Target Detecting With Semantic Information In High Resolution SAR Image

Synthetic Aperture Radar(SAR)is a kind of active microwave remote sensing imaging radar coherent.It is unlike optical remote sensing which can work but in the daytime,and is different from infrared remote sensing affected easily by the surrounding environment and temperature.SAR can be equipped with a variety of platforms(satellite,airplane,etc),which can be imaged at anytime,and is not affected by the weather,temperature and light.Therefore,SAR plays an important role in the field of natural disaster monitoring,topographic mapping,agricultural and forestry resources investigation,environmental assessment,military and other fields.With the increasing number of SAR imaging methods,more and more information is available,and the amount of data acquired is increasing.The TerraSAR-X satellite in Germany can be observed around the earth in more than ten days.The resolution of the SAR image is as high as 1m.Therefore,how to deal with massive and high resolution SAR image data quickly and effectively has become a challenge.At present,the ability to interpret SAR images is far behind the acquisition ability of SAR image data.It has become a key factor restricting the application of SAR,and is also an urgent problem to be solved in the current research.Therefore,the automatic processing technology of SAR images has become a hot topic both at home and abroad,and automatic target detection technology is one of the core issues.SAR image automatic target detection system ATR(Automatic Target Recognition,referred to as ATR)refers to automatically detect the target and location of interest in a relatively short time without the direct intervention of human beings.The existing SAR image target detection is mainly based on the contrast algorithm,and the CFAR algorithm is the most widely used.The targets in SAR images have strong scattering characteristics and shadow features.The strong scattering and shadow features of the same target have a semantic association relationship in the spatial location,and there is a certain context relationship between the target and its surrounding environment.We regard these relations as rules or strategies to guide target detection,and improve the original detection algorithm,so as to design more effective SAR image target detection algorithms.In addition,using the semantic information of sketch line segments in sketch map of high-resolution SAR image,we propose a framework of SAR image target detection based on the interaction of semantic space and pixel space information for high-resolution SAR image target detection.Based on this framework,the detection method of related targets is designed and realized.The main innovative work of this paper is as follows.1.In this paper,a Constant False Alarm Rate(CFAR)detection method based on matching of bright and dark regions is proposed,which is suitable for vehicle detection in high-resolution SAR images.The traditional CFAR algorithm mainly uses the strong scattering characteristics of targets to detect,while the CFAR method used in bright and dark regions uses not only the strong scattering characteristics of targets but also the shadow characteristics of targets.According to the strong scattering characteristics and the spatial position relationship between shadows,the false alarm target can be effectively reduced.Experiments on Mini SAR images show that the proposed algorithm can achieve lower false alarm rate compared with the traditional constant false alarm rate algorithm with the same detection rate.2.For high resolution SAR images,it is difficult to find the accurate background clutter distribution probability model,while the CFAR algorithm relies on the clutter distribution model.This paper proposes a vehicle detection algorithm for high resolution SAR images without the background clutter distribution model.The algorithm uses fuzzy clustering and automatic threshold segmentation algorithm to search the bright area and dark area in the scene,and then selects potential vehicle strong scattering area and potential automobile shelter area.Then match their spatial position relations and calculate their membership from the same vehicle.Finally,the target with high membership is selected by threshold and combined and output.Experimental results show that the method can detect targets effectively without any background clutter probability model.3.In view of the above two algorithms,a coupling constant false alarm rate detection algorithm is proposed to detect multiple targets in bright and dark regions.Compared with the traditional CFAR algorithm,only the strong scattering characteristics of the target can be detected.The CFAR algorithm can detect the strong scattering characteristics of the target and detect the shadow characteristics of the target simultaneously.Because the CFAR algorithm uses region matching instead of template matching,it can detect many different targets.Experimental results show that the algorithm has good detection results.4.Due to the complexity of the scene and the limitation of SAR imaging mechanism,the target in the scene is mixed with the non target.The non target often shows strong brightness,and the inherent speckle noise in the imaging process makes the useful target information imply.Aiming at these problems,a SAR image target detection method based on semantic space and image pixel spatial information interaction is proposed.It can effectively utilize the different characteristics of the target in two spaces to carry out information interaction,and realize the accumulation of semantic information,so as to complete the detection of targets.First,the Sketch map is extracted from the SAR image,and then the homogeneous region in the SAR image is segmented based on information interaction.According to the statistical characteristics of the water level,it is detected from the homogeneous region of the SAR image.The water information is fed back to the candidate target area map in the semantic space,and the candidate bridges and the candidate port target areas are determined through the contextual information of the water scene.Then,the regions with semantic information are mapped to pixel space to realize the detection and location of the relevant targets,and finally the bridge and port targets are detected.In uneven areas,the building targets are detected according to the matching of bright and dark areas.Experimental results show that the algorithm can identify important objects such as waters,bridges,ports and buildings efficiently.