论文部分内容阅读
利用中红外光谱分析技术进行人体血糖浓度检测时,由于高频噪声与低频基线难以避免的混入,以致很难从光谱数据中提取微弱的血糖信息,因此,提出了一种改进的小波分析光谱预处理方法以期去除其影响。该方法首先按照经验将光谱数据进行尺度J的小波分解,而后通过能量谱分析确定最终的分解尺度J。对尺度J下的细节部分进行极大极小阈值滤波,将其他各尺度下的细节部分直接置0,同时在低频部分以二次曲线拟合由人体组织散射引入的基线漂移并加以去除。将该预处理方法应用于人体血糖无创检测实验数据,以交互验证评价模型,预测值与参考值的相关系数为0.88,预测均方根误差为1.14 mmol/L,模型的预测精度得到较大幅度提高。
When using human infrared spectroscopy to detect human blood glucose concentration, it is difficult to extract weak blood glucose information from the spectral data due to high-frequency noise mixed with low-frequency baselines. Therefore, an improved wavelet analysis spectroscopy Approach to remove its impact. The method firstly decomposes the spectral data according to the wavelet decomposition of scale J according to experience, and then determines the final decomposition scale J by energy spectrum analysis. The minima threshold filtering is applied to the detail part of the scale J, and the detail part of other scales is directly set to zero. At the same time, the baseline drift introduced by human tissue scattering is fitted with a quadratic curve at the low frequency part and removed. The preprocessing method was applied to the noninvasive detection of blood glucose in the human body to validate the model. The correlation coefficient between the predicted value and the reference value was 0.88 and the root mean square error of prediction was 1.14 mmol / L. The prediction accuracy of the model was significantly improve.