Classification Of Microscopic Images Of Rice Blast Spores And Sheath Blight Pathogen

Rice has been the main food crop in China since ancient times,and now many provinces in China have planted different varieties of rice,in recent years due to the invasion of diseases caused by a large-scale reduction in rice production,which has produced a certain threat to China’s food security,so it is of great significance to identify rice diseases in time for prevention and control.At present,the identification of rice blast spores and striated blight bacteria is still observed with the naked eye under the microscope,which wastes a lot of manpower,material resources and time,and is subjective.Aiming at the above problems,this study combined image processing,machine learning and deep learning technology to complete the classification of spore microscopic images of rice blast and micro images of blight bacteria,respectively,and mainly carried out the following studies:(1)Microscopic image pre-treatment of rice blast spores and blight bacteria.Preprocessing is achieved in combination with image grayscale,image enhancement,image denoising,and image segmentation.Image enhancement uses gamma transform,histogram equalization,Laplace transform and restrictive contrast adaptive histogram equalization(CLAHE)for comparison experiments,and selects the average gradient and standard deviation as the evaluation index,and selects the CLAHE algorithm as the image enhancement algorithm according to the evaluation index and enhancement effect.Image denoising uses median filtering,mean filtering,Gaussian filtering and bilateral filtering for comparative experiments,and the peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)are selected as evaluation indicators,and the median filtering algorithm is selected as the image denoising algorithm according to the evaluation index.Image segmentation was conducted using Otsu algorithm and adaptive threshold segmentation algorithm for comparative experiments,and Otsu algorithm was selected as the image segmentation algorithm based on the combination of segmentation effect and average cross-union ratio(MIOU).(2)Extraction of micrographic image characteristics of rice blast spore and tiarami.The shape characteristics and texture characteristics are extracted from the rice blast spore microscopy,the area,perimeter,and external elliptical degree of rice blast spores,the minimum external rectangle length,and the shape characteristics,the grayscale matrix,partially binary value The features extracted by the pattern algorithm are measured as a texture feature,a total of20 feature values.The texture characteristics were extracted to the microscopic image of the tiaramidal bacteria,and the characteristics extracted by the grayscale symbiotic matrix and the localized moduct mode algorithm were taken as a texture characteristic,a total of 16 feature values.The extracted feature value is standardized using the Z-Score method.(3)Microscopic image classification of rice blast spores and striated blight bacteria.Three classification algorithms were used to model the microscopic images of rice blast spores and blight bacteria,and the classification accuracy was compared.The classification accuracy using the cosine similarity algorithm was 85.4% and 89.3%,respectively.The highest accuracy rate of classification using the support vector machine algorithm is 98.7% and 98.5%,respectively.The highest accuracy rate of classification using the transfer learning Alex Net algorithm is 100% and99.3%,respectively,which is higher than that of the other two classification methods.(4)Design and implementation of microscopic image classification system for rice blast spores and blight bacteria.Python’s Flask framework is used to establish a classification system,and the rice blast spore microscopic image classification model and the blight fungus microscopic image classification model trained by Alex Net algorithm are used as the classification model of the system,and the system mainly includes the registration module,login module,image upload module and image classification module.The above research results provide technical support and theoretical basis for the classification of microscopic images of rice blast spores and blight species,and also provide new means and methods for studying the classification of microscopic images of spores of rice blast and blight molds.