基于SVM及BPNN的辣椒红外光谱分析

Infrared Spectroscopic Analyses of Pepper based on Support Vector Machine and BPNN

利用小波变换结合反向传播网络（BPNN）和支持向量机（SVM）研究了朝天椒和灯笼椒的傅里叶变换红外（FTIR）光谱,样品1 750-950 cm^-1范围的红外光谱经多尺度一维连续小波变换（CWT）和离散小波变换分析,发现第20尺度的连续小波系数,提取该尺度3个区域的系数作为特征参数建立BPNN和SVM模型。结果表明,BPNN和SVM模型都能很好地区别两种辣椒。第5尺度的离散小波细节系数建立BPNN和SVM模型分类的正确率分别为93.3%、100%。小波变换结合BPNN和SVM用于傅里叶变换红外光谱技术中能够准确识别朝天椒、灯笼椒,为区分不同品种的辣椒提供了快速、有效的方法。

Fourier transform infrared（FTIR） spectroscopy combined with back propagation neural network（BPNN）, wavelet transform and support vector machine（SVM） were used to analyze Capsicum annuum L. var. conoide（Mill.） Irish and bell pepper. FTIR spectra of samples from C. annuum L. var. conoide（Mill.） Irish were obtained. The infrared spectra in the range of 1750-950 cm^-1were extracted by continuous wavelet transform（CWT） and discrete wavelet transform（DWT）. The decomposition level 10 was obviously different. Three regions of this level were selected as feature vector to train BPNN and the SVM models. Discrete wavelet transform detail coefficients（DWTDC） of level 5 were selected to train BPNN and SVM models. The recognition accurate rate of using BPNN and SVM was 93.3% and 100%, respctively. It is proved that FTIR spectroscopy combined with BPNN and SVM can be used to discriminate C. annuum L. var. conoide（Mill.） Irish and bell pepper.