The Hyperspectral Prediction Model Of Ramie Leaf Moisture Content Based On SVR

At present,hyperspectral technology is widely used in agriculture.Spectrometer detection of crop leaves is used to study the spectral characteristics of crops.Not only can biochemical indicators such as chlorophyll content and moisture content be retrieved,but also the growth status of crops can be predicted.Decision-making information is of great significance to the development of sustainable modern agriculture.As a traditional natural fiber crop in my country,ramie’s economic value cannot be underestimated.In domestic and foreign studies,reports on the hyperspectral characteristics of ramie are relatively rare,and the moisture content of ramie has a great influence on the physiological and biochemical properties of crops,and the moisture content is related to the hyperspectral reflectance.The moisture content model enables accurate and real-time monitoring of the target,which can better reveal the physiological characteristics of plants and provide technical support for fine agriculture.In this study,we will take ramie as the research object,and use portable ground spectrometer and supporting blade holder as instruments to study the relationship between ramie blade hyperspectral reflection characteristics and changes in moisture content;using a variety of noise reduction methods,Such as convolution smoothing（Savitzky-Golay,SG）,standard normal variate（SNV）pre-processing,to reduce the influence of surface scattering characteristics on the spectrum,avoid interference of various factors,and make the data more accurate;Use a variety of dimensionality reduction methods,such as feature selection: correlation analysis and vegetation index,feature extraction: principal component analysis to extract ramie leaf hyperspectral characteristic parameters;finally,establish a moisture content prediction model based on ramie leaf hyperspectral characteristic parameters,compare several The support vector machine model based on the kernel function is compared to get the best solution.The research results of the full text are as follows:（1）Statistical analysis of the original hyperspectral reflectance and moisture content of ramie leaves.Observe the characteristics and differences of hyperspectral reflectance in the upper,middle and lower leaf positions of ramie prosperity in the long-term and mature period,and the hyperspectral reflectance curve has a similar trend.The moisture content of the ramie leaves of the variegated and undivided varieties in the prosperous and mature periods was counted.In any case,the moisture content of the lower leaf position is often higher than that of the upper and middle leaf positions.The moisture content is the lowest.This is because the lower leaf position closer to the root is easier to absorb water,while the upper leaf position is easier to evaporate water.（2）Respectively analyze the correlation between the moisture content of the ramie leaves and the spectral data reflectance in the vigorous and mature periods.The maximum correlation in the vigorous period is at 690 nm,r=0.8111;the maximum correlation in the mature period is at 696 nm,r= 0.5664.Comprehensive analysis shows that the regions with the most obvious correlation are concentrated in the 400-727 nm,837-1138 nm,and 2274-2289 nm bands.These three bands with higher correlation can be selected as the construction of subsequent spectral parameters.（3）Perform SG smoothing and SNV data pre-processing on the original hyperspectral data in the prosperous and mature periods,and on this basis,use principal component analysis（PCA）and vegetation index（VI）The method extracts characteristic parameters and performs dimensionality reduction on the data.When PCA reduces dimensionality,the selection criterion for the number of main components is usually based on the eigenvalue greater than 1 and the interpretation rate of the original variable >80%.In the dimensionality reduction results of PCA based on SG data,the main factors extracted during the prosperous period are the first 6 and the first 8 in the maturity period;in the dimensionality reduction results of PCA based on SNV data,the main factors extracted in the prosperous period are the top 13,The maturity period is the top 15.The principal component factors extracted by the two methods are used as the input of the model.In the result of extracting the characteristic values of the vegetation index,the correlation analysis of the vegetation index value and the moisture value was carried out to obtain seven vegetation indexes with good correlation,which were also used as the characteristic parameter values of the subsequent model construction.（4）Establish support vector regression（SVR）prediction model based on different kernel functions,compare and analyze the modeling effects of PCA main factors and vegetation index of SG and SNV data during the prosperous and mature periods,so as to select the best model.According to the experimental results,the RBF-based SNV data method is the best modeler.According to the model evaluation criteria,the coefficient of determination（R2）in the long-term period reaches0.78068,the root mean square error（RMSE）and the relative error(The relative error（RE）was 0.02211 and 0.20% respectively;the determination coefficient（R2）at maturity reached 0.59040,and the root mean square error（RMSE）and relative difference（RE）were 0.02868 and 0.71%,respectively.