基于高分辨率遥感影像纹理特征的面向对象植被分类方法研究

发布时间：2018-06-07 05:13

本文选题：纹理特征提取 + 指纹识别技术　；参考：《云南师范大学》2017年硕士论文

【摘要】：植被是覆盖地表的植物群落总称,是生态系统重要的组成部分。植被分类是遥感应用研究的一个热点问题,高分辨率遥感影像具有丰富的纹理信息,可以有效改善植被分类精度。纹理特征提取是高分遥感图像分类应用中的关键技术之一,现有的纹理特征提取方法普遍存在准确分类率低、计算复杂以及效率低等缺点。本研究以云南省西双版纳州纳板河流域为例,分析高分辨率遥感影像植被纹理特征,提出一种基于指纹识别技术的植被纹理特征提取方法,并辅以纹理特征进行面向对象植被分类,分析纹理特征对面向对象植被分类精度的影响。研究成果总结如下:(1)提出并实现一种基于指纹识别技术的植被纹理特征提取方法基于指纹识别技术提出指纹纹理增强算法,将指纹纹理增强算法分别与灰度共生矩阵和局部二值模型算法相结合实现了基于指纹识别技术的纹理特征提取方法。以云南省西双版纳州纳板河流域的WorldView-2及Pléiades影像为实验数据,采用本文算法提取影像的纹理特征,并与基于RGB影像提取的纹理特征作对比分析。在WorldView-2数据实验中,相比加入基于RGB影像提取的纹理特征的分类结果,加入基于指纹识别技术提取的GLCM纹理特征的分类总体精度达到了91.56%,提高了4.10%,Kappa系数达到了0.90,提高了0.05,加入基于指纹识别技术提取的LBP纹理特征的分类总体精度达到了89.36%,提高了3.41%,Kappa系数达到了0.87,提高了0.04。在Pléiades数据实验中,采用所提出的纹理特征提取方法提取影像的纹理特征,并辅以纹理特征进行面向对象植被分类。相比加入基于RGB影像提取的纹理特征的分类结果,加入基于指纹识别技术提取的GLCM纹理特征的分类总体精度达到了88.60%,提高了2.91%,Kappa系数达到了0.86,提高了0.04。加入基于指纹识别技术提取的LBP纹理特征的分类总体精度达到了84.60%,提高了3.04%,Kappa系数达到了0.81,提高了0.04。对各个纹理特征提取算法采用不同的高分数据的分类结果表明:基于指纹识别技术纹理特征提取方法提取的纹理特征可在很大程度上改善纹理规则地类的分类精度。(2)纹理特征可以明显改善高分辨遥感影像面向对象植被分类精度将提取的纹理特征加入到面向对象植被分类中,与未利用纹理特征的面向对象植被分类结果作对比分析。WorldView-2影像试验中,采用基于RGB影像提取的GLCM纹理特征的总体分类精度为87.46%,提高了7.05%,Kappa系数为0.85,提高了0.09;加入基于RGB影像提取的LBP纹理特征的总体分类精度为85.95%,提高了5.44%,Kappa系数为0.83,提高了0.07;加入基于指纹识别技术纹理特征提取方法提取的GLCM和LBP纹理特征的总体分类精度分别提高了11.15%和8.95%,Kappa系数分别提高了0.14和0.11。在Pléiades影像试验中,加入纹理特征后的面向对象植被分类精度也显著提高。结果表明:运用纹理特征的面向对象植被分类可以显著提高高分辨率遥感影像的植被分类精度。(3)实现基于单一数据源提取多分类特征的面向对象植被精细分类本文基于单一数据源提取影像对象的光谱特征、纹理特征、植被指数特征以及几何特征等植被识别特征,对研究区自然林、橡胶林、香蕉、茶园以及耕地的面向对象分类结果中,自然林分类精度为95.43%,橡胶林分类精度达到94.33%,香蕉分类精度高达93.60%,茶园及耕地的分类精度均达到了83.00%以上。总体来说分类精度较高,实现了面向对象方法框架下基于单一数据源提取多分类特征的植被精细分类。
[Abstract]:Vegetation is the general name of plant community covering the surface, and it is an important part of the ecosystem. Vegetation classification is a hot issue in remote sensing application research. High resolution remote sensing images have rich texture information and can effectively improve the classification accuracy of vegetation. The extraction of texture features is one of the key technologies in the application of high-resolution remote sensing image classification. The existing texture feature extraction methods have the disadvantages of low accurate classification rate, complex calculation and low efficiency. This study took the Nanban River Basin in Xishuangbanna, Yunnan Province as an example, to analyze the vegetation texture features of high resolution remote sensing images, and put forward a method of vegetation texture feature extraction based on fingerprint recognition technology, supplemented with texture special. The results are summarized as follows: (1) a fingerprint recognition method based on fingerprint recognition technology is proposed and implemented based on fingerprint recognition technology to enhance the fingerprint texture enhancement algorithm and the fingerprint texture enhancement algorithm and gray scale respectively. Combining the symbiotic matrix with the local two value model algorithm, the texture feature extraction method based on fingerprint recognition technology is realized. The WorldView-2 and Pl e iades images of the Nanban River Basin in Xishuangbanna, Yunnan province are taken as experimental data, and the texture features of the images are extracted by this algorithm, and the scores are compared with the texture features based on the RGB images. In the WorldView-2 data experiment, compared with the classification results of texture features extracted based on RGB images, the overall accuracy of the classification of GLCM texture features extracted with the fingerprint recognition technology is 91.56%, 4.10%, the Kappa coefficient is 0.90, and 0.05 is increased, and the LBP texture feature extracted based on fingerprint recognition technology is added. The overall accuracy of the classification has reached 89.36%, increased by 3.41%, and the Kappa coefficient reached 0.87. In the Pl e iades data experiment, 0.04. was used to extract the texture features of the image with the proposed texture feature extraction method, and the texture features were used to classify the object oriented vegetation. The classification of the texture features based on the RGB image extraction was added to the classification. As a result, the overall accuracy of the classification of GLCM texture features extracted with fingerprint recognition technology reached 88.60%, increased by 2.91%, and the coefficient of Kappa reached 0.86. The overall accuracy of the classification of LBP texture features extracted by 0.04. based on fingerprint recognition technology was 84.60%, 3.04%, Kappa coefficient reached 0.81, and 0.04. increased. The classification results of various texture feature extraction algorithms using different high score data show that texture feature extraction based on fingerprint recognition technology can improve the classification accuracy of texture classification. (2) texture features can obviously improve the accuracy of object-oriented vegetation classification in high resolution remote sensing images. The extracted texture features are added to the object-oriented vegetation classification. Compared with the object-oriented vegetation classification results that are not used for texture features, the overall classification accuracy of the GLCM texture features based on RGB images is 87.46%, increased by 7.05%, the Kappa coefficient is 0.85, and 0.09, and the addition of RGB is based on RGB. The overall classification accuracy of the LBP texture features extracted by the image is 85.95%, 5.44%, the Kappa coefficient 0.83, and 0.07. The overall classification accuracy of the GLCM and LBP texture features extracted from the texture feature extraction method based on fingerprint recognition technology is increased by 11.15% and 8.95% respectively, and the Kappa coefficient is increased by 0.14 and 0.11. respectively in Pl e iades shadow. In the experiment, the accuracy of the object-oriented vegetation classification after adding texture features has also been improved significantly. The results show that the classification accuracy of the vegetation can be significantly improved by the object-oriented vegetation classification using the texture features. (3) to realize the fine classification of the object oriented vegetation based on the multiple classification characteristics based on the single data source. The classification accuracy of natural forest, rubber forest, banana, tea garden and cultivated land is 95.43%, the classification precision of the rubber forest is 94.33%, and the precision of banana classification is high. Up to 93.60%, the classification accuracy of tea garden and cultivated land has reached over 83%. In general, the precision of classification is high, and the fine classification of Vegetation Based on the multi classification characteristics based on the single data source is realized under the framework of object-oriented method.
【学位授予单位】：云南师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：Q949;P237

【参考文献】