Research On Early Rapid Diagnosis And Disease Monitoring Of Wheat Powdery Mildew

Food security has always been a major strategic issue related to social stability and economic development.Wheat is one of the most important food crops in China,and ensuring its stable and high yield is crucial to food security in China.Wheat powdery mildew is a fungal disease,which has the characteristics of latent infection and no obvious symptoms in the early stage.Once the disease occurs,it will spread rapidly and cause significant losses to wheat production.It has become an urgent need for disease prevention and resistance identification to achieve early rapid diagnosis and disease monitoring of wheat powdery mildew.As a fast,nondestructive and efficient detection method,hyperspectral imaging has the technical advantages of ’ map integration ’,and has been widely used in the diagnosis of crop pests and diseases in recent years.In this study,hyperspectral imaging technology and deep learning method were combined to carry out the early diagnosis of wheat powdery mildew,disease classification and disease spot growth prediction.A portable wheat powdery mildew diagnostic instrument was developed to provide new ideas,new methods and new ways for the efficient diagnosis of wheat powdery mildew.The main research contents are as follows:（1）Early diagnosis and visualization of wheat powdery mildew based on hyperspectral imaging.In order to realize the early rapid identification and visual monitoring of wheat powdery mildew,hyperspectral images of wheat leaves after different inoculation days were collected to obtain the hyperspectral temporal response characteristics of wheat powdery mildew.The classification scatter diagram,principal component characteristic image and load curve of samples at different infection stages were observed and analyzed.Furthermore,9texture variables of 12 characteristic bands and energy,entropy and contrast were extracted,and the early diagnosis models of wheat powdery mildew based on spectral,texture features and their fusion data were constructed respectively.The results showed that Partial Least Squares-Discrimination Analysis（PLS-DA）based on the fusion of spectral and texture features performed the best,and the recognition accuracy of wheat leaves at different infection stages was 91.4 %.In addition,the spectral angle mapper（SAM）method was used to identify the lesion at different infection stages,so as to realize the visual monitoring of the pathogenesis of wheat powdery mildew,and provide theoretical basis and technical support for disease control and resistance identification.（2）Wheat powdery mildew disease classification based on deep learning.The traditional disease classification mostly relies on plant protection experts,with low efficiency and high labor intensity.Using deep learning technology,the degree of stress on wheat leaves is measured by the proportion of powdery mildew spots.The powdery mildew is divided into four grades（health,mild,moderate and severe）.The YOLOv5 s network model is used to train the data set,which takes 11 h.The weight of the model is 13.7MB.When the loss value converges after 800 iterations,the m AP reaches 99.5 %.The identification speed of the test set was 83 FPS,and the accuracy,recall and m AP were 88.4 %,90.4 % and 88.8 %,respectively,confirming the potential of YOLOv5 s network model for powdery mildew leaf classification.（3）Wheat powdery mildew growth prediction model.The traditional identification of wheat resistance is cumbersome and subjective.Therefore,the characteristic parameters of lesion appearance are extracted by combining image processing technology,and the dynamic prediction model of powdery mildew is constructed to determine the growth stage of lesion and master the pathogenesis law of lesion.The single lesion image target was obtained by image segmentation of wheat powdery mildew leaves during the growth period,and then the lesion area was calculated.The minimum external rectangle was used to calculate the length,width and extension length of the lesion.Combined with the mathematical model,the Logistic optimal model of lesion area growth and the Exponential optimal model of lesion extension length growth were established.According to the growth model,the growth cycle of wheat lesion was divided into three periods,namely,the early growth period,the rapid growth period and the late growth period.（4）Development of portable wheat powdery mildew diagnostic apparatus.At present,the market has a large volume of disease diagnosis instruments and cannot realize in-situ realtime diagnosis.Therefore,based on the characteristic wavelength and disease classification model,a portable wheat powdery mildew diagnosis instrument is designed and developed.Based on the modular design idea,the core control module,light source amplification module,signal amplification module,display module,A / D conversion module and camera module are built,and the data interaction and collaborative control are realized by system program control.The instrument performance test results show that the disease recognition rate is 72.5 %,the classification accuracy rate is 85 %,and the recognition speed is 6 s / piece,which provides technical support for the development of more simple and practical wheat powdery mildew diagnosis equipment.