红外光谱结合基于连续小波变换的BPNN识别柑橘碎叶病  

Recognition of Citrus Tatter Leaf Using Infrared Spectroscopy Combined with BPNN Based on Continuous Wavelet Transform

作　　者：赵兴祥[1] 欧全宏[1] 刘刚[1] 郝建明[1] 周湘萍[1] 

机构地区：[1]云南师范大学物理与电子信息学院,昆明市呈贡区雨花片区1号650500

出　　处：《光谱实验室》2014年第5期574-578,共5页Chinese Journal of Spectroscopy Laboratory

基　　金：国家自然科学基金项目（30960179）资助.

摘　　要：利用傅里叶变换红外光谱结合基于连续小波变换的偏最小二乘判别分析（PLS-DA）和反向传播网络（BPNN）识别柑橘碎叶病和正常叶。采用Haar、Mexh和Morlet小波为母小波对样品的红外光谱进行多尺度连续小波变换，经比较发现第7尺度小波系数具有明显的差异。提取该尺度3个区域的小波系数作为特征向量建立PLS-DA和BPNN模型。PLS-DA模型对未知样品的预测正确率为75％，BPNN模型对未知样品的预测正确率为95％,结果表明小波变换结合BPNN用于傅里叶变换红外光谱技术能够准确地识别柑橘碎叶病和正常叶，有望为柑橘病害检测提供快速、有效的方法。Partial Least Squares-discriminant analysis （PLS-DA） and Back Propagation Neural Network （BPNN） based on Continuous wavelet transform （CWT） is applied to identify citrus tatter leaf （CTL） and healthy leaves by Fourier transform infrared spectroscopy （FTIR）. CWT is applied to analyze the FTIR spectra of all samples. By comparison, the decomposition level 7 is obvious differences, and three regions of this level are selected as feature vector. This feature vector is used to train PLS-DA and BPNN models. The accuracy of BPNN （95%） is better than PLS-DA （75%）. It shows that LDA and BPNN algorithms based on CWT can be successfully used for identification citrus tatter leaf and healthy leaves with FTIR spectroscopy. It also provides technology suppoi＇t to detect citrus tatter leaf in early stage quickly and effectively.

关 键 词：傅里叶变换红外光谱 连续小波变换 碎叶病 反向传播网络 

分 类 号：O657.33[理学—分析化学]