Research On Key Technologies Of Visible Spectral Remote Sensing Images Restoration

Visible light is the best band for obtaining high spatial resolution images from aerospace remote sensing.The improvement of its spatial resolution and clarity is of great significance to many applications such as target detection,image classification,and visual interpretation.However,many complex and unavoidable factors induced visible light remote sensing image degradation,the loss of high-frequency key information,and image blurring.In the past few decades,a large number of natural image restoration algorithms have made great progress.However,the existing image restoration algorithms applied to visible light remote sensing images will cause small adaptability,poor generalization,and manifest ringing artifacts.It can be attributed to the characteristics of remote sensing images,which include richer detailed information of ground features,complex spatial distribution information,and higher degree of blurring.Therefore,this thesis focuses on the blur characteristics of visible light remote sensing images,conducts in-depth research and analysis on the degradation mechanism and restoration models of remote sensing images,and proposes a complete set of visible light remote sensing image restoration processing procedures,which solves the problem of noise and ringing artifacts in the restoration results.A series of research progress and achievements have been made in this paper.The main content and research results of this paper are as follows:This article starts with the factors affecting the scale of visible light remote sensing images,we analyze the imaging link and imaging mechanism of visible light remote sensing images,and further study the degradation and restoration models of visible light remote sensing images.Combined with the actual visible light remote sensing image degradation type,the relationship between the blurred kernel and the spatial position is analyzed.Based on the random segmentation and perceptual hashing(PHA)fusion algorithm proposed in this paper,the blur type of visible remote sensing image is distinguished,which builds the foundation for choosing the best blur image restoration strategy.At the same time,according to the blurred discrimination type,a visible light remote sensing image restoration process combining information entropy and intensity priori is established.Aiming at addressing the problems of low accuracy of blurred kernel estimation and serious ringing artifacts of restored results in traditional methods,this paper proposes a machine learning algorithm based on gradient boosting decision tree,and selects the blind reference image spatial quality evaluator(BRISQUE)as the mode selection indicator of blurred kernel.On the while,combining the idea of edge extension establishes the basic processing flow of space-invariant visible light remote sensing image restoration,and verifies the algorithm effectiveness through simulated images.Experimental results show that our space-invariant image restoration algorithm can effectively suppress noise and ringing artifacts.The value of peak signal-to-noise ratio(PSNR)and the signal-to-noise ratio(SNR)index of the restored image reached 30.438 dB and 24.183 dB respectively.Compared with the suboptimal method,the PSNR value is increased by about 17.38%,and the SNR value is increased by about 22.87%.The values of structure similarity index(SSIM),the multi-scale structure similarity index(MS-SSIM),and the information weighted structure similarity index(IW-SSIM)index reached 0.891,0.867,0.805,respectively.Compared with the suboptimal method,the SSIM value is increased by about 8.66%,the MS-SSIM value is increased by about 2.6%,and the IW-SSIM value is increased by about 3.74%.Based on the analysis of the spatial blur characteristics of the image,combined with the situations of blurred kernel changes in the central and the peripheral field of view caused by the actual wide field of view,the fusion of the local intensity prior and the block processing strategy are introduced here to restore the image.Algorithm verification experiments were carried out on two groups of simulated datasets and one group of real-data datasets.The experimental results show that the RSSCN7 restored results by our method has improved PSNR and SNR by 2.7% and 5.4%,respectively.Compared with suboptimal methods,the structural indicators of SSIM,MS-SSIM and IW-SSIM are increased by 8.2%,6.79% and 12.33%,respectively.The UCMERCED experiment shows that the PSNR and SNR values of this method are improved by7.9% and 2.12%,respectively.The index scores of SSIM,MS-SSIM and IW-SSIM increased by 9.35%,4.92% and 9.65%,respectively.Real-data experiment shows that NIQE and LPSI indicators are improved by 23.29% and 75.76%,respectively.The space-variant blurred visible light remote sensing image algorithm proposed in this paper has a proposing restored effectiveness.