The Research On Identification Methods Of Hyperspectral Remote Sensing Images In Complex Scenes

Hyperspectral remote sensing image is a kind of image cube that integrates twodimensional spatial and one-dimensional spectral information of a feature into one,reflecting the characteristics and advantages of image space-spectrum integration.In recent years,hyperspectral remote sensing image processing technology has developed rapidly and played a vital role in many fields such as military reconnaissance,resource exploration and urban planning.However,the sensor is inevitably affected by the imaging environment(underwater and shadow,etc.)when acquiring hyperspectral remote sensing images,leading to problems such as attenuation and spectral variation of the acquired hyperspectral image target features,which seriously affect the subsequent image identification.Existing hyperspectral image identification methods heavily rely on the quality of the source data and often face performance bottlenecks when processing images in complex environments.How to recognise hyperspectral image targets from complex environments is a problem that needs to be addressed urgently.To solve the above difficulties,this thesis proposes a hyperspectral remote sensing image identification method for two complex scenes: underwater and shadow.The research content and results of this paper are summarized as follows.(1)Aiming at the problem that traditional target detection methods are difficult to detect underwater targets due to the attenuation of spectral features of targets in underwater hyperspectral remote sensing images,an underwater target detection method based on the guidance of unmixing model is proposed.Firstly,a linear spectral unmixing technique is used to extract the end elements of the target and the background,then the significant spectral bands of the target are extracted by analyzing the difference between the end elements of the target and the background.Next,the initial detection probability map of the target is obtained based on the fusion of multiple spectral bands.Finally,the noise points of the initial probability map are further optimized with an edge-holding filter.The method is validated on real hyperspectral remote sensing data,and experiments show that the method can effectively detect underwater targets in hyperspectral remote sensing images compared with other traditional target detection methods.(2)A hyperspectral image identification method based on multi-source information fusion is proposed to address the problem that the spectral variation of hyperspectral image targets under shadows makes it difficult for traditional image identification methods to effectively identify objects in shaded areas.Firstly,the shadow region is detected based on the reflection characteristics of the hyperspectral image target,multiple exposures are performed on the shadow region and the multiexposure hyperspectral images are fused to remove the shadows from the source images.Then the obtained purified hyperspectral images are spectrally downscaled using the averaging fusion method.Finally,the hyperspectral image identification method with the preservation of the null spectral structure is proposed.Experimental results on Houston 2013 data show that the method is robust to shadows and effectively improves the identification accuracy of the images.(3)Based on the results of the hyperspectral remote sensing image identification method under complex scenes proposed in this paper,the graphical user interface development of hyperspectral remote sensing image identification software was carried out in Python 3.7 language environment using PyQt5 application framework,and the hyperspectral remote sensing data under real complex scenes were tested on the software at the same time.