Research On Stitching Farmland Imagesabout Unmanned Aerial Vehicle Remoting Sensingtechnology

Unmanned Aerial Vehicle(UAV)remote sensing is a branch of low-altitude remote sensing technology and it has been widely used in the area of military,agriculture,medical science and geography mapping etc as it has the advantages of faster image access speed,higher image resolution with low cost and outstanding security insurance.However,it is difficult to stitch all the single images obtained by an UAV into a panoramic one with the restriction of flight height and digital camera focal distance etc.In order to broaden the horizons further,more than one single image has to be stitched into a panoramic farmland image.This paper focuses on how to stitch 150 farmland remote sensing images collected by an UAV into a panoramic one precisely and efficiently without resolution reduction of the original images through combination of the characteristics of the farmland remote sensing images.The main works are as follows:(1)A method of feature points streamlining is proposed to improve the stitching efficiency.The shortcomings exist in the SIFT algorithm during the process of feature point detection have been optimized in this paper by setting adaptive threshold.At the same time,some redundant feature points beyond the overlapped area have been removed according to the corresponding relation of geographic coordinate based on geodetic coordinate system.The experimental results show that the number of feature points declines by an average of 26% and the time-consuming of coarse matching and precise matching reduces by 51% and 31% respectively without resolution reduction of the original images by using the proposed method,which proves that both the stitching efficiency and stitching speed have been improved.(2)A more effective image fusion algorithm is chosen to complete fusion operation of two farmland remote sensing images.The performances of weighted average fusion,feather fusion and multi-resolution fusion algorithms are compared and analyzed through experiments.The results show that the multi-resolution algorithm can eliminate stitching gapeffectively.Meanwhile,another experimental analysis suggests that four indicators show that more minutiae and texture features of farmland remote sensing image can be better reserved through multi-resolution algorithm when assessing by six indicators.(3)A multi-layer pyramid stitching strategy is proposed to achievestitching of 150 farmland remote sensing images.During the stitching process,the images captured by the UAV are divided into eight layers according to the proposed strategy.Compared with the traditional accumulation stitching strategy,the proposed strategy assigns image distortion to the stitching process of every layer equally,which to a certain extent decreases the distortion.