尿液量測是現在健康檢測中一個重要手段,如今在尿液試紙的幫助下能簡單粗略
的比對出尿液中各物質是否超標,但是根據色卡的比對沒有辦法得到準確的濃度數
值,無法進一步達到醫療中判斷的標準。所以本實驗中使用微型光譜儀採集試紙反應
后的反射率,并通過數據分析方法建立一個物質反射率對濃度的數據庫,從而能快速
得出尿液中的各物質具體濃度值。本研究中基於特征波長演算法建立了一個人工智能
分類算法進行尿液濃度的偵測。在光譜預處理方面通過幾種平滑算法的對比,最後選
擇 Savitzky-Golay 濾波器對原始光譜進行平滑處理。而分類算法中,首先通過主成分分
析提取出盡可能多的特征,比起單一特征波長擁有更多有用的信息。對光譜通過對比
樸素貝葉斯、支持向量機、神經網絡的準確率,最後選擇使用神經網絡分類算法對樣
本進行分類。在最終樣本測試中分類準確度達到 95%左右。

Urine measurement is an important means in current health testing. Now, with the help of
urine test paper, it is easy to roughly compare whether the substances in the urine exceed the
standard, but there is no way to get accurate concentration values according to the color card
comparison. It is impossible to further meet the criteria for judgment in medical treatment.
Therefore, in this experiment, the micro-spectrometer is used to collect the reflectance after the
reaction of the test paper, and a data analysis method is used to establish a database of the
reflectance and concentration of the substance, so that the specific concentration value of each
substance in the urine can be quickly obtained. In this study, an artificial intelligence
classification algorithm was established based on the characteristic wavelength algorithm to
detect the urine concentration. In the aspect of spectral preprocessing, through several
comparisons of smoothing algorithms, the Savitzky-Golay filter is finally selected to smooth
the original spectrum. In the classification algorithm, firstly, as many features as possible are
extracted by principal component analysis, which has more useful information than a single
feature wavelength. By comparing the accuracy of naive Bayesian, support vector machine and
neural network, the model finally classifies the samples using neural network classification
algorithm. The classification accuracy reached about 95% in the final sample test.

[1] 《现代医学检验仪器导论》出版[J]. 上海医学检验杂志,2002(04):256.
[2] 蔡永军. 尿液检测的无试剂光谱法研究[D].天津大学,2004.
[3] 季江,高鹏飞,贾南南,杨蕊,郭汉明,瑚琦,庄松林. 自适应多尺度窗口平均光谱平滑[J]. 光谱学与光谱分析,2015,35(05):1445-1449.
[4] H. G. Schulze, R. B. Foist, K. Okuda, A. Ivanov, and R. F. Turner, “A small-window moving average-based fully automated baseline estimation method for Raman spectra,” Appl Spectrosc, vol. 66, no. 7, pp. 757-64, Jul, 2012.
[5] F. Qian, Y. Wu, and P. Hao, “A fully automated algorithm of baseline correction based on wavelet feature points and segment interpolation,” Optics & Laser Technology, vol. 96, pp. 202-207, 2017.
[6] A. Savitzky, and M. J. E. Golay, “Smoothing and Differentiation of Data by Simplified Least Squares Procedures,” Analytical Chemistry, vol. 36, no. 8, pp. 1627-1639, 1964/07/01, 1964.
[7] H. G. Schulze, R. B. Foist, A. Ivanov, and R. F. B. Turner, “Fully Automated High-Performance Signal-to-Noise Ratio Enhancement Based on an Iterative Three-Point Zero-Order Savitzky—Golay Filter,” Applied Spectroscopy, vol. 62, no. 10, pp. 1160-1166, 2008/10/01, 2008.
[8] 王阳,周云才.朴素贝叶斯分类算法的设计与分析[J].电脑知识与技术,2019,15(11):206-208.
[9] 李路,漆成莉,张鹏,胡秀清,顾明剑.应用主成分分析评估红外高光谱仪器噪声特性[J].光子学报,2019,48(02):154-162.
[10] 吴温博,金文标.基于主成分分析的高光谱图像降维[J].内蒙古科技与经济,2018(22):66-67.
[11] A.R. Barron, "Approximation and estimation bounds for artificial neural networks", Machine Learning, vol. 14, pp. 115-133, 1994.
[12] 蔡云芳,季凯帆,向永源.基于主成分分析的太阳光谱信息提取[J].光谱学与光谱分析,2018,38(09):2847-2852.
[13] 张静,刘忠宝,宋文爱,富丽贞,章永来.基于多类支持向量机的恒星光谱分类方法[J].光谱学与光谱分析,2018,38(07):2307-2310.
[14] 牟岩. 基于子空间的高光谱影像降维及分类研究[D].西安科技大学,2018.
[15] 李晓鹤. 机器学习在光谱特征选择和润滑脂类别鉴定中的应用研究[D].华北电力大学(北京),2018.
[16] 张明月,贺金兰,田尉霞.主成分分析在高光谱图像降维中的应用[J].内蒙古科技与经济,2017(15):88+90.
[17] 刁万英,刘刚,胡克林.基于高光谱特征与人工神经网络模型对土壤含水量估算[J].光谱学与光谱分析,2017,37(03):841-846.
[18] 刘昊. 基于统计机器学习的光谱识别技术[D].长春理工大学,2017.
[19] 张银露. 基于机器学习的染液光谱分类算法研究[D].浙江理工大学,2017.
[20] 李婵,万晓霞,刘强,梁金星,李俊锋.基于主成分分析的光谱重建训练样本选择方法研究[J].光谱学与光谱分析,2016,36(05):1400-1405.
[21] 付婉莹,刘东.基于人工神经网络的光谱反射率重建[J].包装工程,2015,36(07):103-107.
[22] 王茜蒨,黄志文,刘凯,李文江,阎吉祥.基于主成分分析和人工神经网络的激光诱导击穿光谱塑料分类识别方法研究[J].光谱学与光谱分析,2012,32(12):3179-3182.
[23] 李微微. 神经网络对爆炸物太赫兹光谱的识别[D].首都师范大学,2009.