| Fusarium head blight(FHB)produces toxins in the process of infecting wheat,which can cause great harm to health of humans and animals,as well as affect the yield of wheat.The traditional methods identifying FHB mainly relied on the experience of experts and some equipment and instruments,which had disadvantages such as time-consuming,labor-intensive and inefficiency.Therefore,it was very significant to exploit an efficient method to identify FHB in the wheat kernels,which was of great significance for the development of intelligent agriculture in China.The existing research showed that the spectroscopy technique had the characteristics of rapid and nondestructive,and can be applied to the detection of diseases in crops.However,many disease detections only focused on the spectral information(the spectral range is in the visible near infrared)for research.In addition,most of the instruments of this type were expensive.Therefore,this paper attempts to explore a new method of identifying FHB,which using the spectral and image information.At the same time,it also analyzes the influence of different spectral ranges on FHB identification based on a non-imaging portable ground spectrometer.The optimal bands were selected,and wavelengths were combined based on the optimal wavelengths,the combination wavelengths were carried out to explore for identifying FHB.which lay the foundation to develop the cheap and practical multi-spectral cameras.The research progress of this paper is as follows:(1)The hyperspectral imager can provide spectral and image data,the hyperspectral images of wheat kernels were collected in the wavelength range of 400-1000nm.The spectral was extracted and preprocessed,and the optimal wavelengths were selected based on principal component analysis(PCA),successive projections algorithm(SPA)and random forest(RF).Then the spectral and image features are fused to obtain fusion features.The classification models of support vector machine(SVM),random forest(RF)and naive bayes(NB)were established based on the spectral features,image features and fusion features.The results show the best result was obtained by the SPA-RF model based on the fusion features,and the classification accuracy of the prediction set reaches 96.44%.The method proposed in this study takes full advantage of the information in the hyperspectral image,and provides a more effective way to identify the FHB in wheat grains.This method provided a more effective way to identify the degree of FHB in wheat kernel than using spectral or image information alone.(2)The data of the same batch of wheat kernels were collected by non-imaging portable spectrometer(spectrum range 350-2500nm).The 50nm wavelengths were removed in before and after to remove the noise,and the spectral range studied in most studies(Vis-NIR,400-1000nm;SWIR,1000-2450nm)and different spectral ranges(400-1000nm,970-1900nm,1900-2450nm)collected by different sensors inside the spectrometer were analyzed,respectively.After the pretreatment of the spectral,the optimal wavelengths were selected using competitive adaptive weighted sampling algorithm(CARS),PCA and SPA,and the different spectral ranges were analyzed based on the classification models of K-nearest neighbor classification algorithm(KNN),SVM and RF.The results showed that the modeling result in the spectral range of 1000-2450nm(Accuracy=94.56%)was better than the modeling result in the spectral range of 400-1000nm,and the result of the classification model based on the spectral range of 970-1900nm(Accuracy=97.32%)was better than the other two,which indicates that the classification model established based on the near-infrared range was more conducive to the identification of FHB in wheat kernels.At the same time,the selected optimal wavelengths were combined based on relationship between the spectral and the chemical composition of the wheat kernel,the results showed that the combinational optimal wavelengths in the spectral range of 1000-2450nm obtain a relatively ideal modeling result,and the classification accuracy was 98.14%.The combinational optimal wavelengths had the potential to identify FHB.This laid a good foundation for the development of multi-spectral cameras.(3)Based on the visible near-infrared spectral range of imaging hyperspectral spectrometer and non-imaging portable spectrometer,the performance identifying FHB was analyzed and compared.The results show that the accuracy of the model based on data collected by the imaging hyperspectral spectrometer data reaches 93.63% and that of the model based on data collected by the non-imaging portable spectrometer reaches 94.84% when only using spectral information.When the image information was added to the image spectrometer,the accuracy of the model based on fusion information reached 96.44%.It showed that the spectral and image information fusion model was better than using a single spectrum or image information,thus presumably not within the scope of the near infrared spectrum of imaging portable spectrometer using fusion technology can get better as a result.The research of this paper lays a good foundation for the development of multispectral camera.In summary,the method of fusion features and the combination of optimal wavelengths can provide technical support for identifying FHB,and it also has a great application prospect. |
PDF Full Text Request