Traditionally, the rice blast is diagnosed with the naked-eyes. There is an urgent need to provide a method that can identify the early rice blast without symptoms. In the paper, a method for the early rice blast diagnosis based on the Raman spectroscopy was proposed. Considering the compositions of the biological sample are complex, characteristic peaks are severely crossed, the biological fluorescence background and the noise are strong, and the Raman signal is weak. Different data pre-processing methods will lead to different diagnostic accuracies of Raman models, especially for biological samples. This paper proposed a method for modeling a Raman model based on data without pre-processing. In this method, the raw data are decomposed with Empirical Mode Decomposition (EMD) into several Intrinsic Mode Functions (IMF). Then, based on the self-correlation coefficient of the IMFs and the times of the IMFs crossing the zero Raman Intensity line, IMFs are filtered to get the signal components. Taking the characteristic peaks of the β-carotene, the chlorophyll, and the chitin as the initial characteristic variables, the characteristic variables of the signal components were screened based on Successive Projections Algorithm (SPA). Finally, the obtained characteristic variables were used to establish a Partial Least Squares (PLS) regression model for the rice blast classification, and the test classification accuracy was 94.12%, which was higher than that of models based on spectral data pre-processed by Moving Average Smoothing + Baseline offset, Savitzky Golay Smoothing + Baseline offset, Gaussian Filter Smoothing + Baseline offset and the dB5 wavelet, 3-layer decomposition, Stein Unbiased Risk Estimate, the modulus maximum value method +7 points, 3rd-order Polynomial Fitting.

传统上，稻瘟病是用肉眼诊断的。迫切需要提供一种可以识别出没有症状的早稻瘟病的方法。提出了一种基于拉曼光谱的稻瘟病早期诊断方法。考虑到生物样品的成分复杂，特征峰严重交叉，生物荧光背景和噪声强，拉曼信号弱。不同的数据预处理方法将导致拉曼模型的诊断准确性不同，尤其是对于生物样品。本文提出了一种无需预处理即可基于数据建模拉曼模型的方法。在这种方法中，原始数据通过经验模式分解（EMD）分解为几个固有模式函数（IMF）。然后，根据IMF的自相关系数和IMF穿越零拉曼强度线的次数，对IMF进行滤波以获得信号分量。以β-胡萝卜素，叶绿素和几丁质的特征峰为初始特征变量，基于逐次投影算法（SPA）筛选信号成分的特征变量。最后，利用所获得的特征变量建立了稻瘟病分类的偏最小二乘回归模型，测试分类准确率为94.12％，高于基于Moving预处理光谱数据的模型。平均平滑+基线偏移，Savitzky Golay平滑+基线偏移，高斯滤波器平滑+基线偏移和dB5小波，