特征提取与主成份分析的合成孔径雷达图像的分类方法研究

ON CLASSIFICATION METHOD OF SYNTHETIC APERTURE RADAR IMAGE BASED ON FEATURE EXTRACTION AND PRINCIPAL COMPONENT ANALYSIS

合成孔径雷达的图像分类,在军事和民事上具有重要意义。为了更好地实现分类,提出一种基于特征提取与主成份分析的方法(FEPCA)。特征集包括图像的Pauli特征、相干矩阵特征、小波特征、纹理特征。由于原始特征数目过多,因此采用主成份分析压缩特征维数。最后,选择人工神经网络作为默认分类器。实验部分选择WML算法与HA算法作为对比算法,对San Francisco的实测数据与地面真实场景显示,所提出的FEPCA算法为95.94%精确度,高于WML的94.82%与HA的95.73%精确度。因此,该算法优于WML与HA算法。

The classification of synthetic aperture radar images is of significance in military and civil affairs. In order to classify better, we propose a novel method which is based on feature extraction and principle component analysis （FEPCA）. The feature set includes the Pauli features, correlated matrix features, wavelet features and texture features. Since there are too many original features, therefore the principal component analysis is used to compress the character dimension. At last, we choose artificial neural network as the default classifier. In ex- periment part we pick up WML algorithm and HA algorithm as the comparison algorithms. The measured data in and real ground scenario of San Francisco demonstrate that the proposed FEPCA algorithm achieves 95.94% classification accuracy, higher than the that of 94.85% of WML algorithm and 95.73% of HA algorithm. Therefore, our FEPCA algorithm outperforms WML and HA.