基于Spark的遥感影像多时相变化检测系统

发布时间：2018-05-13 15:05

本文选题：遥感影像 + 多分类　；参考：《浙江大学》2017年硕士论文

【摘要】：近年来随着各种新型的传感器不断涌现,遥感技术的提升,我国的高分辨率遥感卫星技术取得了飞速地发展,高分辨率遥感影像的数据级别趋于海量化发展,数据类型也越来越多样化。海量的遥感影像数据带来更多信息的同时也给快速处理带来了很大的挑战。由于卫星周期性旋转的特点,同一个地区在不同时间将会被卫星拍摄到很多次。通过检测同一地区不同时间影像发生的变化,有利于发现该地区地面覆盖变化情况。变化检测算法根据影像分析的层次不同可以分为像素级、特征级和目标级这三类,根据数据分析的机理,变化检测算法可以分为有监督和无监督两类。传统的变化检测都是基于单机来处理,面对遥感影像数据量较小的场景,处理比较方便,但是面对高分辨率遥感影像,处理速度较慢甚至不能处理。本文主要针对高分辨率遥感影像多时相变化检测问题,利甩Spark这一基于内存计算的计算引擎来处理,加快检测速度。本文采用有监督的像素级变化检测方法。面对高分辨率遥感影像数据,首先利用降维算法,找出主成分,然后利用手动标记的方法,找出各地区的样本点,根据像素点信息及空间信息训练出多分类模型,根据模型将原始遥感影像划分成不同区域,然后对比不同区域的变化得出变化分析报表。
[Abstract]:In recent years, with the continuous emergence of various new sensors and the improvement of remote sensing technology, the high-resolution remote sensing satellite technology in China has made rapid development, and the data level of high-resolution remote sensing image tends to the development of sea quantification. Data types are also becoming more and more diverse. The massive remote sensing image data bring more information, but also bring a great challenge to the fast processing. Because of the periodic rotation of the satellite, the same area will be photographed many times at different times. By detecting the changes of different time images in the same area, it is helpful to find out the change of ground cover in the same area. According to the different levels of image analysis, change detection algorithms can be divided into pixel level, feature level and target level. According to the mechanism of data analysis, change detection algorithms can be divided into two categories: supervised and unsupervised. The traditional change detection is based on a single machine. It is more convenient to deal with the scene where the data of remote sensing image is small, but in the face of high resolution remote sensing image, the processing speed is slow or even can not be processed. Aiming at the problem of multi-temporal change detection of high-resolution remote sensing images, this paper is aimed at removing Spark, which is a computing engine based on memory computing, to speed up the detection. A supervised pixel level change detection method is used in this paper. In the face of high resolution remote sensing image data, we first use dimension reduction algorithm to find out the principal components, then use manual marking method to find out the sample points of each region, and then train the multi-classification model according to pixel information and spatial information. According to the model, the original remote sensing image is divided into different regions, and then the change analysis report is obtained by comparing the changes of different regions.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】