Hyperspectral Image Classification Based On Improved Semi-supervised Fuzzy C-means Algorithm

With the rapid development of computer science,communication technology and sensor technology,a huge of remote sensing data have been stored constantly.The appearance of hyperspectral technology makes spectral resolution increasing at a steady pace and spectral information be abundant.Currently,the hyperspectral remote sensing is one of popular and important topics in remote sensing,and is widely applied successfully in precision agricult ure,forest monitoring,land utilization,ocean monitoring,and atmospheric monitoring,and so on.Facing the vast amounts of data,it is an important problem how to extract the focus parts,found subtle differences and changes of hyperspectral images,and get more accurate information,and so on.With the coming of the big data era and cloud computing,data mining technology has gradually been applied to land use,the coastline monitoring,forest monitoring,atmospheric monitoring,etc.Especially in the study of hyperspectral remote sensing image classificat io n,spatial data mining technology plays an irreplaceable role.For the past few years,due to the large increasing of spectral bands and spatial informat io n of remote sensing image,the data quantity is greatly increased.It is extremely difficult or even impossible to get all the data of the category information in such a large amount of data.The data we have got are only with a few category information.A large amount of data is unlabeled.Semi-supervised learning is one of the important research directions in the field of machine learning and pattern recognition.Because it can accomplish classification of a large number of unlabeled samples by using a small amount of labeled samples,it draws widely attention and application.Unsupervised classification ignores the sample class information,so it usual y cannot get the ideal classification accuracy.The supervised classification needs to label a large number of sample points,this has brought a huge workload.So the advantage of semi-supervised classification is revealed gradually.In this paper,the spatial data mining and classification of hyperspectral remote sensing image and its technology are introduced firstly.Then the methods/algorithms of supervised classification,unsupervised classification and semi-supervised classification are presented in detail.Lastly,the classic FCM algorithm as well as its various improved methods in semi-supervised direction are depicted.Based upon above discussion,this paper proposes a new semi-supervised classification algorithm by revising objective function,the cluster center and the membership degree matrix of semi-supervised FCM.The algorithm makes full use of the labeled sample points in iteration process.Then the improved algorithm is applied to classify two hyperspectral remote sensing images: Indian Pines and Pavia University,and UCR data sets.The experimental results show that one can obtain the good classification results by the improved semi-supervised FCM algorithm.