光谱技术在生物和医学检测方面具有积极的应用前景。 由于血液成分的复杂性和类同性, 有关不同动物血液光谱识别分类的技术研究尚未出现较为完善的结论。 基于机器学习理论, 以BP神经网络为工具, 建立了对不同动物血液荧光光谱进行特征提取和识别分类的方法。实验采用Cary Eclipse光谱仪分别采集了鸽、 鸡、 鼠、 羊四种动物不同浓度(1%和3%)的全血与红细胞荧光光谱数据(每个类型样本各50组数据); 基于移动平滑算法对原始数据进行了平滑处理, 以减少实验仪器噪声对特征提取和识别分类的影响; 进一步根据血液光谱数据的特性, 该文出了“组合放大”的特征提取方法, 并建立了BP神经网络分类器进行训练和识别。 相比于常用的光谱数据(单一)特征, 提出的“组合放大”特征和所设计的BP神经网络能对不同动物、 不同类型(全血与红细胞)、 不同浓度(1%和3%)的血液荧光光谱实现100%的准确分类, 同时神经网络测试误差均远小于设定的允许误差值。 研究的动物血液光谱特征提取及识别技术具有较好的普适性和可靠性, 在农业、 食品检查、 以及生物医学检测等方面均可发挥重要作用。

There is no doubt that spectrum technology has a positive role in applied prospects of biological and medical testing. Because of the complexity and the similarity ofblood component, study on recognition and classificationof different animal’s blood is still an open issue. Based on the theory of machine learning, by BP neural network, the authorsproposed a methodoffeature extraction and classification for different animal’s blood fluorescence spectra. In this experiment, fluorescence spectra data of whole blood and red blood cell with different concentration (1% and 3%) is collected, respectively. By neighborhood average method, the original data is denoised in order to reduce the impact of noiseon thefeature extraction and classification. For the specialty of blood fluorescence spectra, the authors proposed a new feature extraction method of “Combination and Amplification method”, and established a BP neural network classifier. Compared with other common spectrafeature, “Combination and Amplification”feature and the BP neural network classifiercan achieve good recognition and classification for different animal’s blood fluorescence spectra, and the test error is much less than allowable variation. The technologies in this paper can play an important role inmedical examination, agriculture, and food safety testing.