遥感图像道路提取研究

发布时间：2018-04-02 14:16

本文选题：遥感图像　切入点：纹理识别　出处：《江苏大学》2014年博士论文

【摘要】：高分辨率遥感图像的应用,可以使我们获取了更加精确、丰富和全面的信息。从遥感图像中抽取出信息,通过识别出感兴趣的目标获取知识,完成图像的理解是遥感图像应用的根本目标。遥感图像可以提供的信息中,道路信息是很重要的一部分。随着道路信息的不断更新,传统的人工操作已无法满足需求,于是,将遥感技术与电子技术及图像识别技术结合起来,研究遥感图像道路的自动提取,对于道路监控和GPS导航及地图及时更新都有重大意义,也是目前国内外研究的重点。论文主要研究遥感图像道路提取的方法,首先探讨了在线数据库中道路遥感图像的识别分类,然后对遥感图像云层进行自动识别的研究,以避免遥感图像中存在云层覆盖的情况,最后分别用两种方法对遥感图像中道路进行提取。针对在线数据库遥感图像分类,提出基于文本与图像信息融合的道路遥感图像识别分类的方法,需要先提取图像的文本特征与图像特征,再将二者的特征进行融合,通过支持向量机的训练,可得到较好的分类结果,该方法可移植到其他图像在线数据库分类识别与构建上。针对遥感图像云层,提取图像的纹理特征,选择4个纹理特征参数——角二阶矩、对比度、相关度和熵对云层图像进行自动识别。针对遥感图像道路,首先使用结构张量的方法进行主方向的计算,并改进了主方向的计算方法,结合Gibbs抽样对道路进行提取,该方法适用于有遮挡的道路,但无法说明道路的重要性；采用圆投影变换进行道路提取,通过与初始模板匹配找到最优模板来提取道路,该方法可以观察到道路的重要性。论文研究的具体内容如下： (1)基于文本与图像信息融合的道路遥感图像识别。在线图像数据库中的图像并不是在相同的实验室环境和相同的技术参数中统一测量的,因此,现有的方法并不能直接对道路遥感图像进行识别。针对这一问题,利用在线图像数据库中的道路遥感图像及其注释来获得高精度的道路遥感图像识别。利用空间金字塔关键字直方图来描述图像的特征,并通过融合图像和文本信息提高了道路遥感图像的识别精度。使用从图像数据库得到的图像信息和文本信息训练支持向量机以得到更高的分类精度,然后在整合所有信息后可以得到支持向量机的后验概率值和最终结果。相较于使用单独的图像特征或单独的文本特征进行识别,该方法具有较好的识别准确率和分类性能,可移植到其他图像在线数据库分类识别与构建上。 (2)基于纹理特征对遥感图像云层自动识别。针对高分辨率遥感图像中云层的自动识别问题,提出一种基于图像纹理特征的云层自动识别方法,通过灰度共生矩阵来对图像中云层和下垫面的纹理特性进行统计分析,选择对云层和下垫面进行有效区分的4个纹理特征参数——角二阶矩、对比度、相关度和熵对图像进行识别,最后通过图像空间域的云层识别方法来对纹理识别结果进行修正,有效提高了云层识别的准确性,为遥感图像道路提取奠定基础。 (3)基于结构张量进行遥感图像道路提取。针对基于结构张量的主方向计算方法计算结果不够精确的缺点,对高斯滤波进行改进并结合canny算子,提出了改进的主方向计算方法,然后基于局部主方向结合Gibbs抽样进行遥感图像道路提取。该算法适用于有遮挡的道路图像,可以比较精确地对道路进行提取。 (4)基于圆投影变换进行遥感图像道路提取。本文中基于结构张量的遥感图像道路提取方法不能表明道路的关键性,采用圆投影变换理论进行遥感图像道路提取,可以比较精确的提取道路,同时从结果图中观察到道路的重要性信息。
[Abstract]:The application of high resolution remote sensing image, so that we can get more accurate, rich and comprehensive information. The information extracted from the remote sensing images, acquire knowledge through identifying the target of interest, image comprehension is the fundamental goal of the application of remote sensing image. The remote sensing images can provide information, road information is a part of very important. With the road information update, the traditional manual operation has been unable to meet the demand, so the combination of remote sensing technology and electronic technology and image recognition technology, automatic extraction of remote sensing image for roads, road monitoring and GPS navigation and map update are of great significance, is also the focus of research at home and abroad.
The main research method of extracting remote sensing image of the road, first discusses the classification of remote sensing image Road online database, and then research on the automatic identification of the remote sensing images of clouds, cloud cover to avoid the presence of remote sensing images, finally on the road in the remote sensing images were extracted by two methods. According to the online database of remote sensing image the classification method of text and image information fusion road remote sensing image classification recognition based on the need to extract text features and image features, and then the characteristics of the two integration, through the training of support vector machine, can get good classification results, this method can be transplanted to other online database and the construction of image classification and recognition in remote sensing images. Cloud image texture feature extraction, corner two moments 4 texture feature parameters, contrast, correlation The degree and entropy of automatic detection of cloud image. On the remote sensing image path calculation method of first use of the structure tensor principal direction, and improved the calculation method of the main direction, combined with the Gibbs sampling of road extraction, this method is applicable to a sheltered Road, but not to illustrate the importance of the road for road extraction; the circular projection transformation, with the initial template matching template to find the optimal extraction path, this method can observe the importance of the road.
The specific contents of the thesis are as follows:
(1) remote sensing identification text and image information fusion based on road image. The online image database image is not a unified measurement in the laboratory environment and the same technical parameters, therefore, the existing methods can not be directly identified on the road of remote sensing image. To solve this problem, using the way of remote sensing image online image in the database and comments to obtain road remote sensing image recognition with high accuracy. To describe the image feature space Pyramid keyword histogram, and through the integration of image and text information and improve the way of remote sensing image recognition accuracy. The use of the image information from the image database and text information of training support vector machine to obtain higher classification accuracy then, in the integration of all the information can be obtained after the support vector machine a posteriori probability and the final results. Compared with the single use of the map The recognition method of image feature or individual text feature has good recognition accuracy and classification performance, and can be transplanted to other image online database classification, recognition and construction.
(2) texture features of remote sensing image automatic recognition based on cloud. Aiming at the problem of automatic recognition in high resolution remote sensing image clouds, proposes a method for automatic identification of image texture features based on the clouds, through the gray level co-occurrence matrix for the analysis of images of clouds and surface texture characteristics, choose 4 texture parameters to effectively distinguish between clouds and surface -- angle of two order moment, contrast, correlation and entropy for image recognition, and finally through the cloud image recognition method of spatial domain to texture recognition result, effectively improve the accuracy of the recognition, lay the foundation for the extraction of remote sensing image.
(3) based on the structure tensor of remote sensing image road extraction. Based on the results of the analysis calculation method of principal direction based on structure tensor is not precise enough shortcomings, combined with improved Canny operator on Gauss filter, puts forward the main direction of the improved calculation method, and then the local principal direction based on Gibbs sampling for remote sensing image. The algorithm for road extraction in a road image block, on the road can be extracted accurately.
(4) the circle projection transform based on remote sensing image road extraction. The key in this paper extraction method can not show the way of remote sensing image path based on structure tensor, remote sensing image road extraction by theory circle projection transform, can be extracted accurately from the road, at the same time in the observation to the importance of the road information.

【学位授予单位】：江苏大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】