基于商空间粒度计算的SAR图像分类

SAR Image Classification Based on Granularity Computing of Quotient Space Theory

SAR图像的分类是实现SAR图像自动理解与解译的关键步骤,而纹理是各种地表的固有属性,为SAR图像的分类提供了大量有用的信息,尤其对于单波段、单极化的SAR图像,纹理信息就显得格外重要.然而,不同纹理特征对SAR图像中不同地表结构内在属性的刻画能力并不一致.如何将不同纹理特征结合起来,以获得应用范围更广且分类效果更好的SAR图像分类方法,是当前SAR图像处理研究中的一个热点问题.文章将商空间粒度计算引入SAR图像的分类中,结合SAR图像特性,提出了一种基于粒度合成理论的SAR图像分类方法.该方法首先利用具有良好推广能力的支撑矢量机基于不同纹理特征获得SAR图像的不同分类结果,并认为这些分类结果构成不同的商空间,再根据粒度合成理论将这些商空间组织起来得到SAR图像的最终分类结果.实验结果验证了这种方法的有效性和正确性以及商空间的粒度计算在SAR图像分析中的应用潜力.

Classification of SAR image is the key process for SAR image perception. Texture is an innate property of all surfaces, and it can provide abundant useful information for classification of SAR image, especially for single-band and single-polarized SAR images. However, different texture features will give different characterization of different land covers in SAR image. How to efficiently combine different textural features in order to get high performance is the study focus of the SAR image classification currently. In this paper, combining the SAR imaging mechanism, the theory of granular computing of quotient space based SAR image classification algorithm is proposed. Such algorithm uses the support vector machine （SVM） with better generalization ability to get different classification results based on different textural features, and construct different quotient space by using these results. The final SAR classification result is got by granularity synthesis method to organize these quotient spaces. The experiment results on SAR images show the validity and the effectiveness of the new algorithm. In addition, the potency of the granular computing of quotient space in the application of SAR image analysis and processing is demonstrated also.