基于光谱分类的土壤盐分含量预测
发布时间：2019-02-18 19:47
【摘要】：基于相似土壤组分和光谱特征,利用土壤光谱反射率数据和曲线特征来进行土壤光谱分类,同时充分挖掘有效信息是光谱分析的重要应用方向之一。借助模糊k-均值聚类方法将土壤光谱数据分成四个类别(分类前先将原始光谱进行范围归一化处理),比较分析了不同类型土壤在光谱分类前后的高光谱特征,然后利用Kennard-Stone法将各类别样本划分为建模集和预测集,将预处理后的建模集光谱数据作为输入量,采用偏最小二乘回归法(PLSR)方法建立全局和各自类别的盐分预测模型。结果表明:光谱分类建模较按土壤系统分类建模和全局建模的精度有明显提高,其预测模型总体的预测决定系数RP2、预测均方根误差RMSEP、相对分析误差RPD和RPIQ(样本观测值三四分位数Q3与一四分位数Q1之差与RMSEP的比值)四个指标分别从0.664、1.219、1.733和1.461提高至0.818、1.132、2.356和2.422,其中RPD提高幅度达23.13%,四个类别所建模型RPD均大于2.0,可以对土壤含盐量进行较为精确的定量研究。研究结果为利用大样本光谱数据建立大尺度区域的盐分等土壤属性预测模型提供一种新的思路和方法。
[Abstract]:Based on similar soil components and spectral characteristics, soil spectral classification based on soil spectral reflectance data and curve features, while fully mining effective information is one of the important applications of spectral analysis. The soil spectral data were classified into four categories by using fuzzy k-means clustering method. The hyperspectral characteristics of different types of soils before and after spectral classification were compared and analyzed. Then the samples of each class are divided into modeling set and prediction set by using Kennard-Stone method. The spectral data of pre-processed modeling set is taken as input quantity, and the global and their salt prediction models are established by partial least square regression (PLSR) method. The results show that the precision of spectral classification modeling is significantly higher than that of soil system classification modeling and global modeling, and the prediction model total prediction decision coefficient RP2, is used to predict root mean square error (RMSEP,). The relative analysis errors RPD and RPIQ (the difference between quartile Q3 and quartile Q1 and RMSEP) increased from 0.664, 1.219, 1.733 and 1.461 to 0.818, 1.132, 2.356 and 2.422, respectively, and the difference between Q3 and Q1 increased from 0.664, 1.219, 1.733 and 1.461 to 0.818, 1.132, 2.356 and 2.422, respectively. The increase of RPD was 23.13, and the RPD of the four models was more than 2.0, so the salt content of soil could be studied accurately and quantitatively. The results provide a new idea and method for the prediction model of soil properties such as salinity in large scale region using large sample spectral data.
【作者单位】： 塔里木大学机械电气化工程学院;塔里木大学现代农业工程重点实验室;塔里木大学植物科学学院;
【基金】：国家自然科学基金项目(41261083,41361048,11464039)资助~~
【分类号】：S151.93
，
本文编号：2426144
[Abstract]:Based on similar soil components and spectral characteristics, soil spectral classification based on soil spectral reflectance data and curve features, while fully mining effective information is one of the important applications of spectral analysis. The soil spectral data were classified into four categories by using fuzzy k-means clustering method. The hyperspectral characteristics of different types of soils before and after spectral classification were compared and analyzed. Then the samples of each class are divided into modeling set and prediction set by using Kennard-Stone method. The spectral data of pre-processed modeling set is taken as input quantity, and the global and their salt prediction models are established by partial least square regression (PLSR) method. The results show that the precision of spectral classification modeling is significantly higher than that of soil system classification modeling and global modeling, and the prediction model total prediction decision coefficient RP2, is used to predict root mean square error (RMSEP,). The relative analysis errors RPD and RPIQ (the difference between quartile Q3 and quartile Q1 and RMSEP) increased from 0.664, 1.219, 1.733 and 1.461 to 0.818, 1.132, 2.356 and 2.422, respectively, and the difference between Q3 and Q1 increased from 0.664, 1.219, 1.733 and 1.461 to 0.818, 1.132, 2.356 and 2.422, respectively. The increase of RPD was 23.13, and the RPD of the four models was more than 2.0, so the salt content of soil could be studied accurately and quantitatively. The results provide a new idea and method for the prediction model of soil properties such as salinity in large scale region using large sample spectral data.
【作者单位】： 塔里木大学机械电气化工程学院;塔里木大学现代农业工程重点实验室;塔里木大学植物科学学院;
【基金】：国家自然科学基金项目(41261083,41361048,11464039)资助~~
【分类号】：S151.93
，
本文编号：2426144
本文链接：https://www.wllwen.com/kejilunwen/nykj/2426144.html
