基于小波纹理与改进FCM的SAR机场类目标提取

Airport Targets Extraction from SAR Imagery Based on Wavelet Texture and Improved Fuzzy Cmeans Clustering

摘要:

定义了几种小波纹理，提出一种基于小波纹理与改进FCM（模糊C均值聚类）的对SAR（合成孔径雷达）图像进行机场类目标自动提取方法.利用小波多尺度分析对影像分解，得到不同方向上小波频带，然后对不同的小波频带进行小波纹理分析——用频率共生矩阵4个关键特征来描述小波纹理特征向量.应用PCA（主成分分析）的方法去相关，进行特征空间的降维处理，并使用改进FCM进行分割，进而提取出目标.实验结果表明：此种无需初始点的非监督算法自动化程度高，机场类目标提取成功率和定位准确率分别达到97.3%和99.3%，为此类目标的定位

Abstract:

The paper presents definitions of several wavelet textures and a novel automatic airport targets extraction method, an unsupervised method without seed point and training areas, from Synthetic Aperture Radar (SAR) imagery. Wavelet channels at different directions through multiscale wavelet transformation were obtained. Then the four crucial frequency cooccurrence matrix features for the textures were computed in each channel of wavelet to get the eigenvectors of the wavelet textures.Principal component analysis (PCA) was adopted to eliminate the correlation of features; the feature reduction was successful and the segmentation performance could be maintained.Finally followed by implementing the improved fuzzy cmeans clustering method to the feature space, the desirable airports extraction was achieved. Experiments show that with the method,the extracting and the location precision are 97.3% and 99.3% respectively.This study provides a good reference for the location of airports in the SAR imagery.