【作者】 刘静；

【导师】 王晓君；

【作者基本信息】 河北科技大学 ， 测试计量技术及仪器， 2016， 硕士

【摘要】 随着我国“高分专项”系统的进一步推进,为我国提供了更大空间产业化发展机会,在技术和应用等方面获得了一系列成果,同时取得了良好的社会和经济效益。植被调查是遥感的重要应用领域。植被分类识别为政府部门对监测和优化植被群落结构提供了科学的依据,对于人类生存环境的可持续发展有着重要意义。本文研究对象主要针对目前“高分”系列运用最广泛的GF-1遥感卫星影像。通过不同地物反射的电磁波在影像上呈现的光谱、纹理等进行特征分析,对遥感影像中的植被进行分类识别。论文主要完成了以下内容:首先对原始遥感影像进行预处理,对预处理的各个步骤的原理及操作方法做了详细的阐述,其中主要对融合算法进行分析,试验后选择效果好的Pansharpen算法,使得融合后的影像清晰度更高,为后面的工作做好铺垫。然后对预处理后的影像进行植被光谱和纹理特征的提取,分析特征提取的算法以及常见的植被指数类型,通过实验发现利用植被指数NDVI效果比其他植被指数效果理想;证明了基DS理论证据融合多特征的遥感图像植被提取方法,效果优于单独使用光谱特征或者纹理特征的方法。其次利用监督的BP神经网络分类方法完成植被分类识别,分析监督分类和非监督分类的各自特点,深入研究多种类型的BP神经网络的优缺点及优化算法,通过仿真结果显示LM算法收敛效果最好,适合做分类识别。最后完成植被分类识别系统的实现,在Visual Studio 2010的平台上利用C#语言,并结合GDAL库编程技术,开发了一个对植被进行提取和分类识别的系统,其特点是用户可以自由选择样本和提取方法进行分类识别。更多还原

【Abstract】 With the further advance of our country’s “High-resolution special project ”system,industrialization development opportunities for our country provides a larger space,a series of achievements are obtained in the aspects of technology and applications,at the same time achieved good social and economic benefits.Vegetation survey is an important application of remote sensing field.The vegetation classification provides a scientific basis for government departments to monitor and optimize the structure of vegetation community,it has important significance for the sustainable development of the human survival environment.In this paper,the object of study mainly aiming at "High-resolution" series is the most widely used GF-1 satellite remote sensing images.The spectral and texture features of electromagnetic waves reflected by different objects are analyzed,and the vegetation in remote sensing images is classified and identified.The main contents of this paper are as follows:Firstly,the original remote sensing image is pretreated.The principle and operation method of each step are expatiated in detail,which mainly analyzes the fusion algorithm,after the test,the selection effect of the Pansharpen algorithm is better,so that the fused image is clearer,and pave the way for the following work.Secondly,the vegetation spectrum and texture feature extraction of the processed image were extracted.The feature extraction algorithm and common type of vegetation index are analyzed,through experiments discovered using vegetation index NDVI effect is more ideal than other vegetation index;It is proved that the method of vegetation extraction with multi features based on DS theory is better than the method of using spectral feature and texture feature alone;Then,the classification method of BP neural network is used to classify the vegetation classification,and the characteristics of supervised classification and unsupervised classification are analyzed,the advantages and disadvantages of the BP neural network and the optimization algorithm are studied deeply.Simulation results show that the LM algorithm has the best convergence effect and is suitable for classification and recognition.Finally,the system of vegetation classification and recognition is completed.On the platform of Visual Studio 2010,a system of extracting and classifying vegetation is developed by using C # language combined with GDAL library programming technology.The feature is that users can freely choose samples and extract methods for classification and identification.更多还原