基于组合差异图和卷积小波神经网络的SAR图像变化检测

Change detection in SAR images based on combined difference image and convolution wavelet neural networks

合成孔径雷达(Synthetic aperture radar,SAR)图像变化检测可以看作是一个分类过程,像素被分类为变化类和不变类。但是,差异图的质量影响了现有方法的检测精度。为了提供高质量的差异图,提出一种基于组合差异图和卷积小波神经网络(Convolutional-Wavelet Neural Network,CWNN)结合的SAR图像变化检测方法。首先,使用对数比算子、均值比算子和差分算子产生差异图,并使用简单的组合方法得到最终的差异图。然后,在差异图上使用分层模糊C均值获得变化类和不变化类的训练样本。最后,使用训练好的CWNN对所有像素进行分类。该方法提高了差异图的质量,为网络提供高质量的训练样本。真实SAR图像数据集上的实验结果表明,该方法提高了变化检测精度。

Synthetic aperture radar(SAR)images change detection can be thought as a classification process,with pixels classified as change classes and invariant classes.However,the quality of difference image affects the detection accuracy of existing methods.To provide a high-quality difference image,we proposed a change detection method based on combined difference image and convolution wavelet neural networks(CWNN).Firstly,the difference images were produced using logarithmic ratio operators,mean ratio operators,and difference operators,and the final difference image was obtained using a simple combination method.Then,hierarchical fuzzy C mean was used on the difference image to obtain training samples.Finally,all pixels were classified using the trained CWNN.This method improves the quality of difference graph and provides high-quality training samples for the networks.