Superpixel Segmentation Method For High Resolution PolSAR Images

Polarimetric synthetic aperture radar(PolSAR)transmits and receives electromagnetic waves by the polarimetric combined antenna.Rich scattering information can be obtained from the receiving electromagnetic waves.Compared to unipolar or dual-polarimetric SAR systems,PolSAR system is able to measure more information about the objects,that is,PolSAR data can describe ground objects more accurately and truly.With the increase of PolSAR image resolution,the high-resolution images can provide more texture information,structural information,but there are some problems with interpretation of PolSAR image by pixel-based method: PolSAR image has serious speckle noise,which makes many pixels be noise points whose categories is uncertain.Using pixel-based method to interpret PolSAR image will cause the noise points to be processed wrongly.In high resolution PolSAR image,adjacent pixels are likely to be the same ground object,so the information contained in these pixels is basically similar.If these pixels are processed alone,it will cause information redundancy.In order to solve the problems caused by high resolution PolSAR image,the concept of superpixel was introduced.Superpixel is the uniform image blocks which are composed of high similar pixels in shape,color,texture and so on.In high resolution image,the amount of data is large.The number of pixels is generally hundreds or even tens of millions.After superpixel segmentation,it becomes to process hundreds of superpixels in the subsequent processing.The concept of superpixel is proposed in the optical image.Combined with the polarimetric characteristic of PolSAR image,superpixel segmentation algorithm is applied to PolSAR image.The main contents are as follows:Firstly,the paper studies the basic theory of PolSAR system,the represent of polarimetric data,and the mutual conversion relationship between multiple representations.In order to extract rich polarimetric information and texture information from PolSAR image,the paper studies the objective decomposition method and texture feature extraction methods commonly used.Then,the basic principle of superpixel is introduced,as well as the traditional SLIC algorithm and the traditional watershed algorithm which are improved combined with the characteristics of PolSAR image.Finally,the evaluation criteria of the superpixel segmentation are studied.The experiment is carried out by two experimental data.The results are qualitatively and quantitatively evaluated.