Research On Hyperspectral Detected Method Of Soil Available Selenium

Selenium(Se)is an essential micronutrient element for crop growth and development.Available selenium in soil plays an indispensable role in crop metabolism,growth and development,quality improvement and yield improvement.Rapid,accurate and efficient detection of soil available selenium is an important basis for mastering soil selenium content,applying selenium fertilizer and producing selenium-rich agricultural products.Up to now,there are few literatures about measuring soil available selenium at home and abroad,and the measurement method has low efficiency and high cost,which is not universal.Aiming at the problem of the determination method of effective selenium in soil,taking the selenium-rich soil in Shitai,Anhui Province as the research object and the effective selenium in the soil as the detection object,this paper proposed the method of using non-imaging hyperspectral technology to determine the effective selenium in soil.First compares the different single pretreatment method for the construction of the influence of soil selenium prediction model is effective,and compares the two different wavelengths of the modeling results,on this basis,select the best way of pretreatment and the most suitable for modeling of spectral bands,the contrast analysis of two different influence to the precision of the model,modeling algorithm based on the study of soil available selenium accurate detection.The main research and conclusions of this paper are as follows:(1)The influence of different pretreatment methods on the prediction model of soil available selenium was studied and the best pretreatment method was selected.Different pretreatment methods will affect the subsequent modeling result in different degree,SG smooth selection of commonly used in this study,a derivative and second derivative,mean,standard normal centralized transformation,logarithmic transformation,correction to trend seven single pretreatment method,including the after drying,grinding and sieving raw soil spectral data of spectrum pretreatment,and comparing pretreatment method,according to the Ridge and RBF-Ridge model combined with the evaluation standard of choose the best pretreatment method.The results showed that the accuracy of the original spectral data was improved after SG pretreatment,and the Ridge model after SG pretreatment had the best accuracy.Its training sets R2,RPD and MAPE were 0.70,1.81 and 3.45 respectively,and the test sets R2,RPD and MAPE were 0.70,1.84 and 1.24 respectively.This model can be used for quantitative estimation of soil effective selenium.(2)The influence of soil spectra based on different bands on modeling accuracy was studied and the optimal spectral bands were determined.The available bands of the original soil spectrum measured by the ground-object spectrometer are generally between 400-1659nm.Due to the internal structure of the instrument,there is an obvious wobble of the soil spectrum around 900nm.In this experiment,spectral bands of 500-900nm,900-1659nm and 500-1659nm are analyzed and compared respectively to determine the most suitable spectral bands for modeling.The results show that when Ridge and RBF-Ridge algorithms are used for modeling,it is feasible to establish regression models for the regional bands of 500-900nm,900-1659nm and 500-1659nm.Under the same pretreatment method,the overall modeling results of Ridge and RBF-Ridge in the 500-1659nm band are better than that of the 500-900nm band,and the overall modeling results of the 500-900nm band are better than that of the 900-1659nm band.There was no significant difference in RPD values between the two models at bands of 500-900nm,900-1659nm and 500-1659nm.Compared with the three bands,Ridge and RBF-Ridge regression based on 500-1659nm provided the best modeling results for the original spectra.(3)The influence of different modeling algorithms on prediction accuracy is studied and the best prediction model is selected.The advantages and disadvantages of the model depend on the selection of the modeling algorithm.In this experiment,two algorithms,Ridge Regression and Partial least Squares regression,were used for modeling analysis and comparison.On the basis of the above study,500-1659nm bands were selected for analysis and abnormal sample values were removed.SNV+SG,DT+SG,SNV+DT,SG+SNV+DT,LG,LG+SNV,LG+DT,LG+SNV+DT,SG+LG-SNV,SG+LG+DT and SG+DT combination of ten kinds of pretreatment method to deal with the original spectrum,and use the Ridge and PLSR model respectively,the result showed that Ridge to the modeling effect of the original spectrum is better than PLSR,two kinds of modeling method in 500-1659nm wavelength of soil selenium fitting between measured and predicted effect no too big difference,This indicates that the 500-1659nm band modeling can effectively alleviate the interference brought by external factors and human factors,and increase the stability of the soil available selenium prediction model.(4)Design and develop the prototype system of soil available selenium hyperspectral detection.On the basis of the above research,a prototype system of soil effective selenium hyperspectral detection based on the visual interface was developed,including four modules of data import,physical and chemical data analysis,spectral data pretreatment and modeling analysis.Through the design of the system,experimental results could be more intuitively displayed.According to the above research,the rapid detection of soil available selenium is basically realized,and the theoretical foundation and technical support are laid for the field in situ detection of soil available selenium.