Water Change Detection Based On Hyperspectral Remote Sensing Image Classification

With the rapid development of remote sensing technology, hyperspectral remote sensing image has been gradually applied to all areas,such as meteorological observation,geological survey,resources investigation,cartographic,fine vegetation classification and military reconnaissance.As the most basic question of hyperspectral remote sensing technology,which is directly restricts the development of remote sensing technology,the main task of hyperspectral image classification is according to the electromagnetic radiation information on remote sensing image characteristics to identify ground objects and distribution. Especially in the water change detection,hyperspectral remote sensing technology is fast,accurate,timely and short cycle,which can also analyze the surrounding waters of the aquatic environment and land cover, and make an important contribution to the ecologically sustainable development.Hyperspectral remote sensing images contain abundant spatial and spectral information,but the strong correlation between bands,much redundant information and“the curse of dimensionality ”phenomenon,which led to traditional hyperspectral remote sensing image classification may result in misclassification phenomenon,or low classification accuracy.This paper studies the hyperspectral remote sensing image classifier that combining both spatial and spectral information,and apply to water change detection.The main contents are summarized as follows:Firstly,it elaborates description of the background and significance of our research,then introduces traditional hyperspectral image classification methods, include there advantages and disadvantages,and lists three experimental data sets which always verify method for hyperspectral image classification.Secondly,the research is based on integrating spatial information and spectral information in hyperspectral imagery classification,which is combine active learning and filtering post-processing. The selection of train samples directly affects the classification accuracy.The classification result of multinomial logistic regression classifier only consider the spectral information but ignore the spatial correlation,guided filter can reserve edge information of the image.In this new method,active learning method select training samples which can include all sample information to train multinomial logistic regression classifier,and use guided filter to correct initial classification result. Experiments show that new method take into account the spatial and spectral information,which can improve the performance of spectral classifier.Again,for training samples selection and classifier design,it gives another new method based on locality preserving projection hyperspectral image classification.Because of the high dimensionality and numerous amount of the data,prior to the classification process need to reduce dimensions,so the original feature space can be projected to low-dimensional feature space in particular mathematical methods.The new method combined with locality preserving projection and linear discrimination analysis to reduction dimensionality. Then K neighbor selection can select training samples within a certain distance around the center to train sparse classifier,which can minimize within-classscatter matrices.Experimental results show that the method have better classification results.Finally,it proposes water dynamic change detection based on the above two hyperspectral remote sensing image classification. methods.Traditional water change detection method can not make real-time,efficient investigation for water change detection,but remote sensing technology can accurately observing changes in the water and surrounding ecological environment.In this paper,we use the above proposed hyperspectral image classification method to classify the categories surrounding waters,fuse other categories to extract water,then detect the water change.Experimental results show that hyperspectral image classification can be better used to observe changes in water area.