Research On Detection Of Moisture Content Of Tomato Leaves Based On Dielectric Spectroscopy And IRIV-IFOA-SVR Algorithm

As one of the most important factors in the growth cycle of plant,water not only affects the growth and development of plant,but also plays a pivotal role in fruit yield and quality.The traditional methods for detecting plant moisture are observation method and drying method.The former method can only estimate the moisture content of the leaf coarsely,and the latter method has the disadvantages of strong destructiveness and large resource consumption.Therefore,the rapid and non-destructive detection of plant water is beneficial for the guidance of plant irrigation,which has very important practical application significance and theoretical research value.In this paper,dielectric spectroscopy was used to measure the dielectric parameters of leaf with different moisture content in electromagnetic field,and the intrinsic relationship with leaf moisture content was established.The main research contents and conclusions are as follows:(1)The principle of dielectric spectroscopy detection was explored in this study,which uses the relative dielectric constant?'and the dielectric loss factor?" as the characterization parameters of the dielectric spectroscopy.Plant leaves are considered to be dielectrics and represented by equivalent circuit of ideal capacitance and resistance,which are used to derive the calculation of?'and?".(2)In this paper,the effects of frequency,voltage and moisture content on the dielectric properties of the leaf were investigated and the optimum test voltage was determined to be 1.0V.At 1.0V,both?'and?"decrease with increasing frequency.When the frequency is less than 20 kHz,the downward trend is obvious.At the same frequency,the larger the water content,the larger the corresponding dielectric parameter.The three kind of parameters(?',?"and combined of?'and?")are used as input variables,and the moisture content is the output variable.(3)The 33 abnormal samples in the dielectric spectroscopy were removed by the Mahalanobis distance method,and the dielectric spectroscopy were preprocessed by Savitzky-Golay(SG),Standard Normal Variate Transformation(SNV)and Multivariate Scatter Correction(MSC).The result shows that the pretreatment has no significant improvement on the prediction effect.The sample set was reasonably divided into a calibration set(200)and a prediction set(67)by the Sample set Partitioning based on joint x-y distance(SPXY).In order to reduce the influence of redundant information on modeling,reduce the amount of model calculation and improve the accuracy of the model,Principal Component Analysis(PCA),Successive Projections Algorithm(SPA),Iteratively Retains Informative Variables(IRIV)and Variable Iterative Space Shrinkage Approach(VISSA)were adopted to select the characteristic variables of the dielectric spectroscopy.(4)In order to deeply analysis the intrinsic relationship between dielectric spectroscopy and moisture content,using Extreme Learning Machine(ELM),General Regression Neural Network(GRNN)and Support Vector Regression(SVR)to establish prediction model based on characteristic variables and full variables.The modeling results show that the SVR model based on the 17 characteristic variables of IRIV extraction combined parameter has the best prediction effect,the corresponding coefficient of determination for calibration set(R_C~2)is 0.9517,root mean square error for calibraton set(RMSEC)is 0.0286,the coefficient of determination for prediction set(R_P~2)is 0.8721,root mean square error for prediction set(RMSEP)is 0.0390,residual predictive deviation(RPD)is 1.8923.(5)The IRIV-SVR model have achieved better prediction result,but the prediction accuracy of the model can only provide the magnitude of moisture content,which is difficult to satisfy practical applications.Subsequent analysis shows that the SVR parameters(penalty factor c,kernel function parameter g and insensitive loss coefficient?)have greater impact on the prediction model.Therefore,Grey Wolf Optimizer(GWO)and the Fruit fly Optimization Algorithm(FOA)were introduced to intelligently optimize the parameters of SVR,and an Improved Fruit fly Optimization Algorithm(IFOA)was proposed for the defects of FOA.Finally,an optimized moisture content prediction model was established.The modeling results show that IRIV-IFOA-SVR model has the best prediction effect on the moisture conten with R_C~2=0.9873,RMSEC=0.0102,R_P~2=0.9720 and RMSEP=0.0186,and RPD=3.9677.In this paper,tomato leaf was taken as research objects,and the feasibility of detecting the moisture content of leaf by dielectric spectroscopy was explored.The moisture content prediction model established by the processing of the dielectric spectroscopy data has excellent prediction performance.Therefore,it is effective and feasible to detect the water content of tomato leaves by dielectric spectroscopy,which provides a reference for plant water detection and a new reference and method for dielectric spectroscopy data processing.