Research On Geometric Correction And Stitching Technology Based On UAV Infrared Image

With the rapid development of UAV,the use of UAV remote sensing technology to monitor the thermal pollution of the sea area caused by the thermal drainage of nuclear power plants provides reliable and effective data for ocean temperature monitoring and protection of the marine ecological environment.In the project,the use of UAV equipped with infrared cameras to obtain images with a larger field of view has been widely developed,but there are still two shortcomings: Infrared cameras using wide-angle lenses with a large field of view are very likely to cause image distortion;The limitation of flight height makes it impossible to have an overall grasp of the study area.Therefore,in order to obtain remote sensing images of large-scale scenes,it is essential to improve and develop aerial infrared image stitching technology.This paper takes the temperature monitoring of the adjacent sea area of Fangchenggang Hongsha Nuclear Power Station as the engineering background,and conducts research on the above two types of problems of image distortion correction and image stitching,and solves the problem of making a schematic diagram of the overall temperature field of the sea area has certain engineering significance for the environmental impact assessment of the later temperature drainage.The main work is as follows:(1)A systematic analysis of the causes of the distortion of the collected image caused by the wide-angle lens of the infrared camera.The traditional equivalent surface model method and grid template method are theoretically analyzed and deduced.In order to study the applicability and correction effect of these two algorithms,a large number of simulation experiments were used to realize the distortion correction of the remote sensing image.(2)Propose an image distortion correction method based on genetic algorithm and Bayesian regularization optimization BP neural network.This method makes full use of the advantages of neural network in digital image processing and establishes the mapping relationship between ideal image and distorted image to achieve image distortion correction.The introduction of Bayesian regularization can effectively avoid the phenomenon that traditional BP neural networks are prone to overfitting.At the same time,the use of genetic algorithms greatly improves the robustness and global optimization capabilities of neural networks.Finally,the infrared image of Fangchenggang Hongsha Nuclear Power Station wasused to verify the algorithm,and compared with the traditional correction method to verify the accuracy and feasibility of the algorithm in solving image distortion problems.(3)Aiming at the problem of splicing without obvious features in engineering environments such as sea areas,a registration method based on POS information of UAV and Fourier Merlin transform is proposed.This method first uses POS information for overall image registration.If there is any deviation in the POS information,the Fourier Mellin algorithm is used to internally correct the position accuracy of the image with deviations in the POS information.Through stitching simulation experiments on some images of nuclear power plant,it is verified that the method has better registration accuracy than registration methods based on UAV POS information only,which can well compensate for the problem of deviation in UAV POS information,which has good engineering applicability.Secondly,a fusion algorithm combining optimal stitching and multi-resolution is proposed,which can effectively solve the problems of "black border" and stitching traces in the process of stitching and fusion,and greatly improve the visual effect of the fusion image.