Research On Multi-source Information Fusion Detection Method For Facility Diseases And Insect Pests

Facility agriculture is an important supporting industry for my country's agriculture.The temperature inside the greenhouse is suitable,the relative humidity is large,and the leaves are relatively humid,which can easily cause outbreaks of diseases and insect pests.Because most greenhouse managers lack the scientific system of plant pathology knowledge,it is difficult to make accurate judgments on the severity of crop diseases and insect pests outbreaks.At the same time,the blind spraying of pesticides by managers will cause pesticide residues and pests to be effectively suppressed.As a result,crop production is reduced and environmental pollution is a problem.In order to predict the pests and diseases and quickly determine the degree of diseases in the greenhouse Pest logistic regression prediction model based on temperature and humidity change trends;spectral information analysis of cucumber leaf disease characteristics,and finally the use of visible light spectrum and terahertz spectrum fusion method for automatic classification of diseased leaves.The main findings are:(1)In the greenhouse,perlite potting was used to cultivate samples,and the greenhouse environmental information collection system was built.In addition to providing daily nutrients for the growth of greenhouse crops,the daily replacement of pest trapping hanging in the greenhouse and the statistical analysis of the species and number of pests on the day were carried out,and the disease status of greenhouse plants was analyzed and statistics and real-time temperature data were recorded.(2)Based on the collection of greenhouse environmental information and samples of plant diseases and insect pests,the daily changes of temperature,room temperature,humidity,and the number of insect pests in May were analyzed,and a pest warning model based on the trend of temperature and humidity changes was established.The occurrence of insect pests showed a trend of rising first and then decreasing with the month;except for July and August,the trend of the number of insect outbreak days predicted by the sensors was consistent with the actual statistical trend.In May,the changes of whitefly,leafminer,and aphids were consistent with the change of temperature and humidity,and the number of whiteflies was the largest.The results of the pest warning model based on the logistic regression prediction commonly used in machine learning show that the prediction accuracy of the three pest verification sets is above 87.5%,and the prediction accuracy of the test set is above 77.5%,and the prediction accuracy is high.It can be used for monitoring and early warning of greenhouse pests.(3)After classifying the collected diseased leaves,a visible light spectrum imaging scanning experiment was conducted.The stepwise regression method and PCA algorithm were used to extract the characteristic wavelengths of the obtained data,and the characteristic wavelengths of powdery mildew were 690.5 nm,555 nm,420.9 nm,614 nm,and 467.8 nm;the characteristic wavelength of cucumber downy mildew was 425.6 nm,700.8 nm,575.3 nm,635.1 nm,487.8 nm,and the first five principal components of cucumber downy mildew and powdery mildew.Based on the support vector machine algorithm,a visible light spectrum disease degree classification system is established.The test set results show that the feature model extracted by the stepwise regression method is better than the PCA.The classification accuracy rates are as high as 77.27% and 69.70%,respectively.The Sigmoid function is in 4 kinds Kernel function has the highest classification accuracy rate.(4)Based on the acquired terahertz data,the time-domain spectral response information of the diseased leaves was analyzed,and the feasibility analysis of the detection and classification of the terahertz system was carried out.Two algorithms,SPA and SCARS,were used to extract the terahertz sensitive bands,and the terahertz features were extracted based on principal component analysis.A disease SVM classification model based on terahertz information was established and tested and verified.The classification accuracy of the SPA algorithm reached 69.70%(powder mildew)and 71.43%(downy mildew);the results showed that: The recognition effect of the terahertz feature model is better than the SCARS method.(5)Based on the acquired hyperspectral and terahertz feature information of cucumber diseased leaves,feature layer fusion of multi-source information was carried out,and the multi-classification method of support vector machine was used to reasonably divide the training set and the test set for subsequent multi-core functions Classification experiment.The results show that the classification and identification of pests based on Sigmoid kernel function and linear kernel function has a high classification accuracy,and the classification detection accuracy of multi-source information fusion reaches 94.42%(white powder disease)and 93.88%(downy mildew).It can be used for automatic identification of the disease degree of cucumber downy mildew and powdery mildew,and the test classification accuracy of multisource information fusion is higher than that of a single information model.