支持向量机的动物血液光谱特征提取和识别分类

利用光谱检测和数据挖掘实现不同种类动物血液光谱数据的精确识别与分类具有重要意义，目前尚未见到较为完善及普适的相关研究报道。实验采集了鸽、鸡、鼠、羊四种动物全血和红细胞溶液(浓度为1%)的荧光光谱数据；基于小波变换的软阈值去噪方法，首先对原始光谱数据进行去噪处理，并确定了717个原始特征(包括荧光峰强度值、荧光峰连线斜率等4类特征)；提出以“区分度统计量”为核心的特征提取方法，结合主成分分析法和平均影响值算法，实现了对717个原始特征到2个识别特征的高效筛选；进一步建立了径向基核函数的支持向量机分类器，对四类不同动物的全血荧光光谱数据实现了准确率为100%的识别分类，对红细胞荧光光谱数据实现了94.69%～99.12%的识别率；最后蒙特卡洛交叉验证的结果表明所提出的思路和方法对于动物全血溶液的识别分类具有较好的泛化能力，能对荧光光谱数据进行准确的识别分类，因此能够在进出口检查、食品安全、医药等领域发挥重要作用。针对动物血液荧光光谱，提出的基于“区分度统计量”的特征提取方法，相比于传统的人为特征选取方法，能够从大量原始特征中自动提取少量且有效的识别特征，具有较强的普适性和高效性，为其他领域的光谱特征提取和识别分类提供了一种新的思路。

Feature Extraction and Classification of Animal Blood Spectra with Support Vector Machine

It is of great significance to study how to use spectral detection technology and data mining technology to realize the accurate identification and classification of different animal blood spectral data ,and it has not yet seen relevant complete research conclusions and methods on animal blood identification and classification .Therefore ,the authors collected fluorescence spectra data of four kinds of animals ,including pigeon ,chicken ,mouse and sheep .Based on the soft threshold denoising method of wavelet transform ,the original spectral data were denoised ,and the 717 original features were determined .Following the ap-proach of"Distinguish statistic"proposed by the authors ,717 original features were extracted into 2 finally input features .Based on support vector machine ,the whole blood solution of different animals were 100% recognized ,while the red cell blood solution of different animals were 94.69% ~99.12% correctly recognized .Finally ,the Monte Carlo cross validation revealed that the method used in this paperhad a great generalization ability for whole blood solution of different animals ,which can play an impor-tant role in the import and export inspection ,food safety ,medicine and other fields .