傅里叶变换红外（FTIR）光谱结合基于小波变换(WT)的线性判别分析（LDA）和反向传播网络（BPNN）研究了甜橙炭疽病和正常果皮。样品的红外光谱经多尺度一维连续小波变换（CWT），发现第10尺度的小波系数存在着明显的差异，提取该尺度三个区域的系数作为特征参数建立LDA和BPNN模型，结果表明LDA模型对样品的识别效果比BPNN模型好。选取1750~950 cm-1 范围内的FTIR光谱进行5尺度离散小波变换(DWT)，选取第5尺度的逼近系数(DWTAC)和细节系数(DWTDC)建立LDA和BPNN模型，结果显示利用细节系数建立的模型比逼近系数识别效果好，LDA和BPNN模型对样品的识别正确率均为95%。结果表明小波变换结合LDA和BPNN用于傅里叶变换红外光谱技术能够准确地识别甜橙炭疽病和正常果皮，为甜橙病害检测提供快速和有效的方法。

Fourier transform infrared (FTIR) spectroscopy combined with linear discriminant analysis (LDA) and back propagation neural network (BPNN) based on wavelet transform (WT) is applied to study citrus osbeck anthracnose and healthy peel. Continuous wavelet transform (CWT) is implemented to the FTIR spectra of anthracnose and healthy peel. By comparison, the decomposition level 10 is obviously different and proposed to extract feature vectors, then three feature regions of level 10 are used to train LDA and BPNN models. The performance of LDA algorithm is better than BPNN. On the other hand, in order to extrude the differences between anthracnose and healthy peel, discrete wavelet transform (DWT) is used to compose all spectra in 1750~950 cm-1 range. Wavelet transform approximation coefficients (DWTAC) and discrete wavelet transform detail coefficients (DWTDC) of level 5 are used to train LDA and BPNN models. Results show that accuracy of both LDA and BPNN based on DWTDC (95%) is better than DWTAC. LDA and BPNN algorithms based on wavelet transform can be successfully used for identifying citrus osbeck anthracnose and healthy peel with FTIR spectroscopy. It also provides technology support to detect citrus anthracnose in early stage quickly and effectively.