Research On Synthetic Aperture Radar Image Segmentation Based On Conditional Random Field

Synthetic Aperture Radar(SAR)is an active earth observation system with all-weather earth observation and ground penetration.It can be installed on aircraft,satellites,spacecraft and other aircraft.Therefore,the SAR system has unique advantages in disaster monitoring,crop estimation,mapping,and military.However,the special imaging mechanism of SAR system and the diversity of image through the backscattering of objects with different roughness make the processing of SAR images difficult.High-precision segmentation of SAR images plays an important role in SAR image understanding.Therefore,high-precision and low-complexity SAR image segmentation algorithms have been one of the important research contents of SAR images.However,the existing SAR image segmentation algorithms still have problems such as high time complexity,poor noise resistance,and low segmentation accuracy.For the above,the contents of this paper showing as follows:1)The SAR image segmentation algorithm based on deep conditional random field and multi-scale line detector is proposed.Firstly,stochastic clique is established through window of different scales,and then a deep conditional random field model is constructed for SAR images.This model can not only suppress the interference of speckle noise,but also prevent the image sharpening caused by local information and the time complexity caused by global information.Disadvantages of high complexity;Secondly,the multi-scale line detector used to suppress the interference of noise on river details,and the SAR image segmentation algorithm combining deep conditional random field and multi-scale line detector is proposed;Finally,experiments verify that the algorithm is not only robust,but also has high segmentation accuracy.2)The region-level conditional random field model and segmentation algorithm for SAR images based on superpixels is proposed.First,the segmentation is performed on the SAR image using the SLIC method.Then,through simple threshold processing,the superpixel with the same label form clique,and the region-level conditions are constructed based on the texture information and grayscale information.finally,the SAR image segmentation is achieved by the method of MAP,the experiments verify that greatly reduces the time complexity.