Remote Sensing Monitoring Of Maize Leaf Spot Diseases Based On Multispectral And Thermal Infrared Imaging Data From UAV

Leaf spot disease is one of the common maize diseases,which damages photosynthesis by damaging the pigment structure of corn leaves,thereby reducing corn yield.Traditional disease monitoring methods are time-consuming and labor-intensive,therefore a fast and effective method is needed to monitor maize leaf spot disease,in order to improve maize yield and agricultural production safety.Remote sensing technology has the characteristics of wide monitoring range,fast monitoring speed,and no damage to crops,so it is widely used in crop disease monitoring.This study utilized two types of unmanned aerial vehicle remote sensing data,namely multispectral images and thermal infrared images,to identify,differentiate,and monitor maize leaf spot disease at the canopy scale.Three aspects of research were conducted: Firstly,identify healthy and diseased maize and construct a high-precision monitoring model for maize leaf spot disease;Afterwards,two maize leaf spot diseases with similar characteristics were distinguished-maize southern leaf blight disease and maize curvularia leaf spot disease,and a differentiation model for maize southern leaf blight disease and maize curvularia leaf spot disease at different stages was established;Finally,monitor the severity of maize southern leaf blight disease and maize curvularia leaf spot disease and their changes over time,and evaluate the disease resistance levels of different maize germplasm based on this.The main research results are as follows:(1)Analyze the canopy spectral response characteristics of healthy corn and diseased corn at four different stages after infection with pathogens and use drone multispectral and thermal infrared images and machine learning algorithms to monitor corn leaf spot disease.Analysis found that in the visible light region,the canopy reflectance of maize infected with leaf spot disease was higher than that of healthy maize;In the near-infrared region,the reflectance of infected corn is lower than that of healthy corn,and the difference gradually increases over time.Based on the Recursive Feature Elimination(RFE)algorithm,the most important remote sensing features for monitoring corn leaf spot disease were selected.It was found that the important spectral indices were mostly related to leaf pigments,while the features related to thermal infrared were of high importance in the four periods.The filtered features are input into random forest algorithm to establish a leaf spot monitoring model,which can successfully distinguish between susceptible maize and healthy maize,and the overall accuracy is more than 0.9 in the middle and late infection stages.(2)Taking maize southern leaf blight disease and maize curvularia leaf spot disease as examples,explore a research framework for distinguishing similar characterization diseases.Considering the pathogen infection patterns and physiological and biochemical characteristics of two diseases,as well as the spectral response of the corn canopy after their action on the corn canopy,remote sensing features were extracted using drone multispectral and thermal infrared images,and combined with feature selection and machine learning classifiers,these two diseases with similar characteristics were distinguished at different stages.The results show that the RFE algorithm can reduce feature redundancy,improve model operation speed and stability,and improve the accuracy of distinguishing maize southern leaf blight disease from curvularia leaf spot disease;After screening the features using the RFE algorithm,the Back Propagation Neural Network(BPNN)model was used to classify corn communities suffering from small spot disease and curvularia leaf spot disease.The overall accuracy in the early,middle,and late stages of infection was above 0.80,which can efficiently and non-destructive distinguish between maize southern leaf blight disease and maize curvularia leaf spot disease.(3)The Disease Index(DI)is used to indicate the severity of the disease,and unmanned aerial vehicle remote sensing data and regression analysis methods are used to evaluate the severity of the disease in maize with maize southern leaf blight disease or maize curvularia leaf spot disease,in order to evaluate the disease resistance of maize.Analysis shows that the optimal feature combination and algorithm for evaluating the severity of these two diseases are not consistent: the best feature for estimating the DI of small spot disease is EVI,the best algorithm is Support Vector Machine(SVM),R2 reaches 0.76,and RMSE is 0.07;The best feature for estimating DI of Curvularia leaf spot disease is NIR,and the best algorithm is the e Xtreme Gradient Boosting(XGBoost)XGBoost algorithm.The R2 reaches 0.82,and the RMSE is 0.07.This model verified the high resistance to maize southern leaf blight disease and high resistance to curved spore mold leaf spot disease characteristics of commonly used field varieties Nongda 108,and discovered inbred maize varieties BM01492,BM01501,and WX569 with high resistance to maize southern leaf blight disease.