SAR Image Despeckling Based On Low-rank Hankel Matrix

Synthetic Aperture Radar is active microwave radiation,which mainly based on the principle of Synthetic Aperture.SAR image is not affected by the day and the weather,also has good surface penetration.The advantages in application and research of ocean detection,crop yield estimation,military detection and other aspects,therefore become a modern remote sensing technology and important research hotspot.But the key problem is that the SAR imaging system is easy to generate speckle in the process of acquiring SAR image,which affects image visual effect and interferes with later processing.Therefore,it is of great significance to study SAR image despeckling.Aiming at the problem of poor despeckling and serious loss of detail information after despeckling,this thesis combined with the characteristics of SAR image and the non-local similarity inherent in low-rank Hankel matrix and proposed two algorithms based on low-rank Hankel matrix,which optimize despeckling effect.Two algorithms have different pertinence,one is aimed at the area with high homogeneity,the other is aimed at the area with more texture.In this thesis,the main work is as follows:(1)A despeckling algorithm based on low-rank Hankel matrix and alternate direction multiplier method is proposed.In the traditional despeckling algorithm,the SAR image is indiscriminately filtered,or the despeckling degree is insufficient.The main reason is that the SAR image’s internal structural similarity is ignored.Therefore,to solve this problem,a despeckling algorithm based on low-rank Hankel matrix and alternate direction multiplier method is proposed which mainly used the non-local similarity of low-rank Hankel structure.In the pre-processing stage,the logarithmic processing of the SAR image signal is firstly carried and converted to the Hankel matrix.Then,robust principal component decomposition and alternate direction multiplier method were used to despeckling.Finally,despeckling SAR image was obtained by inverse Hankel calculation.(2)A despeckling algorithm based on low-rank Hankel matrix and weighted kernel norm minimization is proposed.For the previous algorithm,it mainly applies to the smoother region,but for areas with rich texture information,its applicability is reduced and easily lose details.Therefore,the weight adaptive feature of the weighted kernel norm model is used to reduce details’ loss.After preprocessing with the above algorithm,block matching is used to find similar blocks and obtained all similar blocks then the weight of the weighted kernel norm is initialized and minimized the solution.Finally,image blocks were collected to perform inverse Hankel operation of the matrix to obtain despeckling image.Since the weight of the algorithm is not uniform,it can adaptively update the weight value according to the noise level.Therefore,it is more suitable for areas with rich details,such as mountains and valleys.Experimental results show that the proposed algorithm can effectively retain details compared with other algorithms and achieving a certain degree of speckle suppression effect.