一种联合特征降维的极化SAR影像分类方法

A classification method for PolSAR combined with feature dimensionality reduction

针对PolSAR影像极化信息利用不完全,分类效果有待进一步提高的问题,该文在综合多源极化信息的基础上,提出一种核熵成分分析算法(KECA)特征降维,及BA-SSVM模型训练的PolSAR影像分类方法。构造PolSAR多源目标分解及纹理特征组合,根据熵值贡献率利用KECA开展特征降维,之后采用蝙蝠算法(BA)对光滑支持向量机(SSVM)模型参数自动寻优,实现PolSAR影像分类。通过Flevoland地区的AIRSAR影像及北京地区的Radarsat-2影像的分类实验,验证了该文方法的有效性。在特征降维方面,KECA比传统KPCA算法表现出更好的特征融合效果和非线性适应性;利用BA对SSVM参数进行智能解算,也可有效解决盲搜索问题,提高模型训练精度;通过KECA降维及BA-SSVM智能模型训练,分类效果总体优于传统方法。

Aiming at the problem that PolSAR image classification polarization information was not fully utilized and the classification effect needed to be further improved,this paper proposed a new PolSAR image classification method based on kernel entropy component analysis(KECA)feature dimensionality reduction and BA-SSVM model training.The multi-source target decomposition and texture feature combination of PolSAR were constructed.Then,according to the entropy contribution rate,KECA was used to carry out feature dimension reduction.Finally,Bat algorithm(BA)was used to automatically optimize the parameters of smooth support vector machine(SSVM)model to realize PolSAR image classification.The effectiveness of this method was verified by the classification experiments of AIRSAR image in Flevoland area and radarsat-2 image in Beijing area.In feature dimensionality reduction,KECA showed better feature fusion effect and nonlinear adaptability than traditional KPCA;The intelligent solution of SSVM parameters by BA could also effectively solve the problem of blind search and improved the accuracy of model training;Through KECA feature dimensionality reduction and BA-SSVM intelligent model training,the classification effect was better than the traditional methods.