SAR图像乘性噪声模型分析

Analysis of Multiplicative Noise Models in SAR Imagery

从SAR相干成像的物理散射机制出发,SAR乘性噪声模型认为SAR图像的每个分辨单元强度是由该单元中地物的RCS被一个强度服从单位均值(均值为1)指数分布的乘性噪声调制而成。即可以认为SAR图像是场景中地物的RCS和单位均值指数强度分布噪声的乘积。利用实测SAR图像数据库,首先证实了SAR图像中乘性噪声模型比加性噪声模型更合理。然后,在每个目标/姿态区间提取图像模板,估计每个目标切片对应的乘性噪声,对每类目标的乘性噪声分布进行直方图拟合,并采用拟合优度检验评估了拟合精度,结果表明:乘性噪声确实能用单位均值指数分布较精确描述,这也进一步证实了乘性噪声模型的正确性。最后,给出了分辨率变化(9.6m～0.3m不同)和视角变化(两种不同的视角)时单位均值指数分布对SAR乘性噪声直方图的拟合实验,结果表明单位均值指数分布对于分辨率参数和视角参数变化的情况下都是SAR乘性噪声的较精确模型。

On the basis of the physical scattering concepts of coherent radiation of SAR, the observed intensity at each resolution cell of the SAR images is formulated by the multiplicative noise model as a deterministic RCS modulated a unit mean exponential distributed multiplicative noise. Namely,the observed SAR scene can be regarded as the product of an underlying terrain RCS term and a unit mean exponential distributed noise term. Utilizing the actual SAR database, the validation of the multiplicative model than that of the additive model is firstly verified. Then, we extract the template in each target/pose interval and remove the template from the chip to analyze the residual noise. By using the test statistics to measure the goodness that the histogram for the noise residual fits different distributions, the results show that the unit mean exponential distribution fits the data relatively precisely, which further testifies the correction of the muhiplicative noise model. Finally, some experiments that the histogram for the noise residual fits the unit mean exponential distribution are presented Under different resolutions （9.6m - 0.3m） and depression angels （two angels）, the results show unit mean exponential distribution are all a relatively precise model of the multiplicative noise in SAR images despite the variety of resolutions and depression angles.