SAR Image Segmentation Method Based On Ridgelet Filter And Convolution Structure Learning Model

Synthetic Aperture Radar SAR is an important advance in the field of remote sensing technology,it is used to obtain high resolution images of the Earth's surface.Compared with other types of imaging technology,SAR imaging technology has a very important advantage,it works even there is not cloud,rainfall or fog and other atmospheric conditions and the impact of light intensity.It can work all day,all-weather to access to high-resolution remote sensing data.SAR image interpretation technology have important guiding significance for military,agriculture,geography and many other areas.Depth learning is a kind of machine learning theory proposed in recent years,and the depth learning theory has better adaptability than the algorithm of artificial design.The deep convolution neural network has its unique advantages to the structural feature learning of images,and the mixed aggregation structure of the SAR image has rich structural information.with the study of the convolution neural network and the existing SAR image hierarchical semantic space and Super complete ridgelet redundant dictionary,we propose a method based on the ridgelet filter convolution structure learning for the SAR image mixed aggregation structure.The method can be used to characterize the structural features of the subspace of the aggregated structure pixel,which obtains much better segmentation result.The innovation of this paper is as follows:(1)Inspired by Lenet,combined with the existing research on the ridge wave redundancy dictionary and SAR image sketch,we propose a convolution network structure based on the ridgelet filter guided by sketch line.We sample the mixed space of the SAR image.In the hierarchical semantic space,We statistics sketch line in the direction of the information using the statistical method.And get the first six directions.The ridgelet filter of the known direction is used as the initial filter of the convolution structure model,at last,we use the sample of SAR image to prove the feasibility of our method.(2)Inspired by convolution neural network,based on the existing hierarchical semantic space and Super complete ridgelet redundant dictionary,according to the data fidelity items and structural fidelity items,A SAR image segmentation method based on ridgelet filter and convolution structure learning model is proposed.The method can be used to characterize the structural features of the subspace of the mixed aggregate structure,so that we can obtain a much better SAR image segmentation.(3)For the innovation point(2),we carry out the direction statistics of the SAR image sampling block,we propose a learning model based on the sampling block direction and convolution structure.Based on data fidelity items and structural fidelity items,we designed algorithms to learn more complex filters for further learning.