Hyperspectral Image Fusion Based On Component Substitution And Multiresolution Analysis

Hyperspectral remote sensing images provide spectral context and are often used to characterize the unique properties of ground objects.They have been widely studied and applied in many fields such as environmental monitoring,precision agriculture,mineral exploration and military reconnaissance.Due to physical constraints,a single sensor cannot acquire an hyperspectral image with the desired spectral diversity and high spatial performance.The low spatial resolution greatly limits the application of hyperspectral images in many research fields.The hyperspectral image fusion technique which improves the spatial resolution of hyperspectral images by using the panchromatic images provides an effective way to obtain images with both high spectral and spatial resolutions.As more and more commercial products use high-resolution images(e.g.Google maps,Microsoft bing),the demand for fused data continues to grow.In addition,hyperspectral image fusion is an important preliminary step in image enhancement for many important tasks,such as visual image analysis,object recognition,and change detection.Therefore,hyperspectral image fusion has become a research hotspot in hyperspectral remote sensing image processing.However,many challenges in hyperspectral image fusion still remain:how to effectively extract the missing details of hyperspectral images to reduce spectral distortion and spatial distortion,how to effectively estimate injection gain to realize a balance between pansharpening performance and computational complexity.etc.In order to solve these problems,this dissertation studies the component substitution and the multiresolution analysis based hyperspectral image fusion methods.Hyperspectral image fusion algorithms with high precision and low-complexity are proposed in this dissertation.The main contributions of this dissertation are as follows:(1)Details extraction is a key step in hyperspectral image fusion.Due to the mismatch between the spectral range of hyperspectral image and that of panchromatic image,the traditional component substitution-based hyperspectral image fusion algorithms that extract the spatial details only from panchromatic images often result in severe spectral distortion.We propose a details extraction method for hyperspectral image fusion using saliency analysis and gaussian mixture model.In this method,the structural characteristics of hyperspectral images and panchromatic images are fully considered,and the missing spatial details of hyperspectral images are estimated by using both hyperspectral and panchromatic images.The image segmentation is implemented by saliency analysis and gaussian mixture model to reduce the spectral distortion of the fused images.In the experiments,we applied this proposed method to several highly reliable hyperspectral image fusion algorithms based on component substitution.The results show that the proposed algorithm can not only significantly improve the performance of these original fusion algorithms,but also provides more excellent pansharpening results compared with some state-of-the-art hyperspectral image fusion algorithms.(2)Another key step in hyperspectral image fusion is details injection,which depends on the definition of the injection gain.The global estimation of injection gain is simple to implement,while the context adaptive estimation method can get better fusion results,but the calculation complexity is high.In order to solve this problem,we propose a novel injection gain estimation method for hyperspectral image fusion using image segmentation.A good injection gain should be conductive to maintaining the spatial contrast of the original image in the fused image.It is expected that the pixels with similar spatial details should require the same injection gain.In the proposed algorithm,pixels with similar gradient are segmented in the same region and share the same injection gain.The proposed algorithm is applied to two widely used fusion methods,which belong to component substitution and the multiresolution analysis families respectively.The experimental results show that the proposed algorithm can effectively improve the performance of the original fusion algorithms and achieves the goal of balancing pansharpening performance and computational complexity.(3)The purpose of hyperspectral image fusion is to effectively improve the spatial resolution of the original hyperspectral image while preserving its spectral information.However,the fusion algorithms based on component substitution often cause serious spectral distortion due to the difference between the estimated intensity component and the high-resolution panchromatic image.In order to solve the problem of spectral distortion of pansharpened images,we propose a hyperspectral image fusion algorithm based on matting model.The proposed algorithm provides a stable weight vector to approximate the intensity components by introducing a nonlinear synthesis scheme,and uses the matting model to reconstruct a high-resolution hyperspectral image to reduce spectral distortion.The experimental results show that the fused image obtained by this algorithm can not only greatly improve the spatial resolution of the hyperspectral image but also effectively preserve spectral information of the original hyperspectral image.(4)Hyperspectral images often suffer from the spectral distortion due to the environmental interferences and the inherent limitations of the hyperspectral imagers.For example,the pixels at the boundary of an image may suffer the problem of identical material with different spectral.To solve this problem,a hyperspectral image fusion algorithm based on intrinsic image decomposition is proposed.Intrinsic image decomposition simulates human visual perception,and decomposes the hyperspectral image into a reflectance component reflecting material and spectral color and an illumination component representing detailed texture information.In this algorithm,the details are extracted from the illumination image and the pure spectral information contained in the reflection image is used to reconstruct the high-resolution hyperspectral images.Experimental results show that the hyperspectral image fusion algorithm based on intrinsic image decomposition can effectively improve the spectral fidelity of fused images.