Bayesian Modeling And Analysis Of Target Features In Hyperspectral Images

The classification technology of hyperspectral images can more accurately perceive the information of target objects,and has a wide range of applications in geological surveys,precision agriculture,identification camouflage,and military reconnaissance.However,in the real-life feature classification process,the characteristics of certain specific categories are not significant enough and will greatly reduce the accuracy of the classifier classification.As we all know,hyperspectral images are a stack of two-dimensional images composed of a series of wavelength bands.How to establish a classification algorithm that is not affected by dimensionality and can accurately classify features is a hot research direction.The Bayesian modeling and classification method of hyperspectral images of object features has been verified by experiments,and the classification accuracy has been greatly improved.The innovation of this paper mainly proposes a hyperspectral image classification method combining mathematical statistical characteristics and energy spectral characteristics.Its main content includes the following two parts:First,this paper proposes a probabilistic algorithm for object classification based on a single hyperspectral band,which can obtain a series of hyperspectral images of specific target objects.We obtain hyperspectral images through aerial photography or remote sensors of Earth satellites,but their size is limited,which may affect the accuracy of target classification,especially those small objects lacking pixels,which will greatly reduce the accuracy of classification..Secondly,this paper proposes to study and analyze the hyperspectral characteristics of specific objects based on the normalized hyperspectral energy function in the wavelength domain.In particular,for those targets that are difficult to distinguish using the classic classification method,we use the characteristics of hyperspectral targets for experimental verification,and the results show that the accuracy of classification is indeed improved.The classification accuracy can be characterized by the producer accuracy(PA)and the Kappa coefficient of thehyperspectral image,respectively.The experimental results show that the method can effectively identify and verify targets in hyperspectral images,especially for those objects that are difficult to distinguish by classical methods.In addition,we compare the proposed target classification method based on hyperspectral features with traditional methods,including the spectral information divergence(SID)method and Markov random field-based multispectral angle matching(MSAM-MRF)method.It is concluded that the accuracy of the proposed object-based classification method is20% higher than that of the traditional method.