基于深度学习的湖泊湿地信息提取及时空演变特征研究

发布时间：2018-01-03 22:01

本文关键词：基于深度学习的湖泊湿地信息提取及时空演变特征研究　出处：《东华理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：湖泊湿地是具有多种生态功能的生态系统,对人类和生态环境可持续发展至关重要。由于环境恶化,湿地面积正不断萎缩,对湿地进行保护和恢复迫在眉睫。利用遥感技术提取湿地空间分布信息,实施动态地监测湖泊湿地时空演变,能为湿地保护和决策提供重要数据支持。常用遥感影像湿地信息提取方法少有兼顾光谱与空间信息,难以保证湿地提取的准确性和可靠性。本文在利用遥感影像光谱特征时融入深度学习算法,构建空谱联合估计模型对湿地信息进行提取,获取不同时期湖泊湿地信息专题图分析湖泊湿地时空变化。主要包括:(1)构建基于深度学习的湿地识别模型。引入深度学习算法,建立多尺度卷积神经网络湿地识别模型,制作湿地样本,对基于深度卷积神经网络的湿地识别模型进行学习训练。(2)研究湿地提取空谱联合估计模型。运用深度学习卷积神经网络模型,与归一化差分水体指数和湿度分量相结合,综合考虑湿地空间特征和光谱特征,搭建空谱联合估计模型,对湿地进行提取。(3)将目前常用湿地提取方法与基于深度学习的湿地提取方法对比分析。采用最小距离法、最大似然法、决策树分类法、面向对象法、支持向量机法等与本文方法进行对比分析实验,评定精度,验证基于深度学习的湿地提取方法的可行性和准确性。(4)湖泊湿地时空演变特征研究。本文以鄱阳湖为例,分析湖泊湿地时间空间变化。运用景观生态学方法对2000年、2005年、2010年、2015年鄱阳湖湿地信息进行分析,利用景观格局指数定量分析湿地变化过程,研究其变化趋势,探寻湿地变化规律,为湿地保护和恢复决策提供数据支持,并提出相应保护建议。
[Abstract]:Lake wetland is an ecosystem with many ecological functions, which is very important for the sustainable development of human and ecological environment. It is urgent to protect and restore the wetland. Remote sensing technology is used to extract the spatial distribution information of the wetland and to dynamically monitor the temporal and spatial evolution of the lake wetland. It can provide important data support for wetland protection and decision-making. The common remote sensing image wetland information extraction method rarely takes account of spectral and spatial information. It is difficult to ensure the accuracy and reliability of wetland extraction. In this paper, a joint space-spectrum estimation model is constructed to extract the wetland information by using the spectral features of remote sensing images into the depth learning algorithm. The temporal and spatial changes of lake wetland are analyzed by obtaining the thematic map of lake wetland information in different periods. The main contents include: 1) construct the wetland identification model based on deep learning, and introduce the depth learning algorithm. A multi-scale convolution neural network wetland identification model was established to make wetland samples. The wetland recognition model based on deep convolution neural network is trained. (2) the joint estimation model of wetland extraction space spectrum is studied, and the deep learning convolution neural network model is used. Combined with normalized differential water index and humidity component, combined with spatial and spectral characteristics of wetland, a joint space-spectrum estimation model was built. The wetland extraction method is compared with the wetland extraction method based on deep learning. The minimum distance method, the maximum likelihood method, the decision tree classification method and the object oriented method are adopted. Support vector machine (SVM) method is compared with the method in this paper to evaluate the accuracy of the experiment. Verify the feasibility and accuracy of wetland extraction method based on in-depth learning. 4) study on temporal and spatial evolution characteristics of lake wetland. This paper takes Poyang Lake as an example. The temporal and spatial changes of lake wetland were analyzed. The information of Poyang Lake wetland on 2000, 2005, 2010 and 2010 were analyzed by landscape ecology method. The landscape pattern index is used to analyze the changing process of wetland, to study its changing trend, to explore the rules of wetland change, to provide data support for wetland conservation and restoration decision, and to put forward corresponding protection suggestions.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237;P901

【参考文献】