Research On Prediction Model Of Soil Moisture And Nutrients In Tea Garden Based On Hyperspectral Imaging

China is a major producer of tea,and the planting area of tea ranks first in the world.Currently,people’s attention to the development of the tea industry is mainly focused on the quality of tea,lacking systematic research on tea garden soil and tea tree growth.The key factor that determines the growth of tea plants and affects the yield of tea is the water content in the soil of the tea plantation.Tea,as a leaf crop,requires a large amount of nitrogen,phosphorus,and potassium during its growth.The levels of nitrogen,phosphorus,and potassium in the soil have a direct impact on the growth and quality of tea plants.Traditional soil sampling and analysis methods have shortcomings such as high cost,long cycle,and severe pollution,which cannot meet the requirements of timely monitoring and large-scale management of tea gardens.The emergence of agricultural hyperspectral technology provides a new technical approach for rapid and accurate monitoring of soil quality.This study takes the Jieshou Ecological Tea Garden in Changqing District,Jinan City,Shandong Province as the research area to explore the hyperspectral characteristics of tea garden soil and its relationship with water content and main nutrient content,extract characteristic bands,and construct the optimal model for predicting water content,alkali hydrolyzed nitrogen(AN),available phosphorus(AP),and available potassium(AK).The main research contents and conclusions are as follows:(1)The collection of soil samples,nutrient data,and hyperspectral data was carried out.Use the chessboard method to collect soil samples in tea plantation.Biochemical experiments have successively measured the water content,alkali hydrolyzable nitrogen,available phosphorus,and available potassium content of soil samples in the study area;Soil particle size analysis using laser diffraction method is divided into clay,silt,and sand particles;Use the SOC710-VP portable hyperspectral imager to collect hyperspectral images of soil samples.(2)The hyperspectral data of tea garden soils were pretreated.In the preprocessing stage,five preprocessing methods,Savitzky-Golay(SG),standard normal transformation(SNV),multivariate scattering correction(MSC),first derivative(FD),second derivative(SD),were compared.The results show that the smoothed water content spectral curve of SG is better than the original spectrum and other pretreatment methods.The effect of FD pretreatment on nutrient content spectra of sand and powder was the best.The effect of SG pretreatment on the nutrient content of clay was the best.(3)The characteristic bands of hyperspectral data of water content and nutrients in tea garden soils were determined.Since the spectral data predictors after pretreatment are still low,which indicates that there may be a serious problem of multicollinearity among the wavelength variables,it is necessary to extract the characteristic wavelength.In the phase of feature wavelength selection,correlation analysis(CA),principal component analysis(PCA)and continuous projection(SPA)were used to extract the original spectrum and the pre-processed feature bands.After comparison,the prediction model of soil moisture content in tea garden was established by using the characteristic wavelength extracted by SPA and the prediction model of soil nutrient content in tea garden was established by using the characteristic band extracted by PCA,which can effectively improve the accuracy of the model.(4)A prediction model for soil moisture and nutrients in tea gardens was established.The prediction models of partial least squares regression(PLSR),BP network(BPNN)and least squares support vector machine regression(LSSVM)are established.From the prediction results,it can be seen that the BP network after SG+SPA treatment is the best prediction model for soil moisture content in tea garden.The optimal prediction model for alkaline nitrogen hydrolysis in tea garden soils was BPNN model after FD+PCA treatment.The optimal prediction model of available phosphorus is LSSVM model after FD+PCA treatment.The optimal prediction model for available potassium is LSSVM model after SG+PCA treatment.PSO is used to optimize the model,and the accuracy of the optimized model is significantly improved.