基于RGB和纹理特征的土壤表层水盐信息反演方法  

A Method for Estimating Water-Salinity Information of Soil Surface Using RGB and Texture Features

作　　者：宋岩 柴明堂 李王成[1] 孙利英 吾连恩·赛尔奴 杜天择 SONG Yan;CHAI MingTang;LI WangCheng;SUN LiYing;Wulianen Saiernu;DU TianZe(School of Civil and Hydraulic Engineering,Ningxia University,Yinchuan 750021;Ningxia Key Laboratory of the Internet of Water and Digital Water Governance of the Yellow River,Yinchuan 750021)

机构地区：[1]宁夏大学土木与水利工程学院,银川750021 [2]宁夏回族自治区黄河水联网数字治水重点实验室,银川750021

出　　处：《中国农业科学》2025年第6期1159-1172,共14页Scientia Agricultura Sinica

基　　金：国家重点研发计划(2021YFD1900600);宁夏自然科学基金项目(2023AAC05014);宁夏重点研发项目(引才专项)(2023BSB03021)。

摘　　要：【目的】盐渍土是全球性的生态环境问题。快速精准的监测表层土壤水分、盐分信息为土壤盐渍化的治理和改良提供支持。【方法】本研究在小尺度上,将高精度光学遥感和图像数字处理技术相结合,提出一种基于土壤表观色彩参数(RGB)和纹理特征值预测土壤含水量、含盐量的定量估算方法。首先利用介电常数和含水量、电导率和含盐量之间的关系标定了土壤多参数传感器。其次利用图像数字处理技术,提取土壤的3种色彩参数(RGB)和纹理特征,并通过相关性分析等方法分别确定相关性最大的变量,并构建RGB、纹理特征同土壤含水量、含盐量的最优拟合模型。最后利用传感器法验证反演方法的精准度。【结果】土壤含水量与RGB拟合的三元回归模型拟合效果最好,R^(2)为0.97。土壤含盐量与RGB和纹理特征的拟合中,当含盐量≥6%时,含盐量与土壤的表观白色占比拟合的一元多项式模型拟合效果最好,R^(2)为0.97;当含盐量<6%时,含盐量与纹理特征值中的自相关拟合效果最好,R^(2)为0.93。经过对比计算标定后的多参数传感器法和本文提出的反演方法得到的土壤含水量和含盐量,发现两种方法测量的土壤含水量相对误差范围为0.27%—9.48%,土壤含盐量相对误差范围为0.07%—8.64%,且绝对误差均<1%。【结论】本研究为土壤表观水盐信息的反演提供了一种方案,为土壤表层的水盐的快速、准确的测定提供了理论依据和技术支撑。【Objective】Saline soil poses a global ecological challenge.The rapid and precise monitoring of surface soil water and salt information was crucial for effective control and remediation of soil salinization.【Method】The present study proposed a quantitative estimation method,which combined high-precision optical remote sensing and image digital processing technology at a small scale,to predict soil water content and salt content based on soil apparent color parameters(RGB)and texture feature values.Firstly,the calibration of the soil multi-parameter sensor was based on the relationship between dielectric constant and water content,electrical conductivity,and salt content.Secondly,the image digital processing technology was employed to extract the RGB and texture features of the soil.The most relevant variables were determined through correlation analysis,and an optimal fitting model incorporating RGB,texture features,water content,and salt content was constructed.Finally,the accuracy of the inversion method was verified using the sensor approach.【Result】The trivariate regression model,which fitted the water content and RGB,exhibited the most optimal fitting effect with an R^(2) value of 0.97.For the fitting of salt content to RGB and texture features,a one-variable polynomial model incorporating salt content and soil apparent white ratio demonstrated superior fitting performance when the salt content was greater than or equal to 6%,yielding an R^(2) value of 0.97.Conversely,for salt content below 6%,the autocorrelation(AUT)fitting between salt content and texture feature values was proved to be the most effective approach with an R^(2) value of 0.93.Upon comparing and calculating the water content and salt content obtained through both multi-parameter sensor calibration method and the inversion method proposed in this paper,it was observed that relative error ranges for water content measurement using these two methods fell within 0.27%-9.48%,while relative error ranges for salt content ranged from 0.07

关 键 词：图像处理 频域反射(FDR) 土壤色彩参数(RGB) 纹理特征 土壤水盐含量 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]