Research On FCM Clustering Image Segmentation Method And Its Application In Remote Sensing Image Classification

Image segmentation is a practical image processing technique for processing and analyzing image data in a regular Euclidean space.In the past decades,the Fuzzy C-Means(FCM)algorithm has been widely utilized in the field of image segmentation.However,images are vulnerable to noise during acquisition,transmission,and decoding.The diversity and uncertainty of noise lead to FCM algorithms with high segmentation accuracy and computational efficiency that are poorly developed and utilized at present.Remote sensing image classification is the process of classifying the features in remote sensing images based on the characteristics of pixels.However,due to the large amount of spectral information contained in the remote sensing image itself,complex feature types,and ambiguous edges,the conventional FCM algorithm has a low classification accuracy for remote sensing images.In addition,the FCM algorithm has the following problems: 1)the practical universality of the FCM algorithm is poor,which is mainly reflected in the fact that several parameters need to be set manually;2)the FCM algorithm only considers a single pixel feature in the clustering process,which is less efficient for color image segmentation;3)the FCM algorithm generates more outliers of membership in the clustering process,which makes the algorithm fuzzier;4)the FCM algorithm has a high complexity in the computation process,which is unfavorable for handling large-scale data.In this paper,based on the existing FCM theory,various improved FCM algorithms were studied to solve the above problems,and the main works are as follows:(1)The application of the FCM algorithm in image segmentation and remote sensing image classification is introduced and analyzed,and the theoretical basis and mathematical methods of the FCM algorithm are elaborated.(2)For the problem of the FCM algorithm being sensitive to noise and difficult to handle noisy images,a fuzzy subspace clustering image segmentation algorithm based on local variance and non-local information is proposed,called the FSC_LNML algorithm.Firstly,the algorithm proposes a local variance template to eliminate the noise "over-preservation" of non-local spatial information,and integrates the local variance and non-local spatial information into the FCM objective function to improve the robustness.Secondly,the mean membership linking is used as the denominator of the objective function to reduce the number of iterations.Thirdly,the absolute intensity difference between the original image and the local variance & non-local spatial information and its inverse is used to adaptively constrain the original image and the regular term to improve the practical universality.Finally,the concept of the subspace is introduced to adaptively assign appropriate weights to each dimension of the image to improve the segmentation performance of color images.The experimental results show that the FSC_LNML algorithm reduces the complexity,improves the robustness of the FCM algorithm,and improves the segmentation performance of color images,which is better than the equivalent clustering algorithm.At the same time,FSC_LNML algorithm also has excellent applicability in remote sensing image classification and can better classify large target remote sensing images.(3)For the problem that multiple parameters of the FCM algorithm fail to be calculated adaptively and the algorithm has high fuzziness,a fuzzy c-means image segmentation algorithm with adaptive non-local space constraints and KL information is proposed,called KLFCM_ANLS algorithm.Firstly,the algorithm achieves the adaptive calculation of the search window and neighborhood window of non-local spatial information items by defining the smoothing degree and designing the adaptive matching function to overcome the problem of a fixed size of the non-local spatial information window.Secondly,the KL information is introduced into the objective function,and the hidden Markov model is used to calculate the contextual information of image pixels to reduce the fuzziness of segmentation.Finally,the absolute difference between the local variance of the original image and the non-local spatial information term and its inverse is used to adaptively constrain the original image and the regular term to achieve the adaptive selection of the parameters of the constraint term and improve the practical universality of the algorithm.The experimental results show that the KLFCM_ANLS algorithm has efficient segmentation performance and flexibility,and it can also accurately identify the feature targets in remote sensing images and perform efficient remote sensing image classification.(4)Since the FCM algorithm only considers a single pixel feature in the clustering process,which is less efficient for the segmentation of color images,as well as the algorithm adopts type-1 fuzzy membership function,which fails to better classify remote sensing images with high uncertainty due to blurred edges and complex features,an interval type-2probabilistic fuzzy C-means image segmentation algorithm based on spatial information and feature weighting & clustering weighting,called IT2PFCM_SCFWCW algorithm,is proposed.Firstly,the algorithm uses non-local spatial operations to extract the spatial information of the image and improve the robustness of the algorithm.Secondly,the local feature weighting strategy is applied to determine the weighting coefficients adaptively based on the degree of contribution of different features to the segmentation results to improve the clustering efficiency.Thirdly,the cluster weighting strategy is adopted to dynamically calculate the weights of clusters and eliminate overlapping clusters.Finally,the proposed clustering algorithm is combined with interval type-2 fuzzy sets to better handle image data with high uncertainty.Extensive experiments conducted on synthetic and real images show that the proposed algorithm outperforms equivalent clustering algorithms.In addition,the proposed algorithm also performs better on remote sensing images,and can accurately classify remote sensing images with blurred edges and complex features.