Early Detection Of Chilo Suppressalis Infestation In Rice Based On Hyperspectral Imaging Technology

In recent years,chilo suppressalis has become a major pest affecting rice production.Due to its latent nature it is difficult to determine the degree of rice infestation by chilo suppressalis,which leads to unreasonable or even massive use of chemicals.Therefore,in order to reduce this situation,it is necessary to diagnose the early infestation of rice by dicotyledonous borer.This paper proposes the use of hyperspectral imaging technology combined with image processing methods to distinguish normal and infested rice,and then combined with chemometric methods to establish qualitative and quantitative detection models for rice infected with different levels of chilo suppressalis larvae,to achieve the purpose of accurate application of rice infected with different levels of chilo suppressalis,the specific research content and conclusions are as follows.(1)The differentiation between normal and infected rice chilo suppressalis larvae was achieved based on hyperspectral images.First,the spectral analysis of normal rice stalks and rice stalks infected with chilo suppressalis larvae with smaller insect holes was performed,and a spectral region in the range of 500-1000 nm was determined.Then the first five images were obtained using two image data dimensionality reduction methods,principal component analysis(PCA)and minimum noise separation(MNF),respectively,and the images of PC1and MN3 were selected by their respective comparisons,and the band and weight coefficients were analyzed based on the feature information of the PC1 and MN3 images,and the single-band image at 750 nm wavelength was selected as the first feature image.For the method of mixing distance,the spectral region was divided into the visible spectral region of575-750 nm and the near-infrared spectral region of 750-1000 nm,and the band at 689.9 nm with the largest mixing distance was selected in the visible region,and the band at 778.4 nm with the largest mixing distance was selected in the near-infrared band,and by comparing the grayscale of these two bands and their combined band images,the single-band image at 778.4nm was selected as the second feature image.Finally,two algorithms,global threshold segmentation and iterative threshold segmentation,were used to process the two feature images,in which the best segmentation effect was achieved using the iterative threshold segmentation-based method for the single-band image at the 778.4 nm band,with a false positive count of 3 and an overall correct detection rate of 96.3%.The results show that the hyperspectral images can achieve rapid and precise differentiation between normal and chilo suppressalis larvae infested rice,providing a theoretical and methodological basis for online detection by hyperspectral imaging technology.(2)Based on hyperspectral imaging technology to achieve qualitative identification of the degree of infestation of rice by chilo suppressalis larvae.Firstly,80 rice stalk samples each of normal,lightly infected,moderately infected and heavily infected were obtained,and the spectral features and image features were extracted respectively,where the image features included color features(9 color momenta)and texture features(5 GLCM).Then the sample division KS algorithm was applied to divide into modeling and prediction sets in the ratio of3:1.19 and 20 feature wavelengths were selected from the full-spectrum data using SPA and CARS,and the full-band spectra,feature spectra,grayscale co-generation matrix and color moments were used as input variables for the PCA analysis model.Finally,the full spectrum,SPA,CARS,GLCM,color moments,image features(GLCM and color moments),SPA with image features and CARS with image features,which are used as inputs to PLS-DA and SVM,respectively,are optimally modeled based on the RBF kernel function of SVM combined with the fusion features of the wave filtering algorithm CARS and image features,and the total number of false positives is 1,the correct rate is 99.69%,RMSEC was 0.3098,R_cwas 0.9809,RMSEP was 0.3444,and R_pwas 0.9763.The results showed that the hyperspectral imaging system was able to detect the extent of rice infestation by chilo suppressalis larvae and could quickly locate rice infested with chilo suppressalis larvae at an early stage for early control.(3)A quantitative model study of the degree of chilo suppressalis larvae infestation in rice was realized based on hyperspectral imaging.The chlorophyll contents of normal,lightly infected,moderately infected and heavily infected rice stalk samples were firstly determined,and the distribution patterns of the four types of chlorophyll contents were analyzed by single-and multi-factor correlations between the chlorophyll contents of rice and the degree of infection of chilo suppressalis larvae.Then the SPXY algorithm was applied to get the modeling set with 240 samples and the prediction set with 80 samples,and the spectra were preprocessed using S-G smoothing,MSC,1st derivative,SNV,Baseline,etc.Finally,the full spectrum,color moments,and fusion features(full spectrum and color moments),which are used as the input of PLS and SVR models,respectively,and the optimal one is the model built based on PLS combined with fusion features,and the lower root mean square error is obtained when the number of principal factors is 9.At this time,RMSEC was 4.3387,RMSEP was4.9510,Rc was 0.8649,and Rp was 0.8546.The results show that hyperspectral imaging technology can be combined with chlorophyll content to quantitatively detect the degree of rice infestation by chilo suppressalis larvae and achieve the purpose of accurate application.