基於IMU感測器使用機器學習預測呼吸頻率

Abstract

現今使用IMU感測器預測人體呼吸頻率的研究，主要聚焦在利用主成分分析(Principal Components Analysis)降維與多種濾波器過濾掉非呼吸的訊號，進而預測人體呼吸頻率。我們的實驗設置與方法利用主動降噪的概念，提出了使用可適性濾波器(Adaptive Filtering)以及頻譜圖(Spectrogram)結合機器學習預測人體呼吸頻率。結果顯示，頻譜圖結合機器學習的方法除了可以在人體處於靜態活動(例如坐、站)的條件下能準確預測人體呼吸頻率外，特別在人體處於動態活動(例如走路、跑步)的條件下會比只使用可適性濾波器和其他論文的方法更準確且一致。

Recent research on human respiratory rate estimation using IMUs (Inertial Measurement Units) has mainly focused on applying PCA (Principal Component Analysis) for dimension reduction and using various filters to remove non-respiratory signals, thereby predicting the respiratory rate. Our experimental setup and methodology introduce the concept of ANC (Active Noise Cancellation), proposing the use of adaptive filtering and spectrograms combined with machine learning to predict human respiratory rate. The results show that the method using spectrograms combined with machine learning can accurately predict respiration rate not only during static actions (such as sitting or standing) but also achieves greater accuracy and consistency compared to adaptive filtering methods and previous works, especially during dynamic actions (such as walking or running).