Recognition Of Potato Late Blight By Space Spectrum

Fine agriculture is the goal and direction of modern agriculture development.In the aspect of disease detection,precision agriculture requires fast and accurate data information to provide accurate and reliable support for disease prevention and control,so as to achieve fine management in the process of crop growth.Potato is an economic crop widely cultivated in the world.The wide spread of late blight will lead to its low quality and yield,and even lead to the occurrence of vicious famine.Real-time on-line detection of diseases can effectively reduce losses,and traditional methods have high misjudgment rate and low efficiency.Searching for a fast and accurate identification method is the focus of people's attention.The emergence of hyperspectral imaging technology provides a new diagnostic method for disease detection.The characteristic of the technology of "atlas integration" solves the problems of traditional spectral band number,low resolution and Atlas separation.In view of this,this paper based on hyperspectral imaging technology for potato late blight feature band selection,classification and threshold segmentation.Firstly,256 bands of hyperspectral images of potato late blight leaves were collected by hyperspectral imaging system in the visible-near infrared band.Because hyperspectral data processing has some shortcomings,such as large data redundancy,high correlation between bands,and has the characteristics of "atlas in one",it is necessary to extract the characteristic bands of disease from two aspects of space spectrum.In the aspect of spectral information,the second derivative of the spectrum curve after denoising is obtained,and then the principal component analysis is used to optimize the compression.The final second derivative combined with principal component analysis extracts the characteristic bands of the diseased leaves: 672.73 nm,691.86 nm and 710.99 nm;in the aspect of spatial information,the characteristic bands of the diseased leaves are 624.91 nm,663.16 nm and 684.69 nm.Then,decision tree,K-nearest neighbor classification algorithm and BP artificial neural network model were established to classify and recognize potato leaves in different disease stages based on the reflectivity of characteristic band and the gray value of principal component image.The experimental results show that the recognition rate of all the models is over 80%.Compared with the model based on reflectivity,the model based on gray value of principal component image has better recognition effect.Secondary principal components combined with BP artificial neural network had the highest recognition rate for the early,middle and late stages of the disease,which were 94%,97.2% and 98.1% respectively.Finally,considering the accurate classification and recognition of the severity of disease in the field,this paper uses band ratio image and principal component image to segment the disease area.The experimental results show that the threshold segmentation results based on principal component image and band image are better than those based on image.The threshold segmentation algorithm based on principal component image has a slightly worse effect on edge segmentation,while the threshold segmentation algorithm based on band image has a slightly worse effect on petiole segmentation.The experimental results show that hyperspectral imaging technology can be applied to the diagnosis of potato late blight disease,which is a fast and non-destructive diagnostic technology.The screening of characteristic bands and the establishment of models for hyperspectral data from two aspects of spatial spectrum can provide an effective way to find the best diagnostic means for diseases and provide ideas and methods for disease detection.Threshold segmentation of disease area can be applied to the classification and prevention of fine agriculture.