Spectral-spatial Classification Of Hyperspectral Image Based On Spatial Coordinate Information

Now,the hyperspectral remote sensing technology has been developed well.The hyperspectral images have been used in many fields because of their high resolution and the combination of traditional image dimension and spectral dimension.The hyperspectral image classification,which is one of the hyperspectral remote sensing technologies,has attracted many researchers' attention.Unlike traditional images or multispectral images,the high spectral resolution of the hyperspectral image provides a strong basis for the object recognition.But the high spectral resolution also brings some problems,such as a large amount of data and information redundancy,which make the classification difficult.At the same time,marking samples is costly.It is difficult for algorithms to obtain good classification results with limited training samples.Most recent research indicates that using both spectral information and spatial information can improve the classification accuracy.So finding a better way to combine spectral information with spatial information is also an important task.To solve these problems,this thesis presents several classification methods based on spectral information and spatial information.Specific research contents are as follows:(1)An approach for hyperspectral image classification based on spectral-spatial feature fusion using spatial coordinates is proposed.Firstly,principal component analysis is performed on hyperspectral image to select partial data as the spectral feature.Then,the spatial coordinate feature is used in supervised classification to output the probability feature of samples,and the spectral feature is also used to get the probability feature.Finally,the spectral probability feature and the spatial probability feature are combined to be used in supervised classification to obtain the final classification result.This method uses the space coordinates of the samples to put the spatial information into classification.The spectral probability feature and the spatial probability feature,which are made by classifiers,make it easy to combine spectral information with spatial information.Compared with other methods using spatial information,this algorithm can obtain higher classification accuracy with less running time.(2)A hyperspectral image classification method based on active learning and spatial coordinate is proposed.The result of supervised classification based on spatial coordinate has a strong relationship with the spatial position of training samples.According to this phenomenon,active learning is introduced to be combined with spatial coordinate characteristics.The samples with high uncertainties corresponding to spectral classifier and spatial classifier are selected respectively by sampling scheme and marked manually to retrain the classifiers,which improves the classifiers' performance rapidly.Finally,the fusion of the spectral feature and spatial coordinate is analyzed and classified,so that the algorithm can obtain high classification accuracy using a small number of training samples in a short time.(3)An ensemble learning method for hyperspectral image classification using spatial coordinate is proposed.Firstly,a set of sample probability values are obtained from each base classifier which represent the probability that sample belongs to each category.Secondly,the probability values obtained by each base classifier are multiplied by the sample probability values obtained from the spatial coordinate feature classifier.Then the multiplied probability values are multiplied by each other to get the final probability values of each sample.The category corresponding to the maximum final probability value is selected as the label of the sample.Finally,the labels of all samples are determined by this way.This method integrates spatial coordinate with ensemble learning method.It makes the classification of each base classifier be combined with the spatial information.So the accuracy of final classification is increased.