Super-pixel Segmentation And Merging Method For High Spatial Resolution Remote Sensing Images

With the improvement of the spatial resolution of the remote sensing image,the types and scenes of the objects are more and more complex.The details of the objects are more abundant,and the relationship between the objects is more complex and diverse.The excellent image segmentation algorithm can accurately segment the ground target,and the segmentation results can provide a reliable basic processing unit for the subsequent image processing and analysis.Because of the uncertainty of image segmentation itself and the complexity of high spatial resolution remote sensing image,the research in this field is very challenging.The traditional segmentation methods can not meet the current segmentation needs.Most of the existing segmentation methods can not balance the problem of over segmentation and under segmentation.Too small scale is easy to over segmentation,while too large scale causes under segmentation,which leads to the wrong segmentation.In the following image processing,error segmentation is irreversible,which will directly affect the subsequent processing.In this paper,on the premise that the optimal segmentation scale cannot be determined,a smaller segmentation scale is used for pre segmentation,and different merging methods are added to the image segmentation process to improve the segmentation effect.Super-pixel segmentation has become a key technology in image preprocessing,which can better describe the information of small-scale area,and at the same time,it can achieve the goal of large-scale object merging through three merging strategies of spatial clustering and spectral difference.Based on this,this paper discusses the problem that the segmentation method can not balance the segmentation scale and the segmentation effect is not ideal.Based on the detailed analysis of traditional remote sensing image segmentation methods and the summary of the latest super-pixel segmentation methods,aiming at the problem of scale balance in the process of image segmentation,the following research is carried out:(1)A superpixel segmentation method of high spatial resolution remote sensing image based on hierarchical clustering is designed.Firstly,the adaptive multi-scale reconstruction watershed segmentation algorithm is used to generate super-pixel over segmented image for remote sensing image,and then the hierarchical clustering algorithm is used to merge the superpixel.(2)A high spatial resolution remote sensing image segmentation method combining superpixel and FCM is studied.Firstly,watershed algorithm is used to generate the initial hyperpixel segmentation region,and then FCM algorithm is used to merge the hyperpixels.(3)In this paper,a high spatial resolution remote sensing image segmentation method is proposed,which combines super-pixel and spectral difference.Firstly,the single linear iterative clustering(SLIC)algorithm is used to pre segment the super-pixel set,then the spectral measure difference of pixels is fully considered,and the super-pixel is merged by the spectral difference algorithm.The experimental results show that this method can solve the problem of segmentation of different scale objects in high spatial resolution remote sensing image,and get good segmentation results.It lays a foundation for further remote sensing image processing and application.