Research On The Key Technologies Of Airborne Hyperspectral Data Processing

The high-spatial resolution of airborne hyperspectral(HS)imager has improved the efficiency of applications,such as disaster emergency response and rescue,precision agriculture,monitoring and prevention of forest pest and disease.The HS images are used to extract the important information and solve the problems in the applications.This dissertation mainly studies the key technologies of airborne HS data processing,including non-uniformity correction,compression and spectral unmixing of HS images.The principles and mathematical models of these methods are presented in the dissertation,and the effectiveness and validation of these methods are also verified using HS images.The main contents and innovations of the dissertation are as follows.1)The dissertation analyzes the sources of the non-uniformity of airborne HS images,and studies the classical non-uniformity correction methods at home and abroad.A relative radiation correction method based on the "side-slither" technique(RRC-SS)is proposed by studying the "side-slither" technique and the two-point multi-segment non-uniformity correction method.The principle of the proposed algorithm is presented in detail.The effectiveness and validation of the RRC-SS method are verified using airborne short-wave infrared(SWIR)HS images,and the correction effect of the method is also quantitatively evaluated by three indexes.The experimental results show that the RRC-SS method can successfully eliminate the correction error introduced by the equivalent color temperature difference between the integrating sphere and the sun.In conclusion,the RRC-SS method greatly improves image quality and retains the useful information of original image to the greatest extent,which can improve the accuracy of HS applications such as absolute radiation correction and target recognition.2)The dissertation analyzes the reasons why the HS images need to be compressed,and we propose a lossy compression method based on the unsupervised classification of HS data.The principle and mathematical model of the proposed algorithm are elaborated in detail.The classification accuracy and compression effect of the proposed algorithm are evaluated using the airborne visible near-infrared(VNIR)HS image.The experimental results show that the proposed method can achieve the classification and compression of HS data simultaneously.Moreover,the method has a good performance in classification and compression of HS images.Therefore,the method has a strong advantage in applications that require both compression and classification.3)In this dissertation,the classical spectral unmixing methods are studied and a linear spectral unmixing method based on the panchromatic(PAN)image and HS image is proposed.The mathematical model of the proposed algorithm is presented in detail.The effectiveness of the algorithm is verified using the airborne VNIR HS images,and the advantages and disadvantages of the method are also discussed.The experimental results show that the proposed method can achieve a good result,especially the endmembers in the PAN image are distinguishable.4)In this dissertation,the effect of misregistration between PAN image and HS image on linear spectral unmixing is studied.The experimental results and the related mathematical model show that the spectral unmixing error of the mixed pixel increases with the misregistration error between the PAN and the HS image.5)In the dissertation,a quantitative method for the maximum acceptable misregistration error of the spectrometer is proposed.The maximum acceptable misregistration error is determined according to the spectral unmixing error introduced by misregistration between images.With the PAN image as a reference,the maximum acceptable misregistration error ranges from 0.55 to 1.41 pixels,which provides reference data for the error budget between the PAN and HS detectors of imager.