基于改进栈式核稀疏深度模型的大规模极化SAR地物分类算法

LARGE-SCALE POLSAR TERRAIN CLASSIFICATION BASED ON IMPROVED STACKED KERNEL SPARSE DEEP MODEL

地物分类是极化SAR应用的一个重要分支。传统的地物分类方法需要提取特征,通过分类器进行分类。在栈式稀疏自编码模型的基础上,提出一种鲁棒的极化SAR地物分类算法。采用基于Morlet小波核的最小二乘支撑向量机代替深度模型中常用的Softmax分类器。通过与栈式稀疏自编码网络相结合,在一定程度上克服了传统极化SAR影像地物分类方法受相干斑噪声影响,且结果过于粗糙的缺点,保证了分类结果中非匀质区域的连贯性和匀质区域的一致性。真实极化SAR数据仿真实验结果表明,该算法可以有效地提高分类精度,降低相干斑噪声的对分类精度的影响。

Terrain classification is an important branch of polarization SAR(PolSAR)applications.Traditional terrain classification methods require the extraction of features and then classification by classifiers.On the basis of the stack sparse auto-encoding model,we proposed a robust PolSAR terrain classification algorithm.We used Morlet wavelet kernel-based least squares support vector machine to replace the commonly used Softmax classifier in the deep learning model.Combining with stack sparse AutoEncoder network,it overcame the shortcomings of traditional PolSAR image classification methods,which was greatly affected by speckle noise and the result was too rough.It guarantees the consistency of inhomogeneous regions in classification results and the coherence of homogeneous regions.The simulation results of real PolSAR data show that the algorithm can effectively improve the classification accuracy and reduce the influence of speckle noise on the classification accuracy.