Research On Image Reconstruction Method Of Cloud-Contaminated Optical Remote Sensing Image

Optical remote sensing satellite images contain rich spectral information and are widely used in industrial and agricultural production,resource exploration,and environmental disaster assessment.However,due to weather conditions,optical remote sensing images will inevitably be affected by clouds during the imaging process.Cloud occlusion reduces the quality and visibility of satellite images,resulting in deviations or even missing data information,which seriously affects the utilization rate of remote sensing satellite images and brings difficulties to later processing and application.Therefore,reconstructing the ground feature information under the cloud cover of the optical remote sensing satellite image to obtain clear and accurate cloudless images can effectively improve the utilization rate and interpretation accuracy of the optical remote sensing satellite image,and increase the availability of satellite image data.This paper studies the related theory of cloud removal,cloud imaging mechanism,cloud detection method and cloud removal method.In view of the distribution law of thin and thick clouds on optical remote sensing images,a method of thin cloud regional information reconstruction based on wavelet transform and Thick cloud region information reconstruction method based on Poisson equation.The research content and conclusions of the thesis include the following aspects:1.Research and analysis of commonly used cloud detection methods for remote sensing images.Threshold method,multi-temporal comparison method and objectbased cloud detection algorithm are studied respectively.For the cloud analysis results of each method,the threshold method will misinterpret the image and For objects with similar cloud spectral characteristics,the multi-temporal comparison method requires high quality for reference to cloudless images,and misjudgments and missed detections may occur.The object-based cloud detection algorithm does not require reference images,and can be completely integrated according to the cloud spectral characteristics.Thick clouds on remote sensing images are detected,which provides cloud masks for later research.2.Propose a thin cloud region information reconstruction algorithm based on wavelet transform.The thin cloud image is decomposed by selecting a suitable wavelet basis function and decomposition layers,suppressing the low-frequency information of the thin cloud,increasing or maintaining high-frequency detail information,and finally using wavelet reconstruction to obtain the de-cloud image.After experimental analysis,the algorithm proposed in this chapter is superior to the homomorphic filtering method and the dark pixel method,and can be applied to optical remote sensing images with different spatial resolutions.For different distributions of thin clouds,they can effectively reduce the image quality of thin clouds.Influence,reconstruct the true surface information.3.Propose a thick cloud region information reconstruction algorithm based on Poisson equation.The algorithm uses a global optimization strategy based on the Poisson equation.Because the Poisson equation is very sensitive to the boundary constraints and the guided gradient field,the algorithm in this paper replaces the pixels in the cloud area by the cloudless pixels of the cloud pollution image itself to improve the quality of the guided gradient field.,Use the pixels inside and outside the boundary as constraints,and reconstruct the pixel information of the thick cloud region band by band;reduce the bleed and radiation distortion caused by the difference in color between the inside and outside of the boundary and the difference in seasonal phase.Compared with the pixel replacement method and the direct Poisson equation method,the reconstructed image of this algorithm shows good boundary consistency and radiation consistency.