基于深度学习和同生矩阵的SAR图像纹理特征检索方法

SAR Image Feature Retrieval Method Based on Deep Learning and Synchronic Matrix

由于合成孔径雷达图像(SAR)存在相干斑噪声,采用传统的SAR图像解译工作相当复杂,且传统SAR图像检索方法获得的图像纹理精度和视觉效果不佳。由于SAR图像包含的信号和噪声分布以及纹理信息非常丰富,为了提高SAR图像的检索效率,根据图像的视觉特征提出一种图像检索方法,以改善图像的视觉效果,方便人工直觉观察纹理特征信息;由此,采用深度学习方法,结合模糊理论和神经网络的优点来改善图像处理的性能。首先,根据图像像素单元的统计特征和模糊神经网络语义,提出了一种高效的基于图像纹理特征和深度语义分析的方法,对图像纹理风格优势进行数据语义匹配归类;然后,根据语义特征的特性提出一种检索方法。首先,利用深度数据语义聚类提取SAR图像的纹理特征,然后根据同生矩阵方法对SAR图像进行特征分析;最后,利用深度方法对SAR图像的纹理特征和滤波后的灰度组成的矢量进行检索,进而对图像单元归类。实验结果表明,该方法在SAR图像检索方面能取得较好的效果,且视觉效果和分析效率得到较好的提高,便于分析和应用;而且该方法能抑制相干斑噪声,同时提高SAR图像纹理特征的视觉效果。

For the existence of speckle noise in Synthetic Aperture Radar(SAR)image,however,the traditional SAR image interpretation work is quite complicated.However,the image which is quality and visual effect obtained by the traditional SAR image retrieval method are not ideal of conception of it what is perfect most suitable.Therefore,the signal which is contained in the SAR image is not suitable.And the speckle distribution and texture-information are abundant in themselves.In order to improve the retrieval efficiency of SAR images,an image retrieval method is proposed that according to the visual features of the images,thereby improving the visual effect of the images and facilitating the artificial intuition to observe the images’texture(cells)information;thus,using deep learning to take the advantages of fuzzy theory and neural network and to improve the performance of image processing.Firstly,according to the statistical characteristics of image pixel cells,according to the semantics of fuzzy neural network,an efficient image texture feature and Deep Learning semantic analysis method are proposed to classify and match the image texture style advantage.Secondly,according to the semantic feature.The feature is shown that methods propose a retrieval of it.Firstly,the texture features of SAR images are extracted by Deep Learning Data demantic clustering,and then the SAR images are characterized according to the Synchronic Matrix method.Finally,the texture features of SAR images and the vector of filtered Gray-components are retrieved by Deep Learning method to perform Image Cells’classification.The experimental results show that the proposed method achieves preciser-results in SAR image retrieval,and the visual effects and analysis efficiency are better improved for analysis and application.Moreover,the method is effective in suppressing speckle noise and visual effects on SAR image texture features.It’s an increasing strategy with the effects of SAR image analysis.