论文部分内容阅读
为实现合成孔径雷达(SAR)影像分割中类别数的自动确定,提出一种基于贝叶斯信息准则(BIC)的可变类SAR影像分割算法.该算法以Gamma分布建模SAR影像同质区域内部像素光谱测度的统计分布特性;结合BIC准则构建整幅SAR影像似然函数模型;并在此模型中引入类别数补偿项,继而提高BIC测度对影像分割结果的描述精度.采用期望条件最大化(ECM)算法估计其模型参数;通过遍历所有可能类别数,取最小BIC值对应的类别数作为最佳类别数.采用提出的算法分割模拟和真实SAR影像,模拟SAR影像分割结果的定性和定量分析表明,基于BIC准则的ECM算法可以实现类别数的自动确定,并可得到最优分割结果.通过对真实SAR影像分割结果的定性评价,进而证明了可变类SAR影像分割算法的准确性和可行性.
In order to automatically determine the number of classes in synthetic aperture radar (SAR) image segmentation, a Bayesian Information Criterion (BIC) -based variable SAR image segmentation algorithm is proposed, which uses the Gamma distribution to model homogeneous regions of SAR images The statistical distribution of the internal pixel spectral measure was constructed. The whole SAR image likelihood function model was constructed by combining with the BIC criterion. The classification number compensation term was introduced into the model to improve the description accuracy of the BIC measurement to the image segmentation result. (ECM) algorithm is used to estimate the model parameters. By traversing all possible categories, the number of categories corresponding to the minimum BIC value is taken as the optimal category number. The proposed algorithm is used to segment the simulated and real SAR images to simulate the qualitative and quantitative results of SAR image segmentation The analysis shows that the ECM algorithm based on BIC criterion can automatically determine the number of classes and obtain the optimal segmentation results.According to the qualitative evaluation of the real SAR image segmentation results, the accuracy of the variable SAR image segmentation algorithm and feasibility.