Hyperspectral Image Feature Fusion For Classification And Recognition Of Potato Early Blight And Late Blight

Early blight and late blight of potato are the highest and most threatening diseases in potato planting.The control of disease plays a decisive role in potato yield.However,early blight and late blight are relatively similar and difficult to distinguish,and traditional disease diagnosis methods are easily affected by many factors,causing misjudgment and low diagnosis efficiency.Hyperspectral imaging technology can continuously and dynamically monitor crop conditions in a timely and accurate manner.It also has the characteristics of high resolution,multi-band,narrow band,wide spectral range,and unification of maps.It is extremely valuable for research on crop diseases.In this study,the early and late blights of potato leaves were detected and classified based on hyperspectral imaging technology.The spectral features and image features were combined to make up for the shortcomings of using spectral features and image features alone to accurately identify potato diseases.First,hyperspectral images of potato early blight,late blight and healthy leaves in the 366.66-976.41 nm band were collected using a hyperspectral imaging system.Because hyperspectral images have high-dimensional data,which contains a lot of redundant information,the correlation between different bands is large,which will have a serious impact on the accuracy of data processing.The characteristic wavelength selection can select the most useful information in the full-band spectrum,simplify the model and improve the efficiency of later modeling.Therefore,the paper uses principal component analysis and second derivative method to extract the characteristic band length of the early blight leaf from the spectral dimension of 574.5nm,675.20 nm and 724.50 nm,the characteristic wavelengths of late blight leaves were 572.29 nm,675.11 nm and 780.33 nm.Then,color features are extracted through color moments,texture features are extracted through the gray level co-occurrence matrix,and the image background is removed using a mask and feature fusion is performed,that is,texture features are extracted based on the characteristic wavelength,and color features are extracted based on the mask.Both color features and texture features describe the surface properties of the image.After merging with the spectral features,the internal and external properties of the image are combined.Finally,a BP neural network and SVM model are established to classify and identify the hyperspectral image,hyperspectral image color feature,hyperspectral image texture feature,feature image,feature wavelength and texture feature fusion,and mask and color feature fusion respectively.The results show that early detection of potato early blight and late blight by using hyperspectral imaging technology combined with BP neural network is a feasible and effective method.Compared with the original hyperspectral data,the model recognition rate established by the feature fusion data is higher.The recognition rate of the BP neural network based on the feature image and the texture features under the feature image reached 100%,and the recognition rate of the fusion data based on the mask and color features also reached 97.5%.The classification and recognition effect is good,which proves that Data fusion has certain advantages in the application of crop disease identification based on spectral imaging technology.