Research On Grading For Diagnosis Of Wheat Scab Disease Based On Hyperspectral Imaging

Wheat is the high-yield and widely sowing crops in the world.Wheat scab is most important disease of wheat,which is easy to be attacked by bacteria in the growth process from seedlings to heading stage,and the most serious symptom is rot of the spike.In order to solve the problem of wheat scab resistance,we use the wheat as the research object,whichhad infected withFusarium graminearum,and the visible-near infrared(400 ~ 1000nm)of wheat scab as the data,to study the grading diagnosis model of wheat scab.The research lays the foundation for grading identification of wheat scab,especially the early diagnosis and warning.The primary coverage and conclusion of this paperare as follows:(1)The method of extracting the characteristic wavelength of wheat scabis proposed.We extract 0~420 main componentbased on principal component analysis(PCA1),and obtain 6 characteristic wavelength(548nm、645nm、683nm、739nm、812nm、871nm)by using the weight coefficient method and the difference methodto calculate the second derivative.The results showed that the selected bands were able to identify the grading of wheat scab.(2)Second principal component analysis(PCA2)method based on characteristic wavelength is proposed.According to the spectral image of wheat spike samples,the second principal component analysis is used to extract the region of interest(ROI).First of all,the first principal component analysis of hyperspectral images of wheat spikeis carried out,and select the best one to show the whole contour of wheat as the image of wheat spike contour.Secondly,make the second principal component analysis on characteristic bands and selectthe best one to show the whole contour of wheat spike PC3 as extraction of wheat spike disease image.(3)The grading diagnostic model of wheat scabis established.Take advantage ofmean filter method to realize the extraction of ROI of wheat spike,and calculatethe area aroundthe contour S1 of wheat spikeby the maximum connected region labeling method;Take advantage of local threshold segmentation methodto realize the extraction of ROI of wheat spikedisease area,and use the regional assignment method to calculate the area S2.The result of S2 / S1 is the grading percentage of wheat scab disease.Finally,by compared with the results of artificial grading of wheat spike disease,it is found that the higher degree of coincidence,the better the grading effect is.The result shows that the model can realize the segmentation of wheat scab disease area,which has better accuracy.And the feasibility of this method in grading identification of wheat scab has verified.(4)The automatic batch processing of spectral image data is realized.In order to solve the problem of wheat spike samples during the experiment,the number of wheat spike as much,and the hyperspectral data is large,and the traditional statistical method is inefficient when dealing with the data.Using IDL language to prepare batch processing program to read a large number of hyperspectral images cub file,at the same time,the images of wheat gibberellic disease were obtained by batching the characteristic bands of the secondary principal components analysis.