Abstract: Objective This study aims to evaluate the performance of a self-developed electronic nose under different experimental conditions and explore its application in human odor recognition, particularly for gender classification using convolutional neural network (CNN) algorithms. Methods Firstly, the performance of the electronic nose was evaluated using acetone, ethanol, and isoprene as target gases with a concentration of 2×10⁻⁶ mol/m³, under gas flow rates ranging from 500 to 2 500 mL/min. Multiple factors, including flow rate, enrichment time, storage medium, and storage time, were considered in the experiments. Secondly, odor samples were collected from 25 male and 25 female participants, and gender classification of the gas samples was performed using feature data obtained from the electronic nose, combined with a CNN algorithm. Results The electronic nose performed best at gas flow rates of 1 500-2 000 mL/min. The use of aluminum foil bags resulted in minimal gas attenuation and better preservation. The optimal enrichment time for detection was 10 minutes. The gender classification accuracy based on hand and foot odors was as follows: left hand 96.55%, right hand 93.10%, left foot 100%, and right foot 96.55%. Conclusion This study demonstrates that optimizing experimental conditions significantly improves the performance of the electronic nose. Furthermore, the integration of CNN algorithms proves effective in human odor recognition, achieving high classification accuracy and providing a feasible solution for practical application of electronic noses in human odor analysis.

Key words: high dimensional sensor array, electronic nose, convolutional neural network (CNN), human odor recognition

摘要： 目的 考察自主研制的电子鼻设备在不同实验条件下的性能表现，并探索其在实际人体气味识别中的应用，尤其是通过结合卷积神经网络算法对人体气味样本进行性别分类，进一步验证其在人体气味分析中的潜力。方法 先在气体流速为500~2 500 mL/min的条件下，选择丙酮、乙醇和异戊二烯为测试气体，浓度设为2×10-6 mol/m³进行电子鼻设备性能测试。实验中涉及流速、富集时间、保存介质和保存时间等多个因素。再采集25名男性和25名女性的手足气味样本，利用电子鼻获取的特征数据，结合卷积神经网络算法对气体样本进行性别分类。结果 选择1 500~2 000 mL/min的气体流速时，电子鼻的性能最佳。铝箔气袋保存效果较好，气体衰减最小。富集时间为10 min时，检测能力最强。基于手足气味的性别分类准确率分别为：左手96.55%、右手93.10%、左脚100%、右脚96.55%。结论 通过优化实验条件，电子鼻设备的性能得到了显著提升。同时，结合卷积神经网络算法能够有效进行人体气味识别，并取得较高的分类准确率，为电子鼻在人体气味识别领域的实际应用提供了可行的解决方案。

关键词: 高维传感器阵列, 电子鼻, 卷积神经网络, 人体气味识别