Registration Of Frame Hyperspectral Images Based On UAV

Along with the rapid development of remote sensing technology,requirement for accuracy of earth observation is increasingly high recently.Hyperspectral technology is becoming one of the major topics in quantitative remote sensing research due to its characteristic of unification of geometrical and spectral information.Hyperspectral image can accurately reflect the spectral features and differences of crops,combine spatial distribution and reflective character of surface objects.It was extensively studied and applied in many fields,especially in agricultural applications in which great progress was achieved,with intensive study on crop condition monitoring,pest control and yield estimation.Therefore,the study on hyperspectral technology is of great significance.Currently hyperspectral observation can be acquired from spaceborne,airborne and ground based multi-angle platforms.Unmanned aerial vehicle(UAV)technology effectively covers the shortage of satellite platform,which cannot provide data in satisfactory spatial or temporal resolution,for providing powerful data and technology support for quantitative research of agricultural remote sensing,accurate agriculture and intelligent agriculture.On the UAV platform,frame imager,with the ability of dealing with non-georeferenced data,is of much better prospect compared with traditional push broom imager,whose availability is highly limited by platform stability and load capacity.However,the process of registration is necessary before further application of original band images of frame imager,which suffer from spectral and spatial discrepancy for the reason of different acquisition time.The main objective of this paper is to study band registration method of frame hyperspectral data.Firstly,research situation of image registration was reviewed.Secondly,working principle and features of frame hyperspectral imager were introduced,along with the theoretical basis of registration method based on matching points.In our study,we developed an effective registration method for UAV images from the two aspects of matching strategy and matching method optimization by thorough analysis of the influences of brightness and spatial discrepancies between different bands.Our main study includes:(1)Study on matching strategy in highly textured conditions.Based on the principle data collection of hyperspectral imager,we designed three matching pattern and two experiments to study the reference selection and matching sequence.The result indicates that the changing reference method is of better stability,thus of higher feasibility for registration,and the accuracy can be improved,with the RMSE decreasing from 0.415 pixel to 0.399 pixel,when introducing shot sequence.(2)Researching on matching method in less textured cropland.A modified matching method was proposed aiming at the problem of registration error caused by brightness and contrast difference when using SURF algorithm.The new method employed MI_STD method to improve the extraction of matching points by introducing mutual information algorithm into local matches,taking the advantage of its ability to deal with brightness nonlinear variety problem.In addition,the efficiency of algorithm was improved in ways of constraining searching area by importing phase correlation algorithm.Since both brightness and spatial information of the image are considered,the modified method is of high robustness and sub-pixel accuracy.As a consequence,the newly raised algorithm can effectively solve the problem of local distortion of traditional methods,with a average accuracy of 0.8 pixel.