Research On Hyperspectral Image Target Detection Based On Spectral Information Transformation

Hyperspectral image has a large number of bands and high spectral resolution.Hyperspectral images have many band information,which makes it possible to use spectral information to distinguish different ground objects.This feature makes hyperspectral images have unique advantages in the field of target detection.However,the excessive spectral information makes the information redundant,and it is difficult to extract effective information from numerous spectral information and make full use of spectral information,which brings difficulties for target detection.Therefore,this paper uses some spectral information transformation methods to transform the spectral information of hyperspectral images to make the hyperspectral data more effective for target detection.The essence of hyperspectral target detection is to distinguish the target from the background and increase the gap between them.In this paper,spectral information is transformed from two aspects.On the one hand,hyperspectral data is recycled several times to make full use of hyperspectral target and background spectral information.In the iteration process,the background pixel value of the hyperspectral image is changed,and the difference between the target and the background is enlarged by purifying the background.On the other hand,the hyperspectral image is nonlinear.By nonlinear processing of the data of the hyperspectral image,the structure of the hyperspectral data is changed,the noise is removed and the difference between the target spectra is reduced,so that the hyperspectral data is more conducive to target detection.By using hyperspectral spectral information through multiple iterations,zero vector or vector perpendicular to the target is introduced in the iteration process to change the pixel value of hyperspectral image,so that multi-type background objects can be transformed into one-type background,and the distance between the target and the background can be expanded to achieve the effect of optimizing the target and the background.A matching filtering algorithm for optimizing target and background is proposed.The method to optimize the target and background is Matched Filtering(MF)method.The estimation of MF algorithm is not accurate because the relevant statistical information of target and background class cannot be obtained before the target detection.In order to better obtain the relevant statistical information of the two categories,the first MF algorithm processing of the hyperspectral data was carried out,and the threshold segmentation was carried out to redetermine the target and background,and the MF detector was modified.Then,the MF algorithm is used to iterate repeatedly.In the iterative process,the method of optimizing the target and background is introduced,and the iterative conditions are set to obtain the final detection results.Using two sets of real hyperspectral data and a set of simulated data to target detection experiment,the experimental results show that the proposed optimization target and the background of matched filtering algorithm and MF algorithm compared with some traditional target detection algorithm,better able to restrain background and highlight the target,show that the proposed algorithm can improve the accuracy of hyperspectral target detection.Similarly,the method of optimizing target and background is combined with the method of Constrained Energy Minimization(CEM).Compared with the matched filtering method of optimizing target and background,the two methods are different in correcting the detector.The CEM method only corrects the target.Three groups of real hyperspectral data are used to conduct experiments,and the experimental results show that the method of minimizing the constrained energy of the optimized target and background can achieve better target detection effect.Hyperspectral images are nonlinear and have strong correlation between spectra.In this paper,the Kernel function is introduced into the Minimum Noise separation transform,and the Kernel Minimum Noise Fraction(KMNF)is applied to the hyperspectral target detection algorithm.An important factor affecting KMNF effect is the accuracy of noise estimation.In this paper,two noise estimation methods are used for noise estimation.Finally,two groups of hyperspectral data and two target detection algorithms are used to carry out experiments.The experimental results show that the KMNF condition can better highlight the target and improve the accuracy of hyperspectral target detection.