光谱采集过程中的各种时变噪声影响了动态光谱法血液成分无创检测定量校正模型的精度。 该文采用小波变换法， 在脉搏频段内对指端投射光谱的时域吸光度波形聚焦， 提高了动态光谱数据的信噪比和血液成分含量定量校正模型的精度。 对同一个体连续采集10次光谱数据， 引入小波变换去噪后动态光谱数据的平均相关系数r自0.979 6提升至0.990 3。 对110名志愿者进行血常规体检和指端透射光谱采集， 建立动态光谱数据与血糖浓度生化分析值之间的神经网络模型， 在引入小波变换去噪后， 预测集相关系数自0.6774提升至0.846 8， 平均相对误差自15.8%下降至5.3%。 实验表明， 引入小波变换可以有效地去除动态光谱数据中的噪声， 提高定量校正模型的精度， 推动了动态光谱法无创血液成分检测的发展。

Time-varying noises in spectra collection process have influence on the prediction accuracy of quantitative calibration in the non-invasive blood components measurement which is based on dynamic spectrum (DS) method. By wavelet transform, we focused on the absorbance wave of fingertip transmission spectrum in pulse frequency band. Then we increased the signal to noise ratio of DS data, and improved the detecting precision of quantitative calibration. After carrying out spectrum data continuous acquisition of the same subject for 10 times, we used wavelet transform de-noising to increase the average correlation coefficient of DS data from 0.979 6 to 0.990 3. BP neural network was used to establish the calibration model of subjects’ blood components concentration values against dynamic spectrum data of 110 volunteers. After wavelet transform de-noising, the correlation coefficient of prediction set increased from 0.677 4 to 0.846 8, and the average relative error was decreased from 15.8% to 5.3%. Experimental results showed that the introduction of wavelet transform can effectively remove the noise in DS data, improve the detecting precision, and accelerate the development of non-invasive blood components measurement based on DS method.