Study On Early Detection Methods In Rice Blast Based On Raman Spectroscopy

Recently, the precaution and control of rice blast in cold regions have taken effect. The existing prevention methods are spraying insecticides in the susceptible period or detecting the spores after the occurrence of rice blast. These detection methods not only increase labors, and the consumption of pesticides is also very large. It is neither simple nor environmental. Therefore, there is an urgent need for a low-consumption, simple and green method of detecting rice blast earlier in cold regions, and so as to take an effective precaution and reduce the use of pesticides.Raman spectroscopy is fast, non-destructive and green. Based on these characteristics, we collect rice leaves and stems in cold regions. After the data preprocessing, we identify the typical peak of rice leaves’ Raman spectra from the protein, nucleic acid, carbohydrate and other aspects of cell biophysics. By using principal component analysis, we distinguish the change law of Raman spectroscopy’s peak between the severe disease, mild disease and normal rice leaves. The differences in the spectrums of displacement, peak number and relative intensity is analyzed by using the scatter plot and the receiver operating characteristic curve. These spectrums include normal leaves and mild leaves blast, normal leaves and severe leaves blast, normal stems and stems blast, leaves blast and stems blast.Using Fisher linear discriminant method and BP neural network algorithm, the early detection model of rice blast was established by comparing analyzed results. The disease degree of rice leaves can be distinguished effectively by Fisher linear discriminant method; in BP neural network algorithm the root mean square error can reflect the forecast effect of unknown samples. The coefficient of correlation, R, can reflect the relativity between predicted value and actual value.The coefficient of correlation is 0.98122, and the predictive root mean square error is0.024904. This indicates that reaserches in detecting rice blast by using BP neural network algorithm is workable. This method has important significance in early prevention of rice blast, and it also provides theoretical and technological support for rice blast later preventins.