Research On Land Object Classification And Application Of Hyperspectral Remote Sensing Images Fused With Superpixel Segmentation

With the rapid development of hyperspectral remote sensing technology,while hyperspectral remote sensing images have high resolution,the spatial resolution of images has been greatly improved,which creates favorable conditions for the accurate classification of ground objects.Therefore,it is widely used in crop monitoring,urban growth analysis,major disaster management and mineral mapping.Hyperspectral images bring new opportunities for the recognition and classification of ground objects.At the same time,new challenges are brought to the existing image processing technology: because the image contains more types of ground objects,the texture is more complex,and the increase in the image’s spectral resolution leads to a lager amount of data in the image.When using superpixel segmentation to extract the spatial features of the image,the accuracy and efficiency of segmentation are limited;In the process of hyperspectral imaging,the classification of images is more challenging because of the phenomena of "similar species with different spectral curves" and "different species have similar spectral curves".Based on the in-depth summary and analysis of the research status of superpixel segmentation and hyperspectral image classification,this paper studies the high-performance superpixel segmentation technology and classification technology based on the spectral information and spatial characteristics of hyperspectral images.The innovation work and main tasks of this paper are as follows:(1)A fast region growing superpixel segmentation method for hyperspectral image is proposed,which combines spectral-spatial information of the hyperspectral image to extract the local structure features of image.When the existing superpixel segmentation method is used to segment a hyperspectral remote sensing image,the accuracy and efficiency of the segmentation are not high due to the complexity texture in the image and the spectral characteristics of the image are not effectively used.Therefore,in this paper,the spectral information and spatial information of hyperspectral remote sensing images are used for superpixel segmentation.Several common superpixel segmentation methods are compared in the experiments.The results show that the proposed method has advantages in segmenting objective evaluation indicators,and verifies the accuracy and effectiveness of the method in extracting spatial features of hyperspectral remote sensing images.(2)A hyperspectral remote sensing image classification method based on the fusion of superpixel segmentation is proposed.The problem of low accuracy of traditional classification method based on spectral information is improved by fusing the local structure information of the image.The traditional hyperspectral classification method usually relies on the spectral characteristics of the image to classify ground objects,but the classification accuracy is not high due to the phenomenon of "same object with different spectral curves" and "different object have similar spectral curves".In response to this kind of problem,this paper first uses superpixels to extract the spatial characteristics of the image,establishes an adaptive fusion strategy,fuses the superpixel segmentation results and pixel-based classification results,and improves the overall accuracy of feature classification.Multiple classification methods are integrated in the experiment,and the results show that the proposed method can effectively improve the overall classification accuracy,which verifies the effectiveness of the proposed algorithm.(3)A hyperspectral remote sensing image classification system with the fusion of superpixel segmentation is designed and implemented.The system includes four functional modules: an image loading module,a superpixel segmentation module,a hyperspectral classification module,and a performance evaluation module,respectively completing the selection of hyperspectral remote sensing images,image segmentation under different superpixel sizes,hyperspectral image classification based superpixels and display of classification evaluation indicators before and after fusion.Using this system can clearly show the superpixel segmentation results and the classification results before and after the fusion of superpixels.