高分辨率遥感影像中云和似云目标的自动区分

发布时间：2018-10-15 11:21

【摘要】：云的存在会对遥感影像的处理及目标识别等产生影响,因此,自动提取云对高分辨率卫星影像的应用具有重要意义。高分影像上更加复杂的云的细节形态及似云目标的干扰,使得高分影像的自动云提取难以达到实用水平。本文以雪地为例,选取形状、纹理和边缘3个差异化特征作为云与似云目标区分的关键,提出了一种区分高分辨率遥感影像中云和似云目标的云检测算法。首先利用Wallis滤波对输入影像进行预处理,增强影像中不同尺度的影像纹理模式;然后对影像进行快速稳定的均值漂移分割,利用灰度和纹理特征构成支持向量机的第一层分类器,将分割后的区域对象分成"云"和普通地物,再利用边缘、形状、纹理等特征结合灰度特征构成支持向量机的第二层分类器,将"云"区分为云区和似云目标;最后使用Grab-cut对云检测结果进行边缘迭代精化。本文算法取得了优良的试验结果,证明了算法在似云目标干扰下对高分辨率遥感影像进行精确云检测的能力。
[Abstract]:The existence of cloud will affect the processing of remote sensing image and target recognition. Therefore, automatic cloud extraction is of great significance to the application of high-resolution satellite image. Because of the more complicated cloud details and the interference of cloud-like targets, the automatic cloud extraction of high-score images is difficult to reach the practical level. Taking snow as an example, this paper proposes a cloud detection algorithm to distinguish cloud and cloud-like targets in high-resolution remote sensing images by selecting three different features of shape, texture and edge as the key to distinguish cloud from cloud-like objects. Firstly, Wallis filter is used to preprocess the input image to enhance the image texture pattern of different scales in the image, and then the fast and stable mean shift segmentation of the image is carried out. Using gray and texture features to form the first layer classifier of support vector machine, the segmented region objects are divided into "cloud" and ordinary objects, and then the edges and shapes are used. Texture and other features combined with gray features constitute the second layer classifier of support vector machine (SVM), which divides "cloud" into cloud region and cloud-like target. Finally, Grab-cut is used to refine the edge of cloud detection results. In this paper, good experimental results are obtained and the ability of cloud detection for high resolution remote sensing images under cloud-like target interference is proved.
【作者单位】：浙江省测绘科学技术研究院;武汉大学遥感信息工程学院;
【基金】：浙江省科技技术项目(2015C33010)
【分类号】：P237

【相似文献】