Remote Sensing Image Segmentation Based On Cluster Analysis

With the rapid innovation of remote sensing technology and space technology,the high-resolution images covered by remote sensing platforms are increasingly rich in information,which covers a wide range of earth surface and spectral information,such as the shape and texture of the earth's surface.Abundant data of remote sensing images carry lays a good foundation for many applications,but the rich data containing image at present and can not be transformed into effective information for human use,so people need to increase in remote sensing information extraction of the input,and a development is the basis of many research fields,to be effective the information of remote sensing information carrying and realize high efficiency transformation,and achieve in-depth application in many research fields.First of all,the remote sensing image needs to be effectively segmented,and the quality of the subsequent feature extraction will be directly determined by the image segmentation.Clustering analysis as a multivariate statistical method,the classic,has become a field of data collection and analysis of more frequent use of the tools of cluster analysis theory and promote the application of the current development of machine learning,such as image processing,biology and other fields.Clustering algorithm is an unsupervised learning method,that is,no training samples are needed,which makes clustering method very well applied to image segmentation.The image segmentation technology in this paper is based on the current research results,that is,aiming at the current fuzzy C mean clustering and Mean Shift clustering process,it is difficult to identify the cluster center,and the noise interference is serious.In order to obtain good segmentation effect of remote sensing image,this paper is based on the existing theory and image segmentation algorithm.Because the remote sensing image by shooting environment influences the image often has a certain ambiguity and uncertainty of cloud model theory is to realize the fuzzy and uncertainty theory on the basis of the ultimate goal of the amount of sexual,in one of the amount,and ensure the robustness of the segmentation process,the use of cloud generator has completed the conversion between qualitative and quantitative process.Up to now,there are few researches on the theory of Guan Yun model in the field of remote sensing image processing.This paper will also be new,from the field of signal processing of compressed sensing theory to conduct the research into the remote sensing image segmentation theory,compressed sensing theory has played a good effect in the field of sampling,this theory can ensure the completion of the sampling process in low bandwidth,and effectively avoid the Nyquist sampling limit bandwidth.The compressed sensing theory includes three kinds of matrix,the sparse signal representation,through less data samples to make statements about global information;and the mapping process of sparse signals through the observation matrix,then form the observation data matrix;compressed sensing reconstruction belongs to an important part in the theory of compressed sensing,is the observation signal will be changed is the original signal.In this paper,the application of cloud model and compressed sensing theory in remote sensing image segmentation is studied theoretically and experimentally,and the main work is:(1)The use of reverse fitting multidimensional cloud transform method is used to locate the center of remote sensing image data clustering,effectively solve the traditional fuzzy C clustering segmentation FCM(Fuzzy C-means Clustering,FCM)to a predetermined number of disturbance classification theory.Based on the single nuclear cloud transform method,first proposed the multi-dimensional cloud transform generator theory is applied in the field of image segmentation,this method compared to the single nuclear cloud transform,the final results showed that multidimensional cloud transform can avoid the problem of the frequency distribution model is very good,then ensure that the values obtained with the original cloud model the expected value of smaller difference image.(2)In order to further will be in a similar cloud model in different regions are merged,this paper proposes an algorithm based on the ratio of the maximum expected reverse weighted entropy,based on consideration of different cloud model to show that the aggregation of,are also well aware of relationship between adjacent cloud model,through the cloud after the merger of the expected value vector will become the clustering center of FCM,this algorithm has a good effect for the treatment of gray and remote sensing image.Based on Fuzziness and random cloud clustering,the number of clusters obtained is the cloud model FCM theory(FCMCM)algorithm proposed in this paper,which can achieve the purpose of segmenting remote sensing images very well.The experimental results show that the segmentation efficiency has a qualitative breakthrough.(3)The information of high resolution images is increasingly rich,which covers a large number of earth's surface information and spectral information,such as the shape and texture of the earth's surface.The rich data carried by remote sensing image has become a hot topic in many fields.In this paper,the compression perception theory of signal processing is first implanted in the field of image segmentation.The extraction of image digital features is realized,and redundant dictionary is established,and the digital characteristic equation is multiplied with over redundant dictionary to form a new compound dictionary,and then sparse representation is applied to achieve the goal of dimensionality reduction.After sparse representation,the non-zero element can be called the super-pixel defined in the image segmentation algorithm.The pixel covered different super pixel regions showed a strong statistical,such as texture,color and so on;the super pixels as individuals into the process of clustering algorithm,clustering analysis,but also the need for work on the clustering results mark super pixel basis,to ensure that the segmentation results are not damage to the main structure of image.Experimental results show that the algorithm's spatial complexity has been greatly reduced,and the subjective segmentation effect has also been improved.However,signal reconstruction has not completed the theoretical and practical verification work.Finally,the improved algorithm model proposed in this paper,put forward two kinds of image segmentation algorithm,and the final segmentation accuracy and other parameters were analyzed.The experiments show that the characteristics of remote sensing image segmentation using fuzzy clustering,cloud model segmentation algorithm can achieve better effect based on.