| As a loose material on the earth's surface,soil is an important natural resource and the basis for human survival.With the continuous development of mineral resources and rapid population growth,soil,a limited,non-renewable resource,is constantly being destroyed and reduced.In order to resolve the contradiction between human development and limited land resources,precision agriculture has become an important way to solve resource degradation and improve agricultural production efficiency.The timely acquisition of soil attribute information can provide effect evaluation and program selection for soil improvement and land remediation,which is the basic premise for the implementation of precision agriculture.Hyperspectral remote sensing can fully reflect the internal composition structure of the target object due to its unique high-resolution and extremely narrow band advantages,providing a convenient and non-destructive detection method for acquiring the target's attribute information,which makes up for the traditional indoor chemistry Insufficient detection.It is in line with the core issues of"fast" and "accurate" in precision agriculture.In order to provide satellite remote sensing sensors with spectral characteristics of ground objects and promote the application of hyperspectral remote sensing in precision agriculture.In this paper,the indicators of texture,organic matter and moisture content,which have a strong influence on crop and spectral characteristics,are defined as soil based indicators,taking into account factors such as scale(area size,sampling density)and the degree of human intervention(general farmland and reclaimed farmland).Explore the spectral absorption characteristics of soil basic indicators and sensitive areas in different dimensions,establish a quick estimation model of soil basic indicators,and introduce three optimization algorithms to explore the optimization space and improvement capabilities of the model.The main conclusions are as follows:(1)The overall trend of soil spectral curves under different scales and human intervention levels is relatively consistent,with obvious absorption peaks at 1400nm,1900nm and 2200nm.As the area scale becomes smaller and the sampling density increases,the similarity between the spectra of soil samples gradually increases According to the spectral analysis of soil basic indicators content classification,the spectral reflectance showed an upward trend as the soil particles became finer,while the organic matter and moisture content showed a downward trend as the content increased.The continuum removal method was used to extract the absorption characteristic parameters of soil basic indicators attributes,and the results showed that the use of multiple absorption characteristic parameters to explain soil basic indicators was better than a single absorption characteristic parameter.(2)In the one-dimensional spectrum level,the shape of the spectral curve after mathematical transformation had been changed accordingly,and the absorption peak valley and characteristic position of the original spectral curve were highlighted.Through correlation coefficient analysis,it was found that mathematical transformation combined with differential technology was more conducive to improving the spectral sensitivity and revealing the detailed characteristics of the soil.According to the modeling results,multivariate scattering correction first-order differential could eliminate the influence of soil surface scattering,so the interpretation effect of soil particle content was relatively good;the logarithmic first-order differential transformation amplified the internal details of the soil spectrum and enhanced the sensitivity of its characteristic bands,so the interpretation effect of soil organic matter and moisture content was relatively good.In general,the ability of predicting the properties of soil basic indicators based on the one-dimensional spectrum was average.(3)Compared with the one-dimensional spectrum level,the number of characteristic bands and the size of the sensitive area of the soil basic indicators in the two-dimensional spectrum level had been significantly improved.The correlation equipotential map of the spectral index of soil basic indicators constructed with or without mathematical transformation was quite different.The correlation equipotential map constructed by using the original spectral data was mainly distributed in block shape;while the correlation equipotential map constructed by mathematical transformation was mainly distributed in strips and had a clear hierarchy,which was more conducive for us to find sensitive feature combination bands.From the modeling results of the spectral index,the spectral index model constructed by mathematical transformation was better than the spectral index model constructed by the original spectrum,indicating that the appropriate mathematical transformation could enhance the response degree of the soil basic indicators in the spectral index.(4)Introducing the successive projections algorithm(SPA),the uninformative variable elimination(UVE)and the competitive adaptive reweighted sampling(CARS)to the better model of the soil basic indicators in the one-dimensional spectrum and the two-dimensional spectrum level.The number of characteristic wavelengths of soil basic indicators was greatly compressed,which simplified the complexity and calculation of the model.The accuracy of the optimized model had been improved to varying degrees,among which CARS had the most obvious improvement effect.Secondly,the models based on soil basic indicators at different scales and levels of human intervention in the two-dimensional spectrum level had shown excellent predictive capabilities after optimization.The above results indicate that more favorable information of soil basic indicators could be created in a wide two-dimensional space after the calculation of the spectral index,and the combination of the characteristic wavelength optimization algorithm could more effectively interpret the attribute information of the target.Figure[46]table[22]reference[109]... |
PDF Full Text Request