| SAR has many characteristics, such as all-day, all-weather, strong penetrating power and under the condition of bad environment also can obtain high resolution image, it’s widely used in aerospace, industrial inspection, medical services and other civil and military. However, compared other SAR imaging systems with coherent imaging systems, the image is obtained by the existence random distribution of the large number of speckle noise. The existence of the speckle noise of SAR image influences the visual effect, and has seriously restricted in SAR image interpretation work, such as the subsequent feature extraction and target tracking. Therefore, inhibition the speckle noise of SAR image research has the important significance. In this paper, a new generation of multi-polarization imaging mode and multiple receive mode of RadarSat-2 images, based on the five typical feature of Shuozhou City in Shanxi Province, such as buildings, mountains, water, roads and farmland, the filtering test is carried out, get some results. The main results are as follows:(1)This paper summarizes speckle reduction methods and research status of SAR images; studies on imaging mechanism of SAR image, speckle noise generation mechanism and mathematical model; summarizes typical filtering algorithms and the quality evaluation index of image.(2) This paper uses improved non-local means filtering algorithm. To solve the problem of the distribution weight of Gauss’s Euclidean distance, the new algorithm uses a cosine function to reduce the decay rate of Gaussian kernel, so that the weight can distribution rationalizationly, through neighborhood similar function to improve weight function, it’s improves neighborhood similarity. The test results show that the improved non-local mean filter has better de-noising effect, but seriously loss of edge information, so that the result image is blurring.(3)This paper uses the Lee combined with the DCT algorithm. This algorithm makes use of the Lee filter to deal with the homogeneous regions of the image with strong smoothing in homogeneous region. The DCT algorithm can be used to deal with the heterogeneous regions by using the advantages of speckle noise and edge area which can be distinguished. Finally, the reconstructed image is obtained of SAR speckle suppression. The test results show that the algorithm has better ability to smooth, EPI is from 0.8 to 0.9, SSIM is about 0.25.(4)The filtering algorithm based on non-local Mahalanobis distance is used. Aiming at the problem of correlation between data samples and neighborhood image block structure similarity, it is proposed that the weight function is improved by the combination of the Mahalanobis distance and the structure similarity measure, which strengthens the effect of the algorithm in the edge area. The test results show that the algorithm of smoothing ability is higher than Lee combined with DCT algorithms, lower than the non-local means filtering, EPI will reach around 0.6, SSIM is at least 0.2, and the edge is clearly visible, the ability of de-noising and edge holding ability both have better results.(5)The overall test results show that the filtering de-noising effect of ground object from strong to weak is farmland, mountains, water, roads, buildings, the retaining structural information of ground object from big to small is that buildings, roads, water, mountains and farmland. De-noising filtering algorithm still has the problem of the ability of de-noising and edge preserving can’t balance. For a specific feature, de-noising effect is vegetation mountains higher than that of without vegetation; A mixture of water is greater than pure water; The railway is below the highway; There is no vegetation of farmland is higher than the vegetation of farmland; Bungalow is greater than the tall buildings, the ability of edge preservation is greater than tall, but structural information retention capacity is lower than the tall buildings. |
PDF Full Text Request