Hyperspectral Sea Ice Data Dimensionality Reduction And Sea Ice Detection Based On Texture Information

Sea ice is one of the most important causes of marine disasters in the polar and high dimensions,therefore,the study of sea ice detection has important significance.Compared with the traditional detection methods of sea ice,hyperspectral remote sensing has continuous spectral information and rich spatial information,can also get the sea ice images and spectral curves in favor of sea ice detection.Hyperspectral remote sensing data can acquire rich and detail information,high dimension and high redundancy lead to a sharp increase in data,which brings difficulties to the processing of hyperspectral sea ice images.Therefore,in order to achieve the balance between the data processing efficiency and the sea ice detection results,it is necessary to reduce the dimension of the original hyperspectral sea ice data before the classification of sea ice images.Compared with the data dimension reduction method based on feature extraction,the data dimension reduction method based on band selection preserves the original physical meaning of hyperspectral bands,so we use the band selection method to reduce the dimension of the data.In the processes of data dimension reduction,the more information is there in the band,the more characteristic information,so we should select the bands that have enough information.Similarity,due to the high redundancy of the bands,the correlation between the selected bands in the bands set needs to consider.One of the key problems of hyperspectral sea ice image data dimension reduction is: how to choose the optimal band combination which the amount of information as much as possible and low correlation between bands in the process of band selection.When we classify sea ice based on the spectral information of hyperspectral sea ice images,the prerequisite of the optimal detection performance is that different types of sea ice are not identical and separable in spectral dimension,while the same type of sea ice is the opposite.But due to the physical properties of sea ice and environment factors of sea ice distribution,remote sensing data may appear the phenomenon of spectral confusion in some cases.Therefore,we introduced another important feature--surface texture feature,and combined with spectral features to classify sea ice.It can realize functional complementation and enhance the accuracy of sea ice classification.In view of the above researches,in this paper,we proposed an improved similarity measurement method to reduce hyperspectral data dimension,and discussed the application of image texture features in hyperspectral sea ice classification in view of spectral confusion,the main works are as follow:1)We introduce the basic principle and specific process of hyperspectral image data dimension reduction.In order to achieve the balance between the amount of information and the correlation of bands in the selected optimal band combination,this paper proposed an improved similarity measurement method based on linear prediction for dimension reduction of hyperspectral data.Then,sea ice classification based on the support vector machine algorithm was executed using the selected optimal bands set,and the performance of the methodwas tested.The experimental results indicate that the ISMLP method exhibits better performance overall than other methods,and can be effectively applied in hyperspectral sea ice detection.2)In order to solve the problem of spectral confusion and wrong classification in hyperspectral sea ice detection,image texture features are introduced into the classification of sea ice.Different types of sea ice surface will show different texture features,which can provide the possibility for the classification of texture features.This paper used the gray co-occurrence matrix method which has prominent performance in the texture feature extraction to extracting texture features of different types of sea ice in hyperspectral sea ice images,and different texture features have been analyzed and compared.This paper made a comparative analysis between the combination of spectral/texture information and spectral information only in the sea ice classification detection.The experimental results indicate that combination of spectral information and texture information in the sea ice classification detection achieves better classification results than other method.This method can be used to solve the wrong classification problems caused by spectral confusion.