基于MSRC的遥感影像面向对象分类研究

发布时间：2019-04-18 11:33

【摘要】：遥感影像数据具有覆盖范围大、信息客观真实、成本低、获取方便等优点，已经在各个领域得到了广泛应用。对于交通行业，，利用高分辨率遥感技术，并结合已有的交通信息采集手段，可以为城市交通监测、路网规划与建设、交通路网运行状态判别、行业管理、领导决策等综合决策服务提供有效的技术手段。但随着遥感影像分辨率的不断提高，地物信息的提取技术发展相对滞后，遥感技术在交通行业的应用仍处于初级阶段。高分辨率遥感影像相对于中、低分辨率遥感影像具有更为丰富的空间信息、纹理信息和地物几何信息。随着遥感技术的不断发展，遥感影像的分辨率的不断提高，影像数据的信息提取和分类技术面临着新的问题和挑战。传统的基于像元的分类方法由于其分类精度的限制已经不能满足遥感技术发展的需求，因此面向对象的分类技术应运而生。面向对象分类技术在处理遥感影像时，最小信息提取单元不再是单个像元，而是光谱和纹理特征相似的“均质对象”，因此可以充分利用包含光谱特征在内的其他结构信息，大幅度地提高分类精度与效率。在面向对象分类技术的基础上，①采用标记分水岭算法获取影像对象。提出了一种溢水模型，并用该模型修改标记产生方式，同时利用边缘检测手段，提高微弱边缘的提取能力且极大的抑制过分割现象；②对获得的区域对象提取包括光谱信息、纹理信息、边缘几何信息等特征，通过不同的特征组合分类实验，分析出不同特征对各类地物的分类效果，得出最佳的特征组合；③利用自适应权重的多重稀疏表示分类算法（Multiple Sparse Representation Classification approach,MSRC）自适应地调整各特征的权重，并获得最终分类结果。定量和定性的实验结果对比分析可以看出，基于MSRC的面向对象分类方法能更充分的利用不同特征的协同作用，整体分类精度和Kappa系数得到了明显提高，取得了较好的分类结果。
[Abstract]:Remote sensing image data has been widely used in various fields because of its advantages such as large coverage, objective and real information, low cost, convenient access and so on. For the traffic industry, the use of high-resolution remote sensing technology, combined with the existing means of traffic information collection, can be used for urban traffic monitoring, road network planning and construction, traffic network operational status discrimination, industry management, Comprehensive decision-making services, such as leadership decision-making, provide effective technical means. However, with the continuous improvement of remote sensing image resolution, the development of feature information extraction technology lags behind, and the application of remote sensing technology in traffic industry is still in the initial stage. Compared with middle and low resolution remote sensing images, high resolution remote sensing images have more abundant spatial information, texture information and geometric information. With the development of remote sensing technology and the improvement of remote sensing image resolution, the information extraction and classification technology of image data is facing new problems and challenges. The traditional pixel-based classification method has been unable to meet the needs of the development of remote sensing technology because of the limitation of its classification accuracy, so object-oriented classification technology emerges as the times require. Object-oriented classification technology in processing remote sensing images, the minimum information extraction unit is no longer a single pixel, but a "homogeneous object" with similar spectral and texture features, so other structural information, including spectral features, can be fully utilized. The accuracy and efficiency of classification are greatly improved. On the basis of object-oriented classification (OO), the image object is obtained by using the marked watershed algorithm. In this paper, a overflow model is proposed, and the label generation mode is modified by this model. At the same time, the edge detection method is used to improve the ability of weak edge extraction and greatly restrain the over-segmentation phenomenon. (2) feature extraction includes spectral information, texture information, edge geometry information and so on. Through different feature combination classification experiments, the classification effect of different features on all kinds of ground objects is analyzed, and the best feature combination is obtained. (3) the weight of each feature is adjusted adaptively by (Multiple Sparse Representation Classification approach,MSRC (multiple sparse representation algorithm of adaptive weight), and the final classification result is obtained. The comparative analysis of quantitative and qualitative experimental results shows that the object-oriented classification method based on MSRC can make full use of the synergetic effect of different features, and the overall classification accuracy and Kappa coefficient have been significantly improved, and better classification results have been obtained.
【学位授予单位】：湖南工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】