基于高分辨率遥感影像纹理特征的水土保持措施提取方法研究 英文标题:A Study on Methodology of Soil Conservation Practices Extraction Based on Texture Features of Hi-resolution Remotely Sensed Imagery

黄土高原丘陵沟壑区是黄河流域水土流失治理的重点区域。水土保持监测是水土流失治理的一项基础工作。就土地利用和水土保持措施信息的提取而言，以往的研究主要是以TM等中等分辨率的遥感影像数据为基础。梯田、果园、人工林草地和坝地等水土保持措施呈小片状分布，由于受到分辨率的限制，不能有效提取。以较高分辨率影像作为信息源开展黄土高原丘陵沟壑区水土保持监测不仅十分重要，而且十分迫切。本文以陕北绥德韭园沟流域和陕北延安市麻庄流域为实验区，以2.5米分辨率的SPOT-5影像和1米分辨率的IKONOS影像为实验数据，利用傅立叶变换可以将遥感影像由空间域转换到频率域进行能量叠加的性质，采用最大方向频谱能量值比值和光谱值相结合的方法对具有纹理特征的梯田和果园两种水土保持措施进行了提取研究;探讨了此方法与传统监督分类和非监督分类方法对具有纹理特征的水土保持措施的分类差异，并且对两种传统分类方法分类结果进行了对比分析。通过研究得出以下结论:1)利用图像频率域中的最大方向频谱能量值比值和光谱值相结合的方法对SPOT-5影像和IKONOS影像梯田和果园样区进行提取，精度最低达到78.21%，在一定程度上提高了分类精度。2)对于斑块较大，宽度大于10米，内部光谱值差异大的梯田，SPOT-5影像提取结果较好，精度最低达到78.21%。而IKONOS影像则更适合于对于梯田田面较窄的3-7米梯田提取，精度达到81.36%以上。3)和传统分类方法对比基于纹理特征的提取方法精度高于监督分类和非监督分类。对于SPOT-5影像的梯田提取，监督分类和非监督分类方法分类结果类似，果园的提取监督分类结果略优于非监督分类结果。而对于IKONOS影像梯田提取，两种分类结果精度相差12%以上，监督分类结果明显优于非监督分类结果。本文对具有纹理特征的水土保持措施的提取进行了分析研究，其结果在一定程度上为区域土壤侵蚀调查和效益评价提供了参考。

The hilly and gully region of the Loess Plateau is a key area for soil erosion control in the Yellow River Basin. Soil and water conservation monitoring is a fundamental task for soil erosion management. Previous studies on the extraction of land use and soil and water conservation measure information mainly relied on medium-resolution remote sensing imagery data such as TM. However, soil and water conservation measures including terraces, orchards, artificial forest and grassland, and check-dam fields are distributed in small patches, which cannot be effectively extracted due to resolution limitations. Therefore, carrying out soil and water conservation monitoring in the hilly and gully region of the Loess Plateau using high-resolution imagery is not only highly important but also urgently needed.

This study took the Jiuyuangou Basin in Suide, northern Shaanxi, and the Mazhuang Basin in Yan’an, northern Shaanxi as experimental areas, and used 2.5-meter-resolution SPOT-5 imagery and 1-meter-resolution IKONOS imagery as experimental data. Leveraging the property that Fourier transform can convert remote sensing imagery from the spatial domain to the frequency domain for energy superposition, this study adopted a method combining the maximum directional spectral energy value ratio and spectral values to extract two types of soil and water conservation measures with texture features: terraces and orchards. It also explored the differences in classification performance of this method compared with traditional supervised classification and unsupervised classification methods for texture-featured soil and water conservation measures, and conducted a comparative analysis of the classification results of the two traditional classification methods.

The following conclusions were drawn from this study: 1) The method combining the maximum directional spectral energy value ratio in the image frequency domain and spectral values was used to extract terraces and orchards from SPOT-5 and IKONOS imagery, with a minimum extraction accuracy of 78.21%, which improved the classification accuracy to a certain extent. 2) For terraces with large patches, a width greater than 10 meters, and large internal spectral value differences, SPOT-5 imagery yielded better extraction results, with a minimum accuracy of 78.21%. In contrast, IKONOS imagery is more suitable for extracting 3-7 meter-wide terrace surfaces with narrow ridge widths, with an accuracy of over 81.36%. 3) Compared with traditional classification methods, the texture-feature-based extraction method achieved higher accuracy than supervised and unsupervised classification. For terrace extraction from SPOT-5 imagery, the results of supervised and unsupervised classification were similar; for orchard extraction, supervised classification performed slightly better than unsupervised classification. For terrace extraction from IKONOS imagery, the accuracy difference between the two classification methods exceeded 12%, with supervised classification significantly outperforming unsupervised classification.

This study analyzed and researched the extraction of texture-featured soil and water conservation measures, and its results provide a reference for regional soil erosion surveys and benefit evaluation to a certain extent.

搜集汇总

背景与挑战

背景概述

该数据集是关于基于高分辨率遥感影像纹理特征的水土保持措施提取方法的研究，发布于2021年，属于协议共享数据。研究以黄土高原丘陵沟壑区为实验区，使用SPOT-5和IKONOS影像，通过傅立叶变换和纹理特征结合的方法提取梯田和果园，精度达78%以上，并证明该方法优于传统分类方法。