分数阶微分在土壤盐渍化遥感监测中的应用研究

发布时间：2018-06-04 21:17

本文选题：土壤盐渍化 + 遥感　；参考：《新疆大学》2017年博士论文

【摘要】：土壤作为一种重要的自然资源,为人类生产食物和纤维,并维持地球生态系统平衡;同时,土壤作为人类社会及一切社会活动的基本条件,承担重要的载体作用。然而,土壤盐渍化是干旱、半干旱区所面临的重要生态环境问题之一,土壤盐渍化引起的土壤板结、肥力下降、酸碱失衡、土地退化等后果,严重制约我国农业发展,影响当前我国可持续发展的战略大局。遥感技术因其尺度大、范围广、时效性强、经济性强等特点,很好的弥补了传统盐渍化现象监测方法的不足,为定量监测土壤盐渍化现象提供了崭新的途径。在前人利用遥感技术对土壤盐渍化现象进行监测等研究与应用的基础上,本研究选取位于新疆维吾尔自治区准噶尔盆地西南缘艾比湖流域的典型盐渍化土壤为研究对象,通过土壤样本采集、室内光谱测定、理化性质分析等野外与室内工作,明确该流域土壤理化性质特征,从光谱维和空间维探讨分数阶微分在遥感技术监测土壤盐渍化中的应用的可行性。在光谱维度上,利用Grünwald-Letnikov分数阶微分公式对可控光源条件下测定的盐渍化土壤光谱反射率数据进行0~2阶微分(阶数间隔0.1)处理,结合极差、相关系数、变异系数、光谱信息散度等指标,探讨分数阶微分对光谱反射率数据的影响,并以此为基础,利用偏最小二乘回归及组合建模理论建立土壤含盐量定量估算模型。在空间维度上,构建分数阶微分算子,对Landsat 8 OLI获取的多光谱遥感影像进行滤波处理,结合锐化程度、信息熵、结构相似性指数、峰值信噪比等图像质量评价指标,研究分数阶微分算子在图像增强中的作用,并在选取较为适合艾比湖保护区盐渍化信息空间可视化表征的盐分指数后,探讨分数阶微分滤波对盐分指数提取时产生的影响。主要结论如下:(1)艾比湖流域表层土壤含盐量变异性强,均值为33.838 g/kg,大于盐土20 g/kg的分级标准,流域内土壤盐渍化现象普遍而又严重;阳离子含量Na~+Ca~(2+)K~+Mg~(2+),阴离子含量SO_4~(2-)Cl~-HCO_3~-,结合相关性分析结果,NaCl和CaSO_4是主要的盐分,艾比湖流域土壤主要以碱性和强碱性为主,盐化类型主要以Cl~-盐型和SO_4~(2-)盐型为主。(2)艾比湖流域内,土壤中砂粒含量最多,黏粒含量最少,根据国际制土壤质地分类标准,流域内土壤可分为粉质壤土、壤土、砂质壤土、砂土及壤质砂土;分形维数与黏粒、粉粒、砂粒均呈现较好的非线性关系;利用逐步多元回归方法,建立基于土壤机械组成的分形维数估算模型,该模型为2.844 0.002 0.006D clay sandy(28)(10)x-x,决定系数0.720;根据描述性统计结果,黏粒、粉粒与分形维数随着盐渍化程度的加深而增加,砂粒含量随着盐渍化程度的加深而减少。(3)不同盐渍化程度土壤光谱曲线形态基本相似,在400~500 nm、550~700 nm、1000~1850 nm以及2250~2400 nm这4个范围内,区分明显;在1400 nm和1900 nm附近这两个吸收带,随着盐分含量的增加,光谱中水的吸收谷加深;在2340 nm这个吸收带,吸收深度随着盐渍化程度的降低而加深;在400~1875 nm范围内,反射率与土壤含盐量相关系数均通过了0.01极显著性检验,而在1876~2400 nm,则没有波段通过检验。(4)对于光谱反射率的5种数学形式,与土壤含盐量的相关系数通过显著性检验的波段数量:未微分一阶微分二阶微分;将阶数间隔进行细化后,相关系数曲线呈现一定的渐变趋势,部分细节信息得以体现,相关系数通过显著性检验的波段数量呈现先减少,再增加,而后再减少的趋势;随着微分阶数的增加,光谱微分值趋近于0,数据范围减小;结合消除量纲影响的变异系数,虽然数据范围减小,但是数据的离散程度总体上呈现增加的趋势;光谱信息散度随着阶数的增加呈现非线性的增加趋势,且从0阶到1阶增加的幅度大于从1阶到2阶的幅度,这表明分数阶微分能够细化相关系数、极差、变异系数、光谱信息散度等指标的变化趋势,并且在一定程度上可以增加光谱数据的离散程度,增强光谱间的差异性。(5)针对光谱反射率的5种数学形式0~2阶微分数据,利用偏最小二乘回归建立的105个土壤盐分估算模型中,基于原始光谱反射率均方根变换1.9阶微分数据建立的模型最优,该模型的RMSE_C=26.206 g/kg,R~2C=0.787,RMSEP=24.955 g/kg,R~2P=0.819,RPD=2.352。对简单平均法进行改进,根据RPD的值对每个子模型赋予权重,通过对21个RPD≥2的模型进行加权性质的组合,组合后的模型RMSE_C=26.291 g/kg,R~2C=0.786,RMSEP=24.332 g/kg,R~2P=0.828,RPD=2.413,定量估算土壤含盐量的精度与预测能力得到进一步提升。(6)对于真彩色和标准假彩色合成后的图像,在分数阶微分算子的阶数大于0.8时,R、G、B三个通道的平均峰值信噪比小于20 d B,图像失真程度不可接受;对于分数阶微分滤波后的图像,与原图相比,虽然在一定程度上丢失了图像信息,破坏了与原图的结构相似性,但是PSNR和锐化程度在0.6~0.7阶之间达到平衡,图像增强的效果达到最优。此外,当微分阶数大于1时,分数阶微分算子对图像中的某些边缘双重响应,产生双像素边界。(7)针对艾比湖保护区,盐分指数SI_5(28)(B′R)/G与土壤含盐量的相关系数为0.3532,通过了0.01显著性检验,较为适合保护区内盐渍化信息空间可视化表征;从经过0.1~0.8阶微分滤波增强后图像提取盐分指数SI_5,与土壤含盐量的相关系数呈现先增加再减少的趋势,并在微分阶数为0.4时,达到最大值0.3583;在微分阶数为0.8时,相关系数大幅降低且仅通过0.05显著性检验,表明分数阶微分算子在进行图像增强处理时,微分阶数在一定范围内有利于提取土壤盐渍化信息;而超过这个范围,会对盐分信息的提取产生不利影响。另外,随着微分滤波器阶数的增加,盐分指数空间分布的部分细节信息得以体现,这与分数阶微分算子图像增强的能力息息相关。本研究将分数阶微分引入到遥感数据预处理中,为基于遥感数据提取地物信息提供更多的辅助性方法,也为机载、星载遥感技术在大尺度上对盐渍化专题信息提取、土壤盐渍化现象监测以及遥感专题制图等提供应用参考,以期满足未来对土壤盐渍化现象监测、精准农业等应用的更高要求。
[Abstract]:Soil, as an important natural resource, produces food and fiber for human beings and maintains the balance of the earth's ecosystem. At the same time, soil is an important carrier for human society and all social activities. However, soil salinization is one of the important ecological environmental problems faced by drought and semi dry areas, soil salt. The results of soil consolidation, fertility decline, acid-base imbalance, land degradation, etc., seriously restrict the development of China's agriculture and affect the current strategic situation of sustainable development in China. Remote sensing technology has made up for the shortcomings of the traditional salinization monitoring method because of its large scale, wide range, strong timeliness and strong economy. On the basis of the research and application of remote sensing technology for monitoring soil salinization, this study selected typical salinized soil in the EBI Lake Basin in the southwest edge of Junggar basin, the Xinjiang Uygur Autonomous Region, as the research object, and collected the soil samples. In the field and indoor work, the characteristics of soil physical and chemical properties of the basin are clearly defined. The feasibility of applying fractional differential in remote sensing technology to monitor soil salinization is discussed from spectral dimension and space dimension. On the spectral dimension, the Gr u nwald-Letnikov fractional differential equation is used to determine the controlled light source. The spectral reflectance data of salinized soil are processed by 0~2 order differential (order interval 0.1), and the influence of fractional differential on spectral reflectance data is discussed in combination with the extreme difference, correlation coefficient, coefficient of variation and spectral information divergence. On the basis of this, the soil salt content quantitation is established by using the least two multiplicative regression and combined modeling theory. On the spatial dimension, the fractional differential operator is constructed, and the multi spectral remote sensing images obtained by Landsat 8 OLI are filtered, combined with the sharpening degree, the information entropy, the structural similarity index, the peak signal to noise ratio and so on, the function of the fractional differential operator in image enhancement is studied, and it is more suitable for the selection of the image enhancement. The effect of fractional differential filtering on the extraction of salt index is discussed after the salinization index of the salinization information space of the Ebinur lake protection area. The main conclusions are as follows: (1) the salinity variation of the surface soil in the Ebinur Lake Basin is strong, the mean value is 33.838 g/kg, the soil salinity is higher than the salt soil 20 g/kg, and the soil salinization in the basin The cation content Na~+Ca~ (2+) K~+Mg~ (2+) and anion content SO_4~ (2-) Cl~-HCO_3~-, combined with correlation analysis results, NaCl and CaSO_4 are the main salts, the main soil in the Ebinur Lake Basin is alkaline and strong alkaline, and the main types of salinization are Cl~ - salt and SO_4~ (2-) salt. (2) the sand in the lake basin According to the international classification standard of soil texture, the soil in the basin can be divided into silty loam, loam, sandy loam, sandy soil and loam sandy soil, and the fractal dimension has a good nonlinear relationship with the clay particles, powder and sand, and the fractal dimension based on the soil mechanical composition is established by the stepwise multiple regression method. The model was estimated to be 2.844 0.002 0.006D clay Sandy (28) (10) x-x, and the determining coefficient was 0.720. According to the descriptive statistics, the clay particles, powder and fractal dimension increased with the deepened salinization degree, and the sand content decreased with the deepening of the salinization degree. (3) the soil spectral curves of different salinization degree were basically similar, at 400. The 4 ranges of ~500 nm, 550~700 nm, 1000~1850 nm, and 2250~2400 nm are distinct; the two absorption bands near 1400 nm and 1900 nm increase with the increase of salt content, the absorption valley of the water is deepened; in the 2340 nm absorption band, the absorption depth is deepened with the decrease of the salinization range. The correlation coefficient of soil salt content passed the 0.01 polarity test, while in 1876~2400 nm, there was no wave band through test. (4) the 5 mathematical forms of spectral reflectance, the number of bands associated with the correlation coefficient of soil salt content through the saliency test: the two order differential of the first order of the first order; after refining the interval of the order, the correlation system The number curve presents a certain trend of gradual change, some details can be reflected, the correlation coefficient is first reduced, then increased, then decreasing, and the spectral differential value tends to 0 and the data range decreases with the increase of the differential order. However, the degree of dispersion of the data is increasing in general, and the spectral information divergence increases with the increase of the order, and the increase from the 0 order to the 1 order is greater than that from the 1 to the 2 order, which indicates that the fractional differential can refine the correlation coefficient, the difference, the coefficient of variation, the spectral information divergence and so on. The trend, and to a certain extent, can increase the degree of dispersion of spectral data and enhance the difference between spectra. (5) for the 5 mathematical form 0~2 differential data of spectral reflectance, 105 soil salinity estimation models established by partial least squares regression are built on the basis of the 1.9 order differential data of the original spectral reflectance mean square root transformation. The model is optimal, and the model's RMSE_C=26.206 g/kg, R~2C=0.787, RMSEP=24.955 g/kg, R~2P=0.819, RPD=2.352. are improved for the simple mean method. According to the value of RPD, each sub model is weighted, and the weighted properties of the 21 RPD > 2 models are combined, and the model RMSE_C=26.291 g/kg, R~2C=0.786, etc. G, R~2P=0.828, RPD=2.413, the accuracy and prediction ability of quantitative estimation of soil salt content is further improved. (6) the average peak signal to noise ratio of the three channels of R, G, B is less than 20 d B when the order of fractional differential operator is greater than 0.8, and the degree of image distortion is unacceptable. The image after differential filtering is compared with the original image, although the image information is lost to a certain extent and the structural similarity between the original image is destroyed, but the PSNR and the sharpening degree are balanced between the 0.6~0.7 order, and the effect of the image enhancement is optimal. In addition, when the differential order is greater than 1, the fractional differential operator has some edge double in the image. (7) the correlation coefficient of the salt index SI_5 (28) (B 'R) /G and the soil salt content is 0.3532 for the Ebinur lake protection area, and it passes the 0.01 significant test. It is more suitable for the visualization of the salinization information space in the protected area. The salt index SI_5 is extracted from the image after the 0.1~0.8 order differential filtering enhancement, and the soil is extracted from the soil. The correlation coefficient of soil salt content increases first and then decreases, and reaches the maximum value of 0.3583 when the differential order is 0.4. When the differential order is 0.8, the correlation coefficient is greatly reduced and only through 0.05 significant tests. It shows that the fractional order differential operator is beneficial to the extraction of soil in a certain range when the image enhancement is carried out. In addition, with the increase of the order of differential filter, some details of the spatial distribution of the salt index are reflected, which is related to the ability interest rate of the fractional differential operator image enhancement. The fractional differential is introduced to remote sensing data in this study. In preprocessing, it provides more auxiliary methods for extracting ground information based on remote sensing data. It also provides reference for airborne, spaceborne remote sensing technology to extract salinization information, soil salinization monitoring and remote sensing thematic mapping in large scale, in order to meet the future monitoring of soil salinization, precision agriculture and so on. Higher requirements for application.
【学位授予单位】：新疆大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：S156.41

【相似文献】