Sub-Pixel Mapping Of Hyperspectral Image Based On Linear Feature Detection For Mixed Pixel

Due to the influence of environmental factors or limit of sensor resolution during data acquisition,mixed pixels are commonly found in hyperspectral remote sensing images.The existence of mixed pixels limits the spatial resolution of hyperspectral images.Therefore,effective determination of the spatial distribution of mixed pixels has become a hot issue studied by scholars at home and abroad to achieve sub-pixel classification of hyperspectral images.Most of the existing sub-pixel mapping algorithms only focus on mapping for areal object,and the research on other spatial features in the mixed pixels is less,the effect on the practical application needs to be improved.In order to solve the aforementioned problem,this paper mainly does some researches on the issues of indicating the mixed pixels that include linear features and improving the sub-pixel mapping accuracy by designing reasonable mapping strategies for mixed pixel which are combined with ground objects of different spatial distribution.The main research contents include:(1)Aiming at the problem of the lack of different spatial distribution characteristics in the sub-pixel mapping,a linear feature detection method based on complete line set is proposed.This method determines the distribution of typical linear features and the inclusion of linear features by spectral unmixing.Then the remaining mixed pixels containing linear features are determined based on the maximum linearization index of complete set of lines.The experimental results show that the method has improved accuracy and efficiency.(2)For the mapping problem of mixed pixels with linear features based on template matching,the method of template selection and sub-pixel mapping process are studied respectively.Based on the correlation coefficient selection template,two new template selection methods are proposed:?Further determining the template by fitting a line with linear feature mixed pixel eight neighborhood,to some extent,reduces the scope of template selection;?In order to avoid the uncertainty of template selection,the best template is obtained through line fitting on the mixed pixel eight neighborhoods and the template and calculating the linear correlation.By analyzing the problems of sub-pixel mapping algorithm based on template matching,a linear feature sub-pixel mapping algorithm based on improved template matching is proposed.Furthermore,in order to avoid the influence of template matching algorithm on linear feature mixed pixel mapping,an improved line fitting sub-pixel mapping technique is proposed,which improves the classification accuracy and time efficiency compared with the template matching sub-pixel mapping algorithm.(3)A sub-pixel mapping algorithm based on the spatial distribution of the land cover classes is proposed.The divide-and-conquer mapping strategy combines the linear feature mixed pixel mapping algorithm and the area feature mapping method based on linear optimization are effectively combined to improve the sub-pixel mapping accuracy.The experimental results of simulation data and real data show that the proposed linear feature detection algorithm based on complete line set can not only realize sub-pixel level feature in short time,but also have low computational complexity.The proposed sub-pixel mapping algorithm with linear features can effectively improve the mapping accuracy,which not only ensures the connectivity of the linear features,but also locates the remaining classes in the mixed pixels.The proposed linear-area sub-pixel mapping frame not only considers the sub-pixel mapping of the linear features but also the sub-pixel mapping of the area features,therefore the mapping accuracy is improved on the basis of ensuring the spatial features of different features.