Study On Multi-spectral Image Mosaic Method Of UAV Remote Sensing Fishery Waters

The application of UAV low-altitude remote sensing technology to obtain water quality parameters of fishery waters is one of the development directions of modern fishery farming intelligence.Limited by the focal length of the camera and the flight height,the coverage area of a single remote sensing image obtained by the UAV remote sensing platform is small,and multiple remote sensing images need to be spliced and fused in a large area,and the preprocessing and splicing quality of remote sensing images is the premise of quantitative research on remote sensing information.In this paper,the multispectral image of a certain aquaculture water area in Huzhou collected by UAV remote sensing platform as the research data,and the characteristics of the existing remote sensing image stitching methods and the problems existing in the collection of images are analyzed,and researches the multispectral image preprocessing method and stitching technology,the main research works are as follows:1)Aiming at the pixel distortion problem generated when the UAV remote sensing platform acquires multispectral images,the multispectral camera internal parameters and lens distortion coefficients are obtaind using the Zhang Zhengyou calibration method,and a geometric correction model based on no control points is established using the UAV attitude parameters,which makes the pixel distortion of the corrected remote sensing image smaller and provides a basis for the coarse registration of images based on POS information.Aiming at the vignetting phenomenon of remote sensing multispectral images and the inconsistency of radiance in the same batch of remote sensing images,the vignetting correction of remote sensing images is realized by using the multi-scale Retinex algorithm with spectral recovery,and the radiance consistency correction is realized by using the alternative calibration method,Which make the corrected image have a good visual effect.And the image preprocessing software based on the above methods is designed and implemented.2)Combining the existing remote sensing image stitching technology and the characteristics of multi-spectral images of fishery waters,a remote sensing image stitching improved algorithm based on POS information and SIFT features is proposed.After the preprocessing of UAV remote sensing multispectral image,if there are many image feature points,the algorithm first calculates the image overlap area according to the result of coarse alignment of the POS information,then uses the PCA algorithm to extract the first principal component of the multispectral image as the input of SIFT feature point detection,and the best matching points in one-to-many is screen by spatial distance,and the distance histogram is used to constrain the global similarity attribute to eliminate the incorrect matching points.Finally,the homography matrix is estimated by the RANSAC algorithm.In the case of too few image feature points or no feature points in pure water,the algorithm only solves the homography matrix according to the POS information.3)The multispectral image stitching algorithm proposed in this paper is tested on381 images,and the optimal value ranges of the pixel coordinate distance term influence factor α and the matching threshold T are determined to be in the ranges of 0.4～0.6 and0.45～0.65,respectively.In addition,the number of feature points detected and the matching effect of feature points are compared between the image stitching algorithm in this paper and the stitching algorithm based on SIFT features.The results show that the number of feature points in PCA images is improved compared with the band mean image.The method of eliminating mis-matched points in this paper Compared with the nearest neighbors,the matching rate and the matching correct rate of the nearest neighbor elimination method are increased by 18.5% and 4.9% respectively,and the matching time is reduced by half.4)Aiming at the problem that the image registration is easy to fail in the later stage due to the small overlapping range of the images in the initial stage of splicing,a splicing strategy combining frame-to-frame and splicing to splicing is proposed.82 sequence images are stitched under the guidance of this strategy,and the cumulative error of the stitching algorithm in this paper and the stitching algorithm based on SIFT features are compared.The average errors of latitude and longitude of the image are-0.703 m and2.468 m,respectively.And the multi-resolution fusion algorithm is used for image fusion.The final stitched image not only contains geographic coordinate information,but also has a good visual effect,which satisfies the panoramic image requirements for quantitative remote sensing monitoring of fishery water quality by UAV.