3, pp. 4735–4746, 2023.
    [7] P. Singh, U. Habiba, Z. Shafi, A. Noor, V. K. Pandey, and R. Singh, “Understanding the Concepts of Smart E‐Nose Technology in Combination With Machine Learning for New Era of Food Safety: An Advanced Review, ” Food Safety and Health, vol. 3, no. 4, pp. 518–534, 2025.
    [8] C. Fang et al., “Smart electronic nose enabled by an all‐feature olfactory algorithm, ” Advanced Intelligent Systems, vol. 4, no. 7, p. 2200074, 2022.
    [9] J. W. Gardner and P. N. Bartlett, “A brief history of electronic noses, ” Sensors and Actuators B: Chemical, vol. 18, no. 1-3, pp. 210–211, 1994.
    [10] L. D. Bos, P. J. Sterk, and M. J. Schultz, “Volatile metabolites of pathogens: a systematic review, ” PLoS pathogens, vol. 9, no. 5, p. e1003311, 2013.
    [11] A. D. Wilson and M. Baietto, “Applications and advances in electronic-nose technologies, ” sensors, vol. 9, no. 7, pp. 5099–5148, 2009.
    [12] N. Alanazi, M. Almutairi, and A. N. Alodhayb, “A review of quartz crystal microbalance for chemical and biological sensing applications,” Sensing and Imaging, vol. 24, no. 1, p. 10, 2023.
    [13] T. C. Pearce, S. S. Schiffman, H. T. Nagle, and J. W. Gardner, “Handbook of machine olfaction, ” Willy-VCH, Weinheim, 2003.
    [14] P. K. Misztal et al., “Emission factors of microbial volatile organic compounds from environmental bacteria and fungi, ” Environmental Science & Technology, vol. 52, no. 15, pp. 8272–8282, 2018.
    [15] E. Bonah, X. Huang, J. H. Aheto, and R. Osae, “Application of electronic nose as a non-invasive technique for odor fingerprinting and detection of bacterial foodborne pathogens: A review,” Journal of food science and technology, vol. 57, no. 6, pp. 1977–1990, 2020.
    [16] T. Dias et al., “A lab-made E-nose-MOS device for assessing the bacterial growth in a solid culture medium,” Biosensors, vol. 13, no. 1, p. 19, 2022.
    [17] B. H. Tozlu, C. Şimşek, O. Aydemir, and Y. Karavelioglu, “A high performance electronic nose system for the recognition of myocardial infarction and coronary artery diseases,” Biomedical Signal Processing and Control, vol. 64, p. 102247, 2021.
    [18] J. Qian, M. Lu, F. Tian, and R. Liu, “Study on sensor array optimization of medical electronic nose for wound infection detection, ” IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 69, no. 3, pp. 1867–1871, 2021.
    [19] C. Yang et al., “Analysis of volatile organic compounds and metabolites of three cultivars of asparagus (Asparagus officinalis L.) using E-nose, GC-IMS, and LC-MS/MS,” Bioengineered, vol. 13, no. 4, pp. 8866–8880, 2022.
    [20] R. R. Mohamed, R. Yaacob, M. A. Mohamed, T. A. T. Dir, and F. A. Rahim, “Food freshness using electronic nose and its classification method: A review,” International Journal of Engineering & Technology, vol. 7, no. 3.28, pp. 49–53, 2018.
    [21] X. Xu, H. Cang, C. Li, Z. K. Zhao, and H. Li, “Quartz crystal microbalance sensor array for the detection of volatile organic compounds,” Talanta, vol. 78, no. 3, pp. 711–716, 2009.
    [22] J. Kou, D. Tao, and G. Xu, “A study of adsorption of dodecylamine on quartz surface using quartz crystal microbalance with dissipation,” Colloids and Surfaces A: Physicochemical and Engineering Aspects, vol. 368, no. 1-3, pp. 75–83, 2010.
    [23] T. Julian, S. N. Hidayat, A. Rianjanu, A. B. Dharmawan, H. S. Wasisto, and K. Triyana, “Intelligent mobile electronic nose system comprising a hybrid polymer-functionalized quartz crystal microbalance sensor array, ” ACS omega, vol. 5, no. 45, pp. 29492–29503, 2020.
    [24] D. Compagnone et al., “Quartz crystal microbalance gas sensor arrays for the quality control of chocolate, ” Sensors and Actuators B: Chemical, vol. 207, pp. 1114–1120, 2015.
    [25] R. L. Pérez, C. E. Ayala, J.-Y. Park, J.-W. Choi, and I. M. Warner, “Coating-based quartz crystal microbalance detection methods of environmentally relevant volatile organic compounds, ” Chemosensors, vol. 9, no. 7, p. 153, 2021.
    [26] S. Marco and A. Gutierrez-Galvez, “Signal and data processing for machine olfaction and chemical sensing: A review, ” IEEE Sensors Journal, vol. 12, no. 11, pp. 3189–3214, 2012.
    [27] A. D. Wilson, “Diverse applications of electronic-nose technologies in agriculture and forestry, ” Sensors, vol. 13, no. 2, pp. 2295–2348, 2013.
    [28] T. Liu, W. Zhang, L. Ye, M. Ueland, S. L. Forbes, and S. W. Su, “A novel multi-odour identification by electronic nose using non-parametric modelling-based feature extraction and time-series classification, ” Sensors and Actuators B: Chemical, vol. 298, p. 126690, 2019.
    [29] Y.-H. Liao, C.-H. Shih, M. F. Abbod, J.-S. Shieh, and Y.-J. Hsiao, “Development of an E-nose system using machine learning methods to predict ventilator-associated pneumonia, ” Microsystem Technologies, vol. 28, no. 1, pp. 341–351, 2022/01/01 2022, doi: 10.1007/s00542-020-04782-0.
    [30] V. A. Binson, M. Subramoniam, and L. Mathew, “Prediction of lung cancer with a sensor array based e-nose system using machine learning methods,” Microsystem Technologies, vol. 30, no. 11, pp. 1421–1434, 2024/11/01 2024, doi: 10.1007/s00542-024-05656-5.
    [31] T. W. Xijia Zhang, Wangze Ni, Yongwei Zhang, Wen Lv, Min Zeng, Jianhua Yang, Nantao Hu, Rui Zhan, Guang Li, Zhiqiang Hong, Zhi Yang,, “Sensor array optimization for the electronic nose via different deep learning methods,” Sensors and Actuators B: Chemical, vol. 410, p. 135579, 2024.
    [32] Y. L. Zengyi Tang, Mianyi Xiang, Daxiang Cui, Qichao Li,, “Discrimination of VOCs with e-nose consisting of a single MEMS sensor using lightweight deep learning model based on SqueezeNet,” Sensors and Actuators B: Chemical, vol. 422, p. 136640, 2025.
    [33] S. Kiranyaz, T. Ince, O. Abdeljaber, O. Avci, and M. Gabbouj, “1-D convolutional neural networks for signal processing applications, ” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2019: IEEE, pp. 8360–8364.
    [34] A. O. Ige and M. Sibiya, “State-of-the-art in 1d convolutional neural networks: A survey,” IEEE Access, 2024.
    [35] Z. Ye, Y. Liu, and Q. Li, “Recent Progress in Smart Electronic Nose Technologies Enabled with Machine Learning Methods” Sensors, vol. 21, no. 22, p. 7620, 2021. [Online]. Available: https://www.mdpi.com/1424-8220/21/22/7620.
    [36] X. Liu, G. Wei, A. He, W. Zhang, and S. Jiao, “1D CNN-LSTM based electronic nose algorithm for disinfectant concentration detection,” in 2023 IEEE SENSORS, 2023: IEEE, pp. 1–4.
    [37] Q. D. N. Nguyen, H. V. A. Le, T. Nakano, and T. H. Tran, “Wine quality assessment through lightweight deep learning: integrating 1D-CNN and LSTM for analyzing electronic nose VOCs signals,” Applied Computing and Informatics, 2024, doi: 10.1108/aci-10-2023-0098.
    [38] K. R. E. a. P. Naval, “1D-CNN Model for Fast Assessment of Beef Freshness from Noisy Electronic Nose Signal,” 6th International Conference on Computer Communication and the Internet (ICCCI), Tokyo, Japan, pp. 67–71, 2024, doi: 10.1109/ICCCI62159.2024.10674182.
    [39] A. Paleczek and A. Rydosz, “The effect of high ethanol concentration on E-nose response for diabetes detection in exhaled breath: Laboratory studies, ” Sensors and Actuators B: chemical, vol. 408, p. 135550, 2024.
    [40] R. Laref, E. Losson, A. Sava, K. Adjallah, and M. Siadat, “A comparison between SVM and PLS for E-nose based gas concentration monitoring,” in 2018 IEEE international conference on industrial technology (ICIT), 2018: IEEE, pp. 1335–1339.
    [41] S.-T. Kim, I.-H. Choi, and H. Li, “Identification of multi-concentration aromatic fragrances with electronic nose technology using a support vector machine, ” Analytical Methods, vol. 13, no. 40, pp. 4710–4717, 2021.
    [42] H. Bakiler and S. Güney, “Estimation of concentration values of different gases based on long short-term memory by using electronic nose,” Biomedical Signal Processing and Control, vol. 69, p. 102908, 2021.
    [43] R. Dwivedi et al., “Explainable AI (XAI): Core ideas, techniques, and solutions, ” ACM computing surveys, vol. 55, no. 9, pp. 1–33, 2023.
    [44] M. T. Ribeiro, S. Singh, and C. Guestrin, “" Why should i trust you?"” Explaining the predictions of any classifier," in Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, 2016, pp. 1135–1144.
    [45] S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions, ” Advances in neural information processing systems, vol. 30, 2017.
    [46] R. Sayres et al., “Using a deep learning algorithm and integrated gradients explanation to assist grading for diabetic retinopathy, ” Ophthalmology, vol. 126, no. 4, pp. 552–564, 2019.
    [47] M. Sundararajan, A. Taly, and Q. Yan, “Axiomatic attribution for deep networks, ” in International conference on machine learning, 2017: PMLR, pp. 3319–3328.
    [48] Y. Fukui, K. Minami, G. Yoshikawa, K. Tsuda, and R. Tamura, “Harnessing Explainable AI to Explore Structure–Activity Relationships in Artificial Olfaction, ” ACS Applied Materials & Interfaces, vol. 17, no. 37, pp. 52728–52737, 2025.
    [49] H. Mei, J. Peng, T. Wang, B. Zhang, and F. Xuan, “An interpretable sensor selection strategy for Array optimization and performance enhancement, ” ACS sensors, vol. 10, no. 9, pp. 6700–6713, 2025.
    [50] R. Dutta, E. L. Hines, J. W. Gardner, and P. Boilot, “Bacteria classification using Cyranose 320 electronic nose, ” Biomedical engineering online, vol. 1, no. 1, p. 4, 2002.
    [51] Z. Xu, Q. Kang, T. Cai, T. Huang, P. Zhang, and D. Shen, “A quartz crystal microbalance study of the adsorption kinetics of 1, 2-dichloroethane on metal-organic frameworks [Cu (C24H22N4O3)]· CH2Cl2 film, ” Thin solid films, vol. 556, pp. 325–332, 2014.
    [52] C. Z. Q. Zhou, L. Zhu, and J. Wang, , “A review on rapid detection of modified quartz crystal microbalance sensors for food: Contamination, flavour and adulteration, ” TrAC Trends Anal. Chem, vol. 157, p. 116805, 2022.

    下載圖示
    校外：立即公開
    QR CODE