Research On Sclerotia Disease Of Rape Based On Machine Vision

Rapeseed is an important economic crop in my country.It is widely distributed,but its entire growth cycle is vulnerable to a variety of diseases,especially sclerotinia disease.Rapeseed sclerotinia disease is a fungal disease caused by sclerotinia sclerotiorum,commonly known as "white stalk disease";it can cause the quality of rapeseed to decline,the oil yield is low,and the economic income of agricultural growers is affected.Rapeseed sclerotinia disease can occur from germination to maturity of rape,and its leaves,stems,flowers,siliques and other parts may be infected,and the stems have the greatest impact on yield.Therefore,the ability to quickly identify the severity of sclerotium infection in rapeseed and spray pesticides for prevention and control in a timely manner is essential to inhibit the further expansion of the disease and increase the yield of rapeseed.The traditional method for the classification and detection of rape sclerotium disease mainly relies on the observation and recording of plant protection workers or experienced crop growers.This method is not only time-consuming and labor-intensive,but also vulnerable to weather,man-made factors and other factors,resulting in untimely results and affecting disease prevention.Secondly,based on the development of machine vision,domestic and foreign scholars have mainly studied rape leaves.Therefore,this article mainly studies the leaf and stem parts of rapeseed,and grades the severity of its infection with sclerotinia sclerotiorum,which can provide plant protection workers with decision-making basis for variable application of pesticides,and the user interface can be used to query the control effect after application..For the collection of rape images,manual inoculation and field shooting methods are mainly used;support vector machine(SVM)and VGGNet model algorithms are used to automatically identify the image data sets of healthy and diseased rapeseed leaves and stems;the main research contents are as follows :(1)Aiming at the disease classification of rape leaves,this study mainly adopts the method of artificial inoculation of hyphae,collecting healthy and intact rape leaves for inoculation,and observing the successfully inoculated leaves every 12 h,and using the camera to set the height at the same height.Shoot and record it under the conditions.Normalize the collected pictures uniformly to ensure that the size of the pictures is the same during the program operation,and then pre-process the leaves.The leaves are mainly classified by the proportion of the diseased area.There is no relevant standard for the classification of the disease degree of rape leaves.This article draws on the classification standard of wheat powdery mildew to classify it.Among them,the quality of lesion segmentation directly affects the accuracy of classification;this paper is based on threshold segmentation,watershed segmentation and feature extraction based on HSV color space model,repeated experiments,through comprehensive evaluation of segmentation effect,processing time,etc.,finally Select the method based on the HSV color space model to extract the diseased spots,and then use the contour drawing method to draw the outlines of the diseased spots and the entire leaf and calculate the pixel area ratio,and classify according to the classification standard,and then the leaf image data set The support vector machine(SVM)algorithm is selected for processing,and the operation result is compared with the result of manual measurement.The accuracy of this method can reach 94.25%,and the effect is relatively ideal.(2)For the classification of the disease degree of the rape stems,the source of the image data set contains two parts,one is the pictures of the rape stalks successfully inoculated with sclerotinia disease,and the other is the pictures of the stalks infected with sclerotinia disease in the field.The field environment is complicated.,The collected pictures need to use the Gaussian mixture model to obtain the target area-the stalk,and then use the HSV color space model to extract the lesions.By drawing the smallest rectangle circumscribed by the lesions,the longitudinal extension length of the lesions can be obtained.Refer to the Chinese People The grading results are obtained from the technical specifications for the forecasting and reporting of rape sclerotium disease according to the agricultural industry standard of the Republic,and the accuracy rate can reach 97.04% after verification.Datian needs to process a huge data set.In order to improve efficiency,this paper uses the VGGNet convolutional neural network,which contains 16 convolutional layers and fully connected layers.Aiming at the over-fitting problem of the VGGNet model,on the basis of increasing the data set,the regularization method and the Dropout method are mainly compared,and the optimal method is selected according to the running time and accuracy.Finally,the Dropout method is selected,and its accuracy can reach more than 90%.Finally,a field experiment was carried out.A total of 26 rape fields were collected,and 900 pictures were collected in each field.The field disease index obtained by the program operation was compared with the results of manual query,and the accuracy rate could reach more than 85%.(3)In order to make it easy for users to see the results of program processing intuitively and concisely,this article will design a user interface through the tkinter module that comes with Python.The interface mainly includes the processing process of rape leaves and stems,the classification results and the VGGNet processing process.Partial feature visualization pictures.