土壤实时可见光/近红外光谱有效信息提取算法及其评价机理研究

How to extract effective information from real-time soil spectrum for detecting the specific soil target parameter is the bottleneck for both utilizing spectroscopy technology to solve the practical issue and spectroscopy diagnosis technology itself. This study uses alluvial soil and black soil in China as the research objects, and it is aimed at eliminating the effects caused by soil moisture and its particle size from soil real-time spectrum. Under the condition of no water interference, soil particle size influence eliminating algorithm is derived by analyzing, fusing and comparing the spectrum disassembled using Fourier transform and wavelet packet analysis with that obtained through multiplicative scatter correction, netanalytesignal analysis, etc.; After that, the spectrum is disassembled again according to different target parameters, and the effective spectral information extraction algorithm corresponding to each target parameter is obtained through the loop process of analysis, calculation and verification; Then the coupling relationship of moisture content and particle size effects on real-time soil spectrum is analyzed and their influence eliminating algorithm is developed; The same method is used to extract moisture response spectrum from real-time soil spectrum when analyzing the real-time soil samples with different moisture level, and also the moisture optimal projection matrix and its orthogonal projection matrix are obtained; Accordingly the process pattern can be established for each soil parameter to extract its response spectrum from real-time soil spectrum; And then, the spectral reaction mechanism will be explored for each target soil parameter via working on its corresponding characteristic spectra coming from two type soil to deliberate their common and different features; Finally the effective information ratio of the total information content for each target soil parameter is deduced in real-time soil spectrum, and the information entropy evaluation algorithm for each target can be designed and implemented.

为了解决土壤参数实时光谱无损检测中面临的受水分和粒度干扰显著以及各成分吸收峰严重重叠等问题，以北方潮土和东北黑土为对象，研究如何从土壤实时光谱中有效剔除粒度和水分影响，解析土壤水分和粒度的耦合关系，获取土壤参数的最佳投影变换矩阵或分解基函数，快速提取出反应目标土壤参数含量的有效光谱信息。在无水分干扰情况下，使用光谱变换和小波分析等技术分解光谱，对比并融合多元散射校正、净信号分析等方法，推导土壤粒度影响剔除算法；同理研究不同水分梯度的实时土样光谱，求得最佳投影光谱矩阵，获取水分响应光谱提取算法；解析实时土壤光谱中水分和粒度影响的耦合关系，求得水分和粒度耦合影响剔除算法；建立各土壤参数的响应光谱提取模式，实现土壤参数的实时光谱诊断；研究不同类型土壤中各参数所对应的特征光谱的共性和异性特征，探明各参数的光谱反应机理；分析实时光谱中对应各参数的有效信息与光谱信息总量的关系，设计算法评价信息熵值。

本项目旨在利用土壤实时光谱检测土壤目标参数含量（如土壤水分、土壤全氮、土壤有机质）。在实际生产中，对于较高土壤水分含量的土壤来说，土壤水分含量对土壤实时光谱影响巨大；而针对较为干燥的土壤，其土壤粒度对土壤实时光谱也有较为明显的影响，因此项目开展了土壤水分和土壤粒度对于土壤光谱影响及其剔除方法研究。. 项目设计了科学的数据采集实验，采集了北方潮土和东北黑土作为研究对象；项目通过大量的实验以及实验数据对比和分析、组合运用已有的光谱数据预处理技术，验证并总结了土壤水分和土壤粒度对于土壤实时光谱的影响及其剔除方法。研究发现，使用S-G卷积平滑法、一阶微分光谱法、水分吸收分类修正法、外部参数正交化法以及小波包滤波法对原始光谱进行预处理，均能获得一定的水分剔除效果。然而对于不同类型土壤的实时光谱而言，小波包滤波法具有更加稳定的处理效果。而对于土壤粒度，则使用一阶微分修正法处理后的混合测定集法建模，能够大幅提高模型的精度和稳定性，基本上可消除土壤粒度对土壤光谱的影响。. 项目详细分析了土壤目标参数对应有效光谱信息的评价方法，发现基于互信息可以有效地消除目标变量中的无关信息，寻找到光谱变量与土壤目标参数之间的最大信息，同时降低特征光谱变量数据冗余并最大化特征光谱变量的代表性，从而得到相应土壤目标参数的有效光谱信息。在此基础上，项目综合特征变量选择标准和蚁群算法的特点，提出了基于互信息理论优化的蚁群算法来进行土壤参数含量敏感波段的选择方法，明晰了土壤目标参数对应特征光谱波段的提取模式，并利用多种数据建模技术建立了较高精度土壤参数的实时测定模型并进行了模型验证。. 同时，为了固定核心研究成果，项目根据实际生产需要，研发了土壤全氮-水分含量原位测定仪并进行了性能实验验证，实现了土壤参数的实时检测；并设计开发了土壤光谱大数据平台，实现了土壤光谱数据的综合高效分析和建模，并为项目走向实际生产奠定了基础。

内容获取失败，请点击重试