The Classification Of Hyperspectral Microscopic Image About The Human Body Blood

Hyperspectral imaging is an emerging technology,which can simultaneously obtain the apperarance and spectral information of things.With the development of imaging equipment and algorithms of computer vision,it has become a unique way of diagnosis,applied in the field of military,agriculture,atmosphere,and biomedicine successfully.As for the microscopic hyperspectral imaging,it's convenient for us to explore the micro-world without limitation of naked eyes,such as cell classification,food safety,and cancer diagnosis.This paper mainly focuses on the classification and recognition of blood cell hyperspectral image,including image acquisition,image construction,and image classification.And the primary contributions of this paper are:First,image noise is unavoidable due to the process of imaging.To overcome this issue,we proposed a new image reconstruction method based on guided filter.The framework of the method contains three steps:(1)obtain the best guided image according to the band selection;(2)use the guided image to reconstruct the original image;(3)cell image classification with the support vector machine.The advantages of proposed algorithm are that it can keep the edges of the image and gets a higher classification accuracy than the original one.Second,although the sparse classifer has been widely used in the hyperspectral image classification,the pixel-wise level classification ignores the fact that most of the adjacent pixels belong to the same class.Researchs have proposed to extract the spatial information by joint sparse coding using a fixed window,but this interferes with edge information.So,we proposed a method by combining multi-scale super-pixel segmentation and typical sparse representation classifier,and the last classification result is fused by the multi-scale results.Proposed algorithm can adaptively extract spatial features and gets better classification performance.