为了实现稻瘟病的快速、 准确和无损检测, 力求构建稻瘟病害预测模型。 根据水稻叶片相对病害面积将稻瘟病划分为3个等级, 通过激光诱导法采集不同病害等级的活体水稻叶片叶绿素荧光光谱。 选取502～830 nm波段激光诱导叶绿素荧光光谱(LICF)作为研究对象, 利用Savitzky-Golay平滑法(SG)和一阶导数变换(FDT)对光谱信息进行预处理, 通过主成分分析(PCA)方法获取经SG-FDT预处理后光谱的特征向量, 根据累积贡献率和方差选取前3个主成分进行分析。 将试验样本分为建模样本和检验样本, 以稻瘟病害等级为预测指标, 利用建模样本的133片叶片的光谱和病害信息分别结合判别分析(DA)、 多类逻辑回归分析(MLRA)和多层感知器(MLP)建立稻瘟病的预测模型, 利用检验样本的89片叶片的光谱和病害信息对所建模型进行预测检验, 完成对PCA-DA、 PCA-MLRA和PCA-MLP的对比寻优。 结果表明, PCA-DA, PCA-MLRA和PCA-MLP模型均能完成对稻瘟病害的预测, 但PCA-MLP模型的平均预测准确率能够达到91.7%, 相比PCA-DA和PCA-MLRA模型, 在稻瘟病害3个等级上均具有较好的分类和预测能力。

In order to detect rice blast more rapidly, accurately and nondestructively, the identification and early warning models of rice blast were established in the present research. First of all, rice blast was divided into three grades according to the relative area of disease spots in rice leaf and laser-induced chlorophyll fluorescence spectra of rice leaves at different disease levels were measured in the paddy fields. Meanwhile, 502～830 nm bands of laser-induced chlorophyll fluorescence spectra were selected for the study of rice blast. Savitzky-Golay(SG) smoothing and First Derivative Transform(FDT) were applied for the pretreatment of laser-induced chlorophyll fluorescence spectra. Then the method of Principal Components Analysis(PCA) was used to achieve the dimension reduction on spectral information, three principal components whose variance are greater than 1 and cumulative credibility is 99.924% were extracted by this method. Furthermore, the tentative data were divided into calibration set and validation set, the levels of rice blast were taken as the predictors. Combined with the calibration set which contains the disease and spectral information of 133 leaves, Discriminant Analysis(DA), Multiple Logistic Regression Analysis(MLRA) and Multilayer Perceptron(MLP) were used respectively to establish the identification and early warning models of rice blast. The Prediction examinations of the three models were made based on the validation set which contains the disease and spectral information of 89 leaves. The results show that all the models of PCA-DA, PCA-MLRA and PCA-MLP can carry on the prediction of rice blast, and the average prediction accuracy of PCA-MLP prediction model is 91.7% which is improved compared with PCA- DA and PCA- MLRA.