基于GF-1遥感湿地类型提取研究

发布时间：2018-04-18 14:33

  本文选题：湿地 + 遥感　； 参考：《中南林业科技大学》2017年硕士论文

【摘要】：目前,湿地资源作为一种与森林、海洋生态系统同样地位的重要自然资源,它的变化和可持续保护利用是地球科学核心内容的重要部分。利用遥感技术在湿地区域进行监测成为越来越重要的手段,可以解决湿地研究中一些科学问题,如湿地类型信息、湿地景观信息、湿地变化特征等。通过这些研究可以推动湿地区域生态资源保护和发展,从而保障政府对湿地保护工程和保护区建设的科学、正确的决策。研究选择东洞庭湖作为主要的研究区,以GF-1遥感影像为数据源,对所采用的遥感数据源进行预处理,再进行最佳波段组合、最佳融合方法分析,然后将遗传算法技术和Fisher判别法引入湿地类型提取,一方面优化支持向量机算法实现湿地类型高精度提取,另一方面采用Fisher判别法对湿地类型进行简单、高效、自动提取;最后对四种算法的提取结果进行分析评价,旨在分析出适用于GF-1遥感影像湿地类型提取的最优算法,完善湿地类型提取算法体系,为今后湿地遥感研究提供依据。主要研究结果如下:(1)最佳波段组合分析研究研究综合考虑光谱特征与信息量大小,通过标准差、信息熵、最佳指数3个定量评价指标以及目视效果判断,确定GF-1遥感影像最佳波段组合为RGB=432。(2)最佳融合方法分析研究融合效果评价采用了定量评价的方式,从光谱继承性和空间融入度两方面分析融合效果,采用了均值、相关系数、熵、标准差、梯度五个指标分别对主成分变换融合、Gram-Schmidt融合和基于平滑滤波亮度调整融合的结果进行定量评价。在光谱继承性和空间融入度上SFIM融合法都要优于其他融合方法,SFIM既能提高空间分辨率,又很好的保留了光谱信息,有利于信息提取,对于GF-1号遥感影像来说SFIM是一种较好的融合方法。(3)遗传算法优化的支持向量机对支持向量机和遗传算法优化的支持向量机提取结果进行精度评价,支持向量机的总体精度83.79%,kappa系数0.7985,遗传算法优化的支持向量机的总体精度88.14%,kappa系数0.8527,两者总体精度相差4.35个百分点,kappa系数相差0.0542,提取时间基本一致。充分说明,遗传算法优化的支持向量机在湿地类型提取上的有效性,且提取精度明显提高。(4)Fisher判别法自动提取在GF-1号遥感影像湿地类型提取时间上,Fisher判别法收敛性大为改善,数据迭代次数明显减少,提取速度提升显著,提取结果只需要49秒。在大批量影像处理上Fisher判别法优势明显,既能满足了总体精度要求,也能大大缩短湿地类型提取时间。(5)湿地类型提取算法比较通过实验得出:总体精度上遗传算法优化的支持向量机与面向对象决策树最高为88.14%,其次是Fisher判别法总体精度85.17%,支持向量机总体精度最低83.79%;遗传算法优化的支持向量机Kappa系数最高0.8572,其次是面向对象决策树和Fisher判别法Kappa系数分别为0.8217、0.8129,支持向量机的Kappa系数最低0.7985,这充分说明遗传算法优化的支持向量机在湿地类型提取的总体精度优于Fisher判别法和支持向量机,Kappa系数优于其他三种提取方法,且改善明显。Fisher判别法的提取时间最短49秒,其次是面向对象决策树165秒,遗传算法优化的支持向量机和支持向量机提取时间分别为253秒、249秒,这说明Fisher判别法在湿地类型提取时间上优于其他三种分类方法,且提升显著。
[Abstract]:At present, the wetland resources as a kind of important natural resources and forests, marine ecosystem the same position, it changes and sustainable utilization is an important part of the scientific core of the earth. Monitoring become more and more important means in the wetland area by using remote sensing technology, can solve some scientific research directions, such as wetland type information wetland landscape, information, wetland changes. These studies can promote wetland ecological resources protection and development, in order to protect the government of wetland protection and protection area construction engineering science, the correct decision. On the East Dongting Lake as a case study area, using GF-1 image as data source for remote sensing data source. The pretreatment, then the optimal band combination analysis of optimal fusion method, and then the genetic algorithm and Fisher discriminant method is introduced into the wetland The type of extraction, a optimization algorithm of support vector machine to achieve high precision extraction of wetland types, on the other hand, using the Fisher discriminant method of wetland types were simple, efficient, automatic extraction; at the end of the four algorithms of the extraction results were evaluated to analyze optimal algorithm suitable for GF-1 remote sensing image extraction of wetland types, improve wetland the type of extraction system, provide the basis for future wetland remote sensing research. The main results are as follows: (1) the optimal band combination analysis considering the spectral features and the size of the information, the standard deviation of information entropy, the best judgment index 3 quantitative evaluation index and visual effect, determine the best combination for the band GF-1 remote sensing image RGB=432. (2) the best fusion methods of analysis and evaluation study on the effect of fusion using quantitative evaluation method, integration of the two aspects of the melt and the space from the spectrum of inheritance Moreover, the average correlation coefficient, entropy, standard deviation, gradient index of five principal component transform fusion, Gram-Schmidt fusion and quantitative evaluation of smoothing filter brightness adjustment based on the fusion results. The spectrum of inheritance and space integration SFIM fusion method is superior to other fusion methods, which can improve the SFIM the spatial resolution, and it is good to retain the spectral information, is conducive to the information extraction for GF-1 remote sensing image SFIM is a better fusion method. (3) support vector machine optimized by genetic algorithm of support vector machine and genetic algorithm support vector machine extraction results to evaluate the accuracy, the overall accuracy of 83.79% vector machine, kappa coefficient is 0.7985, the overall accuracy of 88.14% support vector machine optimized by genetic algorithm, the kappa coefficient is 0.8527, the overall accuracy is 4.35 percentage points, kappa coefficient is 0.0542, The extraction time is basically the same. Shows that the effectiveness of support vector machine optimized by genetic algorithm in wetland extraction, and the extraction accuracy is obviously improved. (4) Fisher discrimination method of automatic extraction in the extraction time of GF-1, the remote sensing image of wetland types, Fisher discriminant of convergence is greatly improved, the number of iterations of data significantly reduced the extraction rate significantly enhance the extraction results, only 49 seconds. In large quantities of image processing Fisher discriminant method has obvious advantages, which can not only satisfy the requirements of the overall accuracy, can greatly shorten the extraction time. Wetlands (5) wetland type extraction algorithm by comparing experiment results: the overall accuracy of support vector machine optimized by genetic algorithm with the object oriented decision tree up to 88.14%, followed by Fisher discriminant analysis, the overall accuracy is 85.17%, the overall accuracy of support vector machine minimum 83.79%; genetic algorithm support vector machine Kappa factor 0.8572, followed by object oriented decision tree and Fisher discriminant Kappa coefficients were 0.8217,0.8129, support vector machine Kappa the lowest coefficient of 0.7985, which fully shows that the support vector machine optimized by genetic algorithm method and support vector machine in overall accuracy is better than Fisher wetland extraction, the Kappa coefficient is better than the other three kinds of extraction methods. The extraction time and improved.Fisher discrimination method of the shortest 49 seconds, followed by object oriented decision tree for 165 seconds, genetic algorithm and support vector machine SVM extraction time was 253 seconds, 249 seconds, indicating that the extraction time Fisher discriminant method is superior to the other three kinds of classification methods in wetland types, and enhance significant.

【学位授予单位】：中南林业科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X87;X171
，

本文编号：1768767