Hyperspectral Monitoring Of Maize Leaf Blight Disease Based On Wavelet Transform

As one of the three major crops in the world,maize is also the main food crop and feed crop in China.Its planting area and total yield are second only to rice and wheat.In recent years,the occurrence of maize diseases has become more and more serious.As one of the main diseases harming maize production,maize leaf spot is an important disease in maize production in China.Traditional diagnostic methods mainly rely on manual observation on the ground and a large number of field surveys to determine the degree of maize leaf spot infection,which not only has the shortcomings of large workload and low efficiency,but also has a greater subjectivity.Hyperspectral remote sensing technology has the characteristics of macroscopical,accurate,non-destructive and fast,which provides a strong guarantee for timely monitor ing of crop diseases.In this study,the spectral reflectance of 105 samples of maize leaf blight was analyzed and its regular ity was studied.The wavelet energy coefficients of characteristic bands were extracted by wavelet transform,and the quantitative detection model of maize leaf blight area was constructed by partial least squares(PLSR).The results showed that:(1)The general variat ion law of spectral reflectance of different disease degrees is basically the same,and with the deepening of disease degree,the difference of spectral reflectance is more obvious.With the increase of spectral wavelength,the difference decreased first and then increased.After the sharp increase of spectral reflectance in near infrared region,the disease degree and spectral reflectance of maize leaf spot showed a significant positive correlat ion.As the most intuit ive expression of disease severity,the disease area could be sensitively responded to the change of the area of maize leaf spot.(2)Compared with the original spectrum,the wavelet transform method can significantly improve the correlation between the area of maize leaf spot and the spectrum.After five kinds of wavelet transform,the highest correlation coefficient is higher than the original spectrum,Bior 1.5 wavelet transform spectrum is the highest,the highest correlation coefficient is 0.92;Coif3 wavelet transform spectrum is only slight ly higher than the original spectrum of 0.78.The results showed that wavelet transform could enhance the correlation between the spectrum and the area of maize leaf spot to a certain extent.(3)Quantitative estimat ion model of maize leaf spot area was constructed based on the wavelet energy coefficients of characteristic bands extracted from five wavelet bases and partial least squares method.The regression analysis showed that the accuracy of the model was Bior1.5 wavelet basis(R~2=0.887),Haar wavelet basis(R~2=0.873),Sym2 wavelet basis(R~2=0.868),Mexh wavelet basis(R~2=0.859)and coif3 wavelet basis,respectively.(R~2 = 0.816),the model based on Bior 1.5 wavelet basis performed the best overall performance(R~2c = 0.887,RMSEc = 1.329,RPDc = 4.967;Rv~2v = 0.797,RMSv = 1.414,RPDv = 3.687).The results showed that the accuracy of the model basedon wavelet transfor m was higher than that of the original spectrum,and the area of maize leaf spot could be estimated effectively.