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

项目名称： 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

项目编号： No.31471409

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 农业科学

项目作者： 郑立华

作者单位： 中国农业大学

项目金额： 85万元

中文摘要： 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。

中文关键词： 土壤养分；近红外光谱；信息抽取；无损检测；农业信息化

英文摘要： 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.

英文关键词： Soil nutrition;NIR spectra;Information extraction;Nondestructive detection;Agricultural informatization