基于资源3号影像的阳澄湖围网区自动提取算法研究

发布时间：2019-01-12 09:15

【摘要】：随着中国水产渔业的快速发展,长江流域的浅水湖泊围网养殖业极速扩张,极大地增加了经济效益,但超高密度的围网养殖也对湖泊生态造成了严重破坏。在此背景下,合理科学地规划并控制湖泊围网区已被湖泊管理部门列为湖泊水质提升和优化的重点工作,而获取并掌握湖泊围网养殖的时空分布信息是湖泊生态管理部门在科学管理围网养殖区时制定措施的依据。由于遥感技术具有实时、大范围和客观等优势,遥感监测逐渐成为获取湖泊围网养殖区空间分布信息的主流方法。而目前对湖泊围网养殖区的遥感识别方法主要是人工解译湖泊的遥感影像,解译时间长,受人工影响大。为进一步提高解译效率,部分学者开始研究机器解译围网区的方法,但大多数方法只使用遥感图像的纹理特征,且识别时需要人为确定特征阈值,方法自动化程度低。本文以长江流域典型的围网型湖泊——阳澄湖为研究区,基于高分辨率的资源3号卫星(ZY-3)影像,运用模式识别理论,设计了围网养殖区自动识别方案。该方案主要包括遥感影像预处理,边界信息增强、图像噪声滤除、阈值分割、特征提取、分类识别6个步骤。主要研究内容如下:(1)基于图像处理的影像增强和分割算法。首先,分析了围网边界增强的必要性,并利用梯度变换和灰度变换进行围网区边界信息增强,结果表明,该方法有效的增强了围网区的边界信息;然后,描述了自适应滤波算法的滤波原理和滤波流程,并运用自适应滤波算法滤除了遥感影像中的椒盐噪声和高斯噪声;最后,研究了阈值分割的基本原理,使用了小波变换精确寻找到进行阈值分割的阈值,结果表明,当阈值为130和180时,图像能很好的完成阈值分割任务。(2)基于傅里叶变换的围网区特征提取。首先,通过分析围网区的光谱信息,明确了阳澄湖围网区和水体最具可分性的波段是近红外波段;然后,基于资源3号阳澄湖影像的近红外波段,利用二维离散傅里叶变换对湖泊遥感影像中的每一小块区域进行变换,选取了以直流信号为中心的频率谱特征;最后对二维频率谱进行奇异值分解,并将得到的奇异值重新组成特征向量,达到对原特征向量降维的目的。(3)基于最近邻分类的围网区识别。首先,通过分析不同分类方法的优缺点,明确了最近邻分类法作为阳澄湖围网区识别方法;然后研究了最近邻分类法的基本原理,分析了最近邻分类法的基本操作步骤,并使用最近邻分类法对阳澄湖围网养殖区进行了分类识别;最后,使用混淆矩阵,以人工目视解译的围网区为评价标准对算法的分类结果进行了评价,提取精度为86%。
[Abstract]:With the rapid development of aquatic fishery in China, the shallow water lake seine aquaculture industry in the Yangtze River valley has expanded extremely rapidly, which has greatly increased the economic benefits, but the super-high density seine culture has also caused serious damage to the lake ecology. In this context, the rational and scientific planning and control of the lake seine area has been listed as the key task of lake water quality improvement and optimization by the lake management department. Therefore, obtaining and mastering the spatial and temporal distribution information of lake seine culture is the basis for the lake ecological management department to formulate measures in the scientific management of seine culture area. Because remote sensing technology has the advantages of real-time, large-scale and objective, remote sensing monitoring has gradually become the mainstream method to obtain spatial distribution information in lake seine culture area. At present, the method of remote sensing recognition of lake seine culture area is mainly artificial interpretation of lake remote sensing image, interpretation time is long, and is greatly affected by artificial. In order to further improve the interpretation efficiency, some scholars have begun to study the method of machine interpretation of the seine area, but most of the methods only use the texture features of remote sensing images, and the recognition needs to determine the threshold of features artificially, so the automation of the method is low. In this paper, Yangcheng Lake, a typical seine lake in the Yangtze River Basin, is used as the study area. Based on the high resolution ZY-3 image, the automatic identification scheme of the seine culture area is designed by using the pattern recognition theory. The scheme mainly includes six steps: remote sensing image preprocessing, edge information enhancement, image noise filtering, threshold segmentation, feature extraction and classification and recognition. The main contents are as follows: (1) Image enhancement and segmentation algorithm based on image processing. Firstly, the necessity of the enhancement of the seine boundary is analyzed, and the edge information of the seine is enhanced by gradient transformation and gray transformation. The results show that the method can effectively enhance the boundary information of the seine. Then, the filtering principle and filtering flow of adaptive filtering algorithm are described, and the pepper and salt noise and Gao Si noise in remote sensing image are filtered by adaptive filtering algorithm. Finally, the basic principle of threshold segmentation is studied, and the threshold of threshold segmentation is accurately found by wavelet transform. The results show that when the threshold is 130 and 180, The image can accomplish the task of threshold segmentation well. (2) the feature extraction of seine area based on Fourier transform. Firstly, by analyzing the spectral information of the seine area, it is clear that the most divisible band between Yangcheng Lake seine area and water body is near infrared band. Then, based on the near infrared band of Yangcheng Lake image of Resource-3, every small region of lake remote sensing image is transformed by two-dimensional discrete Fourier transform, and the frequency spectrum characteristic centered on DC signal is selected. Finally, the singular value of the two-dimensional frequency spectrum is decomposed, and the singular value is recomposed into the eigenvector to reduce the dimension of the original eigenvector. (3) the identification of the seine area based on the nearest neighbor classification. Firstly, by analyzing the advantages and disadvantages of different classification methods, the nearest neighbor classification method is defined as the identification method of Yangcheng Lake seine area. Then the basic principle of nearest neighbor classification is studied, the basic operation steps of nearest neighbor classification are analyzed, and the most nearest neighbor classification method is used to classify and identify the seine culture area of Yangcheng Lake. Finally, the classification results of the algorithm are evaluated by using the confusion matrix and the artificial visual interpretation of the seine area as the evaluation criteria. The extraction accuracy is 86%.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】