面對生物揮發性有機物快速辨識於微生物偵測與疾病篩檢的迫切需求，本研究旨在開發並驗證一套以極限學習機的晶體結構演算法（CS-ELM）為核心演算法的輕量化電子嗅覺邊緣運算系統。此系統搭載於樹莓派第五代嵌入式平台，整合16通道石英晶體微量天平，進行即時生物揮發物分析與辨識。電子鼻所蒐集之原始感測數據，首先經過基線移除及正規化等訊號預處理步驟，續針對預處理後數據萃取其斜率、積分值、最大值、最大差異值以及一階與二階導數等多維時序特徵，以建構豐富的特徵集供後續模型分析。最終，以前述特徵驅動CS-ELM演算法進行分類模型的建構與優化。
在麴菌辨識應用中，本系統針對三種常見麴菌：煙麴菌、黃麴菌與土麴菌，在不同培養天數（第一、二、三天，含空白對照組，共計十類別）所釋放的揮發性有機物進行了系統性量測與分析，共蒐集294筆感測數據。實驗結果顯示，部署於樹莓派上的CS-ELM演算法，在經過嚴謹的十倍交叉驗證後，對麴菌培養第二天之樣本展現出良好的辨識效能，準確率可達0.9897，整體平均辨識準確率亦達到0.88。此輕量化模型不僅確保了較高精度，其推論時間僅需21.7秒，CPU使用率為86.8%。模型的效能評估亦包含F1分數、敏感度、ROC曲線及混淆矩陣等多個指標。
針對COVID-19患者呼吸道揮發物的辨識應用，本研究收集了42位COVID-19確診患者及58位健康對照者的呼氣樣本，並利用本電子鼻系統與傳統氣相層析質譜儀進行了比較分析。研究發現，氣相層析質譜儀結合CS-ELM分析可達到0.97的準確率，但其分析流程耗時超過25分鐘。相較之下，本研究開發的電子嗅覺邊緣運算系統搭載CS-ELM演算法，量測週期為150秒，數據推論時間為數十秒，辨識準確率可達0.85。
總結而言，本研究開發之基於CS-ELM的輕量化電子嗅覺邊緣運算系統，在特定麴菌培養階段辨識及COVID-19呼吸氣體篩檢方面，均展現了其應用潛力與可行性。研究成果初步驗證了此技術路徑在所選應用場景中的有效性，其系統設計與實驗數據可為後續在相關領域進行更深入的探索，如擴展目標分析物範圍或進一步優化硬體功耗與反應時間等研究，提供參考依據。

This study aims to develop and validate a lightweight electronic olfaction edge computing system, based on the Crystal Structure Algorithm with Extreme Learning (CS-ELM), for the rapid identification of biological volatile organic compounds (VOCs) relevant to microbial detection and disease screening. This system is implemented on a Raspberry Pi 5 embedded platform, integrating a 16-channel quartz crystal microbalance (QCM) array for real-time analysis and identification of biological VOCs. Raw sensor data collected by the electronic nose first undergoes signal preprocessing steps including baseline removal and normalization. Subsequently, multi-dimensional time-series features such as slope, integral value, maximum value, maximum differences, and first and second derivatives are extracted from the preprocessed data to construct a rich feature set for subsequent model analysis. Finally, these features are used to drive the CS-ELM algorithm for the construction and optimization of classification models.
In the application of Aspergillus species identification, the system was used for the systematic measurement and analysis of VOCs released by three common Aspergillus species: Aspergillus fumigatus, Aspergillus flavus, and Aspergillus terreus, at different cultivation days (Day 1, Day 2, and Day 3, including a blank control group, totaling ten categories). A total of 294 sensor data entries were collected. Experimental results showed that the CS-ELM algorithm deployed on the Raspberry Pi, after rigorous 10-fold cross-validation, demonstrated excellent identification performance for samples from the second day of Aspergillus cultivation, achieving an accuracy of up to 0.9897, with an overall average identification accuracy of 0.88. This lightweight model not only ensured high precision but also required an inference time of only 21.7 seconds and a CPU usage of 86.8%. The model's performance evaluation also included multiple indicators such as F1-score, sensitivity, ROC curve, and confusion matrix.
For the identification of respiratory VOCs in COVID-19 patients, this study collected breath samples from 42 confirmed COVID-19 patients and 58 healthy controls. A comparative analysis was conducted using the developed e-nose system and traditional gas chromatography-mass spectrometry (GC-MS). It was found that while GC-MS combined with CS-ELM analysis could achieve an accuracy of 0.97, its analytical process was time-consuming, exceeding 25 minutes. In contrast, the electronic olfaction edge computing system developed in this study, equipped with the CS-ELM algorithm, required a measurement cycle of only 150 seconds and a data inference time of tens of seconds, achieving an identification accuracy of 0.85.
In conclusion, the CS-ELM-based lightweight electronic olfaction edge computing system developed in this study has demonstrated its application potential and feasibility in specific Aspergillus cultivation stage identification and COVID-19 breath gas screening. The research findings preliminarily validate the effectiveness of this technological approach in the selected application scenarios. Its system design and experimental data can provide a reference basis for subsequent, more in-depth exploration in related fields, such as expanding the range of target analytes or further optimizing hardware power consumption and response time.

[1] R. C. Barton, “Laboratory diagnosis of invasive aspergillosis: from diagnosis to prediction of outcome,” Scientifica, vol. 2013, no. 1, p. 459405, 2013.
[2] R. Dobiáš, D. A. Stevens, and V. Havlíček, “Current and future pathways in Aspergillus diagnosis,” Antibiotics, vol. 12, no. 2, p. 385, 2023.
[3] C. R. Thornton, “Breaking the mould–novel diagnostic and therapeutic strategies for invasive pulmonary aspergillosis in the immune deficient patient,” Expert Review of Clinical Immunology, vol. 10, no. 6, pp. 771-780, 2014.
[4] A. D. Subali, L. Wiyono, M. Yusuf, and M. F. A. Zaky, “The potential of volatile organic compounds-based breath analysis for COVID-19 screening: a systematic review & meta-analysis,” Diagnostic microbiology and infectious disease, vol. 102, no. 2, p. 115589, 2022.
[5] A. Kwiatkowski, S. Borys, K. Sikorska, K. Drozdowska, and J. M. Smulko, “Clinical studies of detecting COVID-19 from exhaled breath with electronic nose,” Scientific reports, vol. 12, no. 1, p. 15990, 2022.
[6] Y. Choi, B. Kang, and D. Kim, “Effective detection of indoor fungal contamination through the identification of volatile organic compounds using mass spectrometry and machine learning,” Environmental Pollution, vol. 363, p. 125195, 2024.
[7] M. Arora et al., “Machine learning approaches to identify discriminative signatures of volatile organic compounds (VOCs) from bacteria and fungi using SPME-DART-MS,” Metabolites, vol. 12, no. 3, p. 232, 2022.
[8] Z. Rezaee, S. S. Mohtasebi, and M. S. Firouz, “An electronic nose system supported by machine learning techniques for rapid detection of aspergillus flavus in pistachio,” Journal of Food Measurement and Characterization, vol. 18, no. 7, pp. 5757-5765, 2024.
[9] E. K. Soylu, S. Safi, M. A. Atalay, and M. Peker, “AI-Assisted Fungal Infection Detection Using Impedimetric Odour Biosensors,” Enginoscope, vol. 1, no. 1, pp. 1-12, 2025.
[10] M. L. Bastos et al., “Breaking barriers in Candida spp. detection with Electronic Noses and artificial intelligence,” Scientific Reports, vol. 14, no. 1, p. 956, 2024.
[11] P. Borowik et al., “Analysis of Wheat Grain Infection by Fusarium Mycotoxin-Producing Fungi Using an Electronic Nose, GC-MS, and qPCR,” Sensors, vol. 24, no. 2, p. 326, 2024.
[12] R. Vanaraj, B. IP, G. Mayakrishnan, I. S. Kim, and S.-C. Kim, “A Systematic Review of the Applications of Electronic Nose and Electronic Tongue in Food Quality Assessment and Safety,” Chemosensors, vol. 13, no. 5, p. 161, 2025.
[13] A. J. Moshayedi, A. S. Khan, M. Chen, and P. P. Piccaluga, “ENose: a new frontier for non-invasive cancer detection and monitoring,” Journal of Cancer Metastasis and Treatment, vol. 11, pp. N/A-N/A, 2025.
[14] L. Cheng, Q.-H. Meng, A. J. Lilienthal, and P.-F. Qi, “Development of compact electronic noses: A review,” Measurement Science and Technology, vol. 32, no. 6, p. 062002, 2021.
[15] N. Kemnitz et al., “Effects of Contagious Respiratory Pathogens on Breath Biomarkers,” Antioxidants, vol. 13, no. 2, p. 172, 2024.
[16] K.-L. Yu et al., “Exhaled Breath Analysis Using a Novel Electronic Nose for Different Respiratory Disease Entities,” Lung, vol. 203, no. 1, p. 14, 2025.
[17] M. C. Dixon, “Quartz crystal microbalance with dissipation monitoring: enabling real-time characterization of biological materials and their interactions,” (in eng), J Biomol Tech, vol. 19, no. 3, pp. 151-8, Jul 2008.
[18] J. Kankare, “Sauerbrey Equation of Quartz Crystal Microbalance in Liquid Medium,” Langmuir, vol. 18, no. 18, pp. 7092-7094, 2002/09/01 2002, doi: 10.1021/la025911w.
[19] R. A. Hall et al., “CO2 acts as a signalling molecule in populations of the fungal pathogen Candida albicans,” PLoS pathogens, vol. 6, no. 11, p. e1001193, 2010.
[20] D. R. Pentland, J. Davis, F. A. Mühlschlegel, and C. W. Gourlay, “CO2 enhances the formation, nutrient scavenging and drug resistance properties of C. albicans biofilms,” npj Biofilms and Microbiomes, vol. 7, no. 1, p. 67, 2021.
[21] S. D. Lowther et al., “How efficiently can HEPA purifiers remove priority fine and ultrafine particles from indoor air?,” Environment international, vol. 144, p. 106001, 2020.
[22] I. Oren, N. Haddad, R. Finkelstein, and J. M. Rowe, “Invasive pulmonary aspergillosis in neutropenic patients during hospital construction: before and after chemoprophylaxis and institution of HEPA filters,” American Journal of Hematology, vol. 66, no. 4, pp. 257-262, 2001.
[23] J. Beauchamp, J. Herbig, R. Gutmann, and A. Hansel, “On the use of Tedlar® bags for breath-gas sampling and analysis,” Journal of breath research, vol. 2, no. 4, p. 046001, 2008.
[24] M. Fido, S. Hersberger, A. T. Güntner, R. Zenobi, and S. Giannoukos, “Systematic study of polymer gas sampling bags for offline analysis of exhaled breath,” Journal of Breath Research, vol. 18, no. 4, p. 046009, 2024.
[25] R. E. Shirey, “SPME commercial devices and fibre coatings,” in Handbook of solid phase microextraction: Elsevier, 2012, pp. 99-133.
[26] E. Gionfriddo, “The Role of Microextraction Techniques in Facilitating Gas Chromatography Separation of Complex Mixtures,” 2021.
[27] J. Liu et al., “Blueberry and cranberry anthocyanin extracts reduce bodyweight and modulate gut microbiota in C57BL/6 J mice fed with a high-fat diet,” European journal of nutrition, vol. 60, pp. 2735-2746, 2021.
[28] W. M. N. H. W. Salleh, S. Khamis, and S. A. Abed, “Characterization of volatile components of Chassalia chartacea and its acetylcholinesterase inhibitory activity,” Chemistry of Natural Compounds, vol. 57, pp. 376-377, 2021.
[29] C. B. Mupparaju, M. Nalluri, A. S. Rongali, and B. Ananthan, “Food Safety Assurance: Crystal Strcture Algorithm with Extreme Learning Machine for Early Detection of Contaminated Products,” in 2024 3rd International Conference on Artificial Intelligence For Internet of Things (AIIoT), 2024: IEEE, pp. 1-6.
[30] J. Wang, S. Lu, S.-H. Wang, and Y.-D. Zhang, “A review on extreme learning machine,” Multimedia Tools and Applications, vol. 81, no. 29, pp. 41611-41660, 2022.
[31] S. Ding, X. Xu, and R. Nie, “Extreme learning machine and its applications,” Neural Computing and Applications, vol. 25, pp. 549-556, 2014.
[32] S. Talatahari, M. Azizi, M. Tolouei, B. Talatahari, and P. Sareh, “Crystal structure algorithm (CryStAl): a metaheuristic optimization method,” IEEE Access, vol. 9, pp. 71244-71261, 2021.
[33] M. Azizi, S. Talatahari, and P. Sareh, “Design optimization of fuzzy controllers in building structures using the crystal structure algorithm (CryStAl),” Advanced Engineering Informatics, vol. 52, p. 101616, 2022.
[34] J. V. Kumar and S. M. Shaby, “Design and optimization of triangular microstrip patch antenna using extreme learning machine (ELM)‐based improved Crystal Structure Algorithm (CryStAl) for C‐band application,” International Journal of Communication Systems, vol. 37, no. 7, p. e5721, 2024.
[35] Y. Yu, X. Si, C. Hu, and J. Zhang, “A review of recurrent neural networks: LSTM cells and network architectures,” Neural computation, vol. 31, no. 7, pp. 1235-1270, 2019.
[36] A. Sherstinsky, “Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network,” Physica D: Nonlinear Phenomena, vol. 404, p. 132306, 2020.
[37] K. Han, A. Xiao, E. Wu, J. Guo, C. Xu, and Y. Wang, “Transformer in transformer,” Advances in neural information processing systems, vol. 34, pp. 15908-15919, 2021.
[38] T.-T. Wong and P.-Y. Yeh, “Reliable accuracy estimates from k-fold cross validation,” IEEE Transactions on Knowledge and Data Engineering, vol. 32, no. 8, pp. 1586-1594, 2019.
[39] D. M. Ruszkiewicz et al., “Diagnosis of COVID-19 by analysis of breath with gas chromatography-ion mobility spectrometry-a feasibility study,” EClinicalMedicine, vol. 29, 2020.
[40] S. Grassin-Delyle et al., “Metabolomics of exhaled breath in critically ill COVID-19 patients: A pilot study,” EBioMedicine, vol. 63, 2021.