OLI与HSI影像融合的土壤盐分反演模型

发布时间：2018-08-22 13:41

【摘要】：土壤盐渍化问题是黄河三角洲地区主要的土地退化问题,借助遥感技术快速、准确地掌握土壤盐渍化信息,对农业可持续发展具有重要意义。该文以黄河三角洲垦利县为研究区,利用超球体色彩空间变换算法,将环境一号卫星HSI高光谱影像与Landsat 8 OLI多光谱影像进行融合,选择土壤盐分的特征波段,结合土壤盐分的实测数据,建立统计分析模型(多元线性回归、偏最小二乘回归)和机器学习模型(BP神经网络、支持向量机和随机森林),对土壤盐分进行遥感反演。结果表明:OLI影像的统计分析模型和机器学习模型精度均较低,精度最高的随机森林模型相关系数仅为0.570;HSI影像的反演模型精度高于OLI,BP神经网络模型相关系数为0.607;融合影像反演模型精度明显高于HSI影像和OLI影像,土壤盐分含量的实测值与机器学习模型预测值具有良好的相关性,BP神经网络模型、支持向量机模型和随机森林模型的决定系数R~2分别达到0.966、0.821和0.926,模型反演精度较高。研究表明,多光谱和高光谱影像融合能显著提高土壤盐分遥感反演精度,机器学习模型的反演效果明显优于统计分析模型。研究结果对黄河三角洲典型地区的土壤盐分反演具有积极的理论和实践意义。
[Abstract]:Soil salinization is a major problem of land degradation in the Yellow River Delta. It is of great significance to grasp soil salinization information quickly and accurately by means of remote sensing technology. Taking Kenli County in the Yellow River Delta as the study area, using hypersphere color space transform algorithm, the HSI hyperspectral image of the environment 1 satellite and the Landsat 8 OLI multispectral image were fused to select the characteristic bands of soil salinity. Based on the measured data of soil salinity, a statistical analysis model (multiple linear regression, partial least square regression) and a machine learning model (BP neural network, support vector machine and random forest) were established to retrieve soil salinity by remote sensing. The results show that the accuracy of the statistical analysis model and the machine learning model of the Oli image are low. The correlation coefficient of random forest model with the highest precision is only 0.570%. The accuracy of the model is higher than that of OLII-BP neural network model (0.607), and the precision of fusion image inversion model is obviously higher than that of HSI image and OLI image. There is a good correlation between the measured values of soil salt content and the predicted values of machine learning model and the BP neural network model. The determination coefficients of support vector machine model and stochastic forest model are 0.9660.82 and 0.926, respectively. The inversion accuracy of the model is high. The results show that multi-spectral and hyperspectral image fusion can significantly improve the precision of soil salt remote sensing inversion, and the retrieval effect of machine learning model is obviously better than that of statistical analysis model. The results have positive theoretical and practical significance for soil salinity inversion in typical areas of the Yellow River Delta.
【作者单位】：山东农业大学信息科学与工程学院;山东农业大学资源与环境学院;北京工业技术学院;
【基金】：“十二五”国家科技支撑计划项目课题(2013BAD05B06,2015BAD23B0202) 国家自然科学基金(41271235) “双一流”奖补资金(SYL2017XTTD02)
【分类号】：S127;S156.4

【相似文献】