Rapid Determination Methods And Equipment Developed For Field Soil Properties Using Visible-near Infrared Spectroscopy

With the development of economy and technology,traditional extensive agriculture was failed to satisfy the needs of people for environment,resources,health and economic benefits.Currently precision agriculture becomes an important direction in modern agriculture for its aboundant advantages.For precision agriculture,it is necessary to adjust the input of various agricultural materials to obtain the largest economic and ecological benefits according to the condictions of soil and crop of each management unit.Soil is the basis of agricultural production,providing the necessary nutrition and environment for crop growth.Thus,rapid determination of soil fertility is the premise of precision agriculture.Traditionally,soil samples have to be brought back to the laboratory for chemical analysis.Although the measurement accuracy is high,it requires a lot of labor costs and time consuming and could not satisfy the needs for large area monitoring of soil properties.Recently,soil visible-near infrared?Vis-NIR?spectroscopy technology develops rapidly,which overcomes the shortcomings of traditional methods,such as high cost and labourious.Therefore,it is widely used in agricultural production.In this study,the research area was located in Xuchang,Henan Province.240 soil samples were collected for spectral measurement and chemical analysis,and then were used to explore the feasibility of different spectral pre-processing methods and modeling methods for the prediction of soil organic matter?SOM?,total nitrogen?TN?and p H.I also explore the methods of removing the moisture effects for field spectra.Based on the research above,a vehicle-mounted soil property rapid detection equipment was developed with core components including tablet PC,Ocean Optics USB2000+VIS-NIR-ES spectrometer,probe,GPS module,etc.then I designed and developed a control system and a field soil information management system that matching to the soil property rapid detection equipment.The main contents and results of this study include the following parts:?1?Comparing the effects of different pre-processing methods and two modeling methods for the estimation of three soil properties.The pre-processing methods included:Savitzky–Golay?SG?smoothing,Standard Normal Variate?SNV?and continuum removal;Partial least squares regression?PLSR?and machine learning algorithms Cubist were selected as modeling method in this study.The results showed that SG smoothing performed the best in the PLSR and Cubist model.The prediction accuracy for three soil properties using PLSR model was better than that of Cubist model.Based on the laboratory spectra,SG smoothing combined with PLSR model produced good predictions for SOM(R²=0.772,RMSE=1.38 g·kg^-1),TN(R²=0.732,RMSE=0.081 g·kg^-1)and p H?R²=0.722,RMSE=0.168?.Direct standardization?DS?method was used to transformed the filed spectra and then the spectra were used to build prediction model using PLSR method.The result showed that the model prediction accuracy was improved after DS transformation.The prediction accuracy for SOM was increased from R²=0.548 and RMSE=1.82 g·kg^-1 to R²=0.718,RMSE=1.37g·kg^-1;for TN,it was increased from R²=0.473,RMSE=0.103 g·kg^-1 to R²=0.651,RMSE=0.061 g·kg^-1 and from R²=0.5,RMSE=0.226 to R²=0.674,RMSE=0.177 for p H after the transformation.?2?Using Ocean Optics USB2000+VIS-NIR-ES modular fiber spectrometer as the sensor,reinforced tablet PC as the control system carrier,and Columbus V-800MARK?GPS receiver as the positioning device to develop a vehicle-mounted soil property rapid detection equipment and its control system.The equipment is connected to the agricultural vehicle through three-point suspension system,achieving the function of on-the-go soil property detection.The matching control system has a good human-computer interaction performance,and provides users with functions of spectrometer parameter setting,model optimization,spectra and prediction data display.When testing in the field,the device can achieve all functions as expected.And the detection results showed that the device had certain predictive capabilities for three soil properties with R² for SOM,TN,and p H prediction of 0.461,0.416,and 0.388,respectively.?3?Based on the B/S architecture,a set of field soil information management system was developed using Vue.js framework,My SQL database,Java Servlet,JDBC API and Arc GIS Server.The core function of the system is to receive the data transmitted from the soil property rapid detection equipment,display its real-time running status at the front end and provide spatial distribution information of soil properties.The rapid detection equipment transmits the detection results to the server through GPRS signals during the field operation.The signals are received by the server and parsed into data,which is stored in the database;the front-end obtains the data from the database and realizes the real-time display of the equipment operating status and detection results.The above research results can achieve the rapid determination of field soil properties to a certain extent,and provide real-time data support for agricultural production.Overall,this research has made a successful attempt in the development of soil detection equipment.However,the current prototype needs more improvement to satisfy the actual production needs.