无人机遥感农田全景图像拼接技术研究

发布时间：2018-04-07 17:19

本文选题：无人机　切入点：农田遥感图像　出处：《西北农林科技大学》2017年硕士论文

【摘要】：无人机遥感技术属于低空遥感技术。该技术凭借获取影像速度快、采集图像分辨率高、成本低、安全性好等优势,现已广泛应用到了军事、农业、医疗、地理测绘等领域。但受到无人机飞行高度、数码相机焦距等因素的限制,无人机获取的单幅图像难以形成对地面农田区域的整体认知。为了进一步扩大视场,就需要将无人机获取的多张单幅图像拼接成一幅完整的农田全景图像。本文以无人机采集到的150幅地面农田遥感图像作为拼接对象,结合农田遥感图像的特点,研究如何在不降低原始图像分辨率的前提下,精确、高效完成农田全景图像的拼接。主要研究内容总结如下:(1)提出精简特征点方法,以提高两幅图像的拼接效率。针对SIFT算法在特征点检测过程中的问题,通过设定自适应阈值进行优化;同时,按照大地坐标系下待拼接图像的经纬度对应关系,剔除重叠区域以外的冗余特征点。实验结果表明,该算法在不降低两幅拼接图像质量的前提下,使得特征点数量平均减少26%,粗匹配耗时减少51%左右,精匹配耗时平均减少了31%,有效提高了农田遥感图像特征点的匹配效率及拼接速度。(2)选择有效的图像融合算法,完成两幅农田遥感图像融合。通过实验将加权平均融合、羽化融合和多分辨率三种经典图像融合算法性能进行了分析和比较。结果表明,基于多分辨率的融合算法能够有效去除拼接缝隙。同时,使用6个评价指标对三种融合算法进行评价时发现,2/3个指标均表明相较于其余两种融合算法,多分辨率算法的融合效果能够较好地保留农田图像的细节特征及纹理特征。(3)提出多层金字塔拼接策略,完成150幅农田遥感全景图像的拼接。农田遥感全景图像拼接过程中,结合分层思想,将无人机采集到的农田遥感图像序列分为8层进行拼接。与累加式的拼接方法相比,多层金字塔拼接策略将全景图像失真均匀分配于多层拼接过程中,一定程度上降低了农田全景图像的失真现象。
[Abstract]:UAV remote sensing technology belongs to low altitude remote sensing technology.With the advantages of fast image acquisition, high resolution, low cost and good security, the technology has been widely used in military, agricultural, medical, geographic mapping and other fields.However, due to the limitation of UAV flying altitude and digital camera focal length, it is difficult for UAV to acquire a single image to form a whole understanding of the farmland area on the ground.In order to further expand the field of view, it is necessary to mosaic multiple single images obtained by UAV into a complete panoramic image of farmland.In this paper, 150 ground field remote sensing images collected by UAV are taken as mosaic objects. Combining with the characteristics of farmland remote sensing images, this paper studies how to complete the mosaic of farmland panoramic images accurately and efficiently without reducing the resolution of the original images.The main research contents are summarized as follows: (1) A simplified feature point method is proposed to improve the efficiency of two image stitching.Aiming at the problem of SIFT algorithm in the process of feature point detection, the adaptive threshold is set to optimize the algorithm. At the same time, according to the longitude and latitude correspondence relation of the image to be stitched in geodetic coordinate system, redundant feature points other than overlapped region are eliminated.Experimental results show that the algorithm reduces the number of feature points by 26 points and reduces the coarse matching time by about 51% without reducing the quality of the two stitched images.The average time consuming of fine matching is reduced by 31 and the matching efficiency of feature points and stitching speed of farmland remote sensing images are improved effectively. An effective image fusion algorithm is selected to complete the fusion of two farmland remote sensing images.The performance of three classical image fusion algorithms, weighted average fusion, emergence fusion and multi-resolution image fusion, are analyzed and compared experimentally.The results show that the fusion algorithm based on multi-resolution can effectively remove the splicing gap.At the same time, when we use six evaluation indexes to evaluate the three fusion algorithms, we find that two / three of three indexes show that compared with the other two fusion algorithms,The fusion effect of the multi-resolution algorithm can better preserve the detailed features and texture features of farmland images. A multi-layer pyramid mosaic strategy is proposed to complete the mosaic of 150 farmland remote sensing panoramic images.In the process of farmland remote sensing panoramic image mosaic, the sequence of farmland remote sensing images collected by UAV is divided into eight layers.Compared with the accumulative stitching method, the multi-layer pyramid mosaic strategy distributes the distortion of panoramic image evenly in the process of multi-layer mosaic, which reduces the distortion of farmland panoramic image to a certain extent.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】