高分辨率道路遥感影像中震害信息的提取

发布时间：2018-11-22 20:16

【摘要】：地震的破坏性不仅会对社会经济和环境造成严重的破坏,而且对生命安全产生巨大的威胁。第一时间获取地震后灾害情况将有利于救援指挥部门制定救援工作方案,从而将损失和威胁降至最低。但地震造成道路坍塌、断裂等损毁,及其次生灾害造成道路严重受阻、掩埋等情况,导致救援人员、救援车辆无法进入灾区,救援工作将无法展开。近年来,随着遥感影像的空间分辨率的提高和传感器的不断更新,使得人类对地理信息的探测更加容易。通过遥感技术能够及时给交通抢修部门提供道路损毁程度、震害分布等信息,这对减轻灾害影响和抢救伤亡人员具有非常重要的意义。遥感系统在获取信息中受时间、光谱、空间以及分辨率等条件的限制,很难精确地观测和记录复杂又丰富的地理信息,而在获取观测数据时也会受到大气、云层和区域的复杂度等多种因素的影响,难免会存在一定的误差。论文采用二次多项式对影像中的畸变源进行几何纠正和相干增强各向异性扩散模型进行平滑处理,得到高质量的遥感影像,为后续信息提取可靠的基础数据。根据提取遥感信息的不同,其分割的尺度参数也不相同。当分割尺度选择不合理时会引起“欠分割”、“过分割”、“边缘不匹配”等问题。论文首先利用分形网络演化方法对原始影像进行小尺度分割;然后利用粒子群算法的全局搜索能力,从预分割的小尺度对象中确定最优初始聚类中心,在对小尺度对象聚类合并时,建立具有对象空间信息和对象间相关信息的目标函数;论文在不同分割尺度下对给出的算法进行了分割实验,并用eCognition Developer 8.7软件和分水岭算法进行了对比和定量评价。实验结果表明,论文给出算法分割效果更优,可得到适应不同尺度地物的分割结果,降低了多尺度分割方法对尺度参数的过度依赖。在对分割对象进行模糊分类时,论文通过分析损毁道路特征信息,在分类时根据特征的不同引入权重系数,从而提高了主要的、区分度好的特征的权重,并降低次要特征的权重。对遥感图像进行了分类实验,实验结果表明,相对于采用相同的权值的进行分类,对不同的特征赋予不同的权重时分类精度更高。
[Abstract]:The damage of earthquake will not only cause serious damage to social economy and environment, but also threaten the safety of life. The first time to obtain the disaster situation after the earthquake will be helpful for the rescue command department to draw up the rescue work plan, thus reducing the loss and threat to the minimum. However, the earthquake caused road collapse, fracture and other damage, and secondary disasters caused serious road obstruction, burial and other conditions, resulting in rescue workers, rescue vehicles can not enter the disaster area, rescue work will not be able to begin. In recent years, with the improvement of spatial resolution of remote sensing images and the continuous updating of sensors, it is easier for human to detect geographical information. The remote sensing technology can provide the road damage degree, earthquake damage distribution and other information to the traffic emergency repair department in time, which is of great significance to reduce the impact of disasters and rescue casualties. Remote sensing systems are constrained by time, spectrum, space and resolution in obtaining information, which makes it difficult to accurately observe and record complex and rich geographic information, and is also subject to the atmosphere when acquiring observational data, It is inevitable that there are some errors due to the influence of the complexity of clouds and regions. In this paper, the quadratic polynomial is used for geometric correction of distortion sources and smoothing of coherent enhanced anisotropic diffusion model. High quality remote sensing images are obtained and reliable basic data are extracted for subsequent information. According to the different extraction of remote sensing information, the scale parameters of the segmentation are also different. When the selection of segmentation scale is unreasonable, the problems of "undersegmentation", "over-segmentation" and "edge mismatch" will be caused. Firstly, the fractal network evolution method is used to segment the original image on a small scale. Then, using the global searching ability of PSO, the optimal initial clustering center is determined from the presegmented small scale objects, and the objective function with object spatial information and object correlation information is established when clustering and merging small scale objects. In this paper, the segmentation experiments are carried out under different segmentation scales, and the comparison and quantitative evaluation of the proposed algorithm are carried out with eCognition Developer 8.7 software and watershed algorithm. The experimental results show that the segmentation effect of the algorithm is better, and the segmentation results adapted to different scale objects can be obtained, and the over-dependence of multi-scale segmentation method on scale parameters is reduced. In the process of fuzzy classification of segmented objects, the paper analyzes the information of damaged road features, and introduces the weight coefficients according to the different features in the classification, thus increasing the weight of the main, well-differentiated features. And reduce the weight of secondary features. The experimental results of remote sensing images show that the classification accuracy of different features with different weights is higher than that of classification with the same weights.
【学位授予单位】：西安建筑科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】