基于自适应阶次的分数阶混沌系统对干旱区土壤盐分的反演研究

Soil salinization is a phenomenon of soil degradation, and its formation is a dynamic evolution process of complex dynamics. There is a complex nonlinear relationship between salt and spectral reflectance curve. However, there is currently no research on the nonlinear characteristics and nonlinear variation laws of the saline soil hyperspectral, which restricts the revelation of the control mechanism of saline soil characteristics and affects the accurate inversion of soil salinity. This project selects the saline soil in different regions of Xinjiang, it is based on the field measured hyperspectral data, soil physical and chemical parameters, and multi-source multi-temporal remote sensing image data, the fractional chaotic system is used to investigate the nonlinear variation laws of the saline soil hyperspectral. By (1) Analyzing differences of chaotic dynamic error and attractor between hyperspectral and salt at different fractional orders, the adaptive optimal fractional chaotic system is constructed to enhance the resolution between the peaks of hyperspectral, and reveal the nonlinear variation features of the saline soil hyperspectral. (2) An improved dragonfly algorithm based on chaos theory is used to explore the optimal wavelength variable that characterizes the hyperspectral properties of soil. (3) Establishing a neural network model that combines fractional-order chaotic systems and grey theory to achieve high precision quantitative inversion of soil salinity. This study provides a new idea for revealing the nonlinear variation characteristics and laws of the oasis saline soil in arid area, and provides a scientific and reliable basis for the improvement of saline soil.

土壤盐渍化是一种土壤退化现象，其形成是一个复杂动力学的动态演变过程，盐分与光谱反射率曲线之间呈现复杂的非线性关系。但目前缺乏对盐渍土高光谱的非线性特点及其非线性变化规律的研究，制约了对盐渍土特性控制机制的揭示，并影响了对土壤盐分的准确反演。本项目选择新疆不同区域的盐渍土，以野外实测高光谱、土壤理化参数、多源多时相的遥感影像数据为基础，采用分数阶混沌系统探讨盐渍土高光谱的非线性变化规律。拟从（1）分析高光谱和盐分在不同分数阶时的混沌动态误差和吸引子差异，构建自适应的最优分数阶混沌系统，增强高光谱的谱峰间分辨率，揭示盐渍土高光谱的非线性变化特点。（2）采用基于混沌理论的改进蜻蜓算法，探索表征土壤高光谱特性的最优波长变量。（3）建立融合分数阶混沌系统和灰色理论的神经网络模型，实现对土壤盐分的高精度定量反演。为揭示干旱区绿洲盐渍土的非线性变化特点及其规律提供新思路，为盐渍土改良提供科学可靠的依据。

土壤盐渍化的产生是非常复杂的动态演变过程，目前缺乏对盐渍土高光谱的非线性特点及其非线性变化规律的研究。以混沌系统为理论基础的非线性科学，能对盐渍土高光谱这类具有非线性特性的问题进行探究与分析。本项目按研究计划取得的主要成果包括：（1）以不同盐渍化程度的土壤为数据源，建立分数阶Lorenz、Chen-Lee、Sprott混沌系统数学模型，增强高光谱的谱峰间分辨率。观测混沌动态误差，分析其运动轨迹、相平面、吸引子的动力学行为特征，探讨采样点在不同分数阶的混沌动态误差和吸引子差异，找寻混沌分布的变化趋势，并结合概率神经网络模型、灰色系统理论、蔡氏电路、可拓物元模型，有效辨识出输入信号的差异特征。（2）采用Grünwald-Letnikov分数阶微分处理土壤高光谱数据，挖掘隐藏在分数阶处的有用信息，避免传统整数阶微分造成的信息丢失，揭示出盐渍土野外高光谱的非线性特点及变化规律。（3）建立基于分数阶微分的显著性水平波段偏最小二乘模型（FOD-SLB-PLSR），不论何种干扰程度的土壤，总盐与主要水溶性盐基离子的最优模型都出现在高阶分数阶，其预测性能优于基于全波段和整数阶微分建立的模型。（4）基于不同方法提取高光谱的特征波段，应用PLSR和3种机器学习（GRNN、ELM、RF）建模，比较线性和非线性模型预测总盐的精度，估算效果最好的是GRNN，三个区域的最佳模型均为R-FOD-CC1-CARS-GRNN，RPD值都大于1.8，且轻、中、重度盐渍化区域提取的特征波段分别为8、9和11个，占全波段的0.45%、0.51%和0.63%%。在项目资助下，负责人共发表论文15篇（一作13篇，通讯2篇），其中在“IEEE Transactions on Geoscience and Remote Sensing、Remote Sensing、Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy”等国际著名期刊上发表SCI论文11篇（二区及以上8篇），3篇CSCD，1篇北大核心；申请软件著作权2项；负责人晋升为教授，获云南省中青年学术和技术带头人后备人才。本研究为土壤盐分高光谱反演模型精度提高提供了新思路，这对绿洲盐渍土的改良有一定的指导意义。