【摘要】 为提升基于金属氧化物半导体(Metal Oxide Semiconductor, MOS)传感器阵列估计气味源距离的精度，同时避免传统距离估计方法对手动参数设定的依赖性，提出了一种基于长短时记忆(Long Short-Term Memory, LSTM)网络的气味源距离估计方法。所提方法利用LSTM网络对长序列特征的获取能力，从MOS传感器阵列的信号中自动学习距离指标，从而实现端到端估计。搭建了气味扩散仿真平台，生成气味扩散仿真数据集用于网络训练、参数调优和验证测试。结果显示，所提出的模型在10 m范围内的平均估计误差为0.16 m,比基于统计特征的估计方法误差降低了一个数量级。最后分析了不同LSTM超参数对距离估计精度的影响，并且就网络对未知气味扩散环境条件的泛化适应能力进行了验证。更多还原

【Abstract】 To improve the accuracy of odor source proximity estimation(OSPE)based on metal oxide semiconductor(MOS)sensor array, while avoiding the dependence on manual parameters setting of traditional distance estimation indicators, an OSPE method based on long short-term memory(LSTM)network is proposed. The proposed method utilizes the LSTM network’s ability to learn long-sequence features, automatically obtaining distance indicators from the MOS sensor signal to realize end-to-end estimation. An odor diffusion simulation platform is built to generate odor diffusion datasets for network training, optimization and verification. The results show that the average estimation error of the proposed model is 0.16 m within a range of 10 m, which is lower than that of method based on statistical characteristics by an order of magnitude. The influence of different LSTM hyper-parameters on the accuracy of distance estimation is analyzed, and the generalization ability of the network under unknown odor diffusion environments is also verified.更多还原