Disease Prediction And Identification Of Panax Notoginseng Disease In High Incidence Period Based On Machine Learning

Panax notoginseng grows in humid,warm,and shaded environments.This type of environment is prone to induce various diseases.Its incidence has a decisive effect on the yield and quality of Panax notoginseng.As the amount of implants increases,the types of diseases And the area is increasing year by year.At present,in the direction of forecasting the incidence of notoginseng,the research still stays at qualitatively describing the relationship between the incidence of notoginseng and meteorological factors;in the direction of identifying the disease on the leaves of notoginseng,it still relies on artificial subjective recognition or the use of shallow models for detection,but these methods There are problems such as slow prediction and recognition,low accuracy,poor generalization and robustness,and it is difficult to efficiently grasp the law of the incidence of Panax notoginseng,so that it is impossible to effectively warn and prevent the occurrence of diseases in a timely manner.To this end,this subject has carried out a research on disease prediction and identification in the high-incidence period of Panax notoginseng disease based on machine learning.The main contents of the research are as follows:(1)Take the annual May-September meteorological data set and disease data as the training set and the verification set.The meteorological data includes 10 indicators such as soil temperature,temperature in the shed,and soil heat flux,each of which has 7 200 indicators.Samples,and use field statistics of the incidence of Panax notoginseng as the incidence data.The random forest algorithm is used as the basic learning machine,and the mean square error is used as the node splitting standard.Through node splitting and recursive execution to select the optimal branch operation,the best random forest notoginseng incidence preliminary prediction model is selected from 100 sub-models.Then the gradient descent algorithm is used to optimize the model to reduce its variance.(2)Collect images of Panax notoginseng leaf disease under various environmental conditions such as rainfall,cloudy,sunny,and night,and expand the sample size to 32,400through data enhancement,and use this as a training set and a test set.Disease types include images of Panax notoginseng leaf disease including blight,yellow rust,anthracnose,powdery mildew,round spot,and virus disease.Input models with different weights of s,m,l,x for training,and the target detection model with the best performance is selected through indicators such as the mean Average Precision of the area under the Precision-Recall curve(m AP),the Generalized Intersection over Union(GIo U),and the prediction accuracy.The results show that:(1)The Pearson correlation coefficient between soil temperature and humidity in the shed is between 0.25 and 0.75 through the analysis of main effects.It is negatively correlated,and its Pearson correlation coefficient is between-0.75 and-0.25;through gradient descent optimization,the relationship between meteorological factors and disease incidence is analyzed qualitatively and quantitatively.Among them,the degree of positive correlation between soil temperature is the largest,and the weight is 21.686.The soil heat flux above the seven canopies has the greatest degree of negative correlation,with a weight of-13.834;Introduce the gradient descent algorithm to optimize the basic learning machine model and improve the prediction ability and stability of the panax notoginseng incidence prediction model,compared with the actual disease incidence of Panax notoginseng,the predicted disease incidence rate after model optimization is 3.79% higher than that before optimization,and the predicted incidence rate change trend is consistent with the real situation.The research results have reliable predictive ability in predicting the incidence of Panax notoginseng disease during the high incidence period through field meteorological factors,and can provide theoretical basis and technical support for reducing the cost of facility environmental control and intelligent management of Panax notoginseng disease.(2)After 5,000 epochs,the results show that the m model performs best in each performance evaluation index.This model can better learn the characteristics of different types of Panax notoginseng leaf diseases,and the disease detection frame can maximize the approximation of the Panax notoginseng leaf disease Department,through the establishment of characteristic equations,the recognition accuracy rate can reach more than 95%,and realize more accurate and faster recognition of different types of Panax notoginseng diseases;The change of each index per 1,000 epochs shows that the more complex structure of the model extracts more information,but the m-weight model is the best for effective information mining and learning ability;The accuracy of disease recognition of the same Panax notoginseng leaf image under different external conditions shows that the m model has the highest recognition accuracy,and has better generalization ability and robustness under complex external environmental conditions.