Research On Neurodynamic Approach To Image Fusion Based On Linear Constrained Least Square Method

Since 1970 s,with the wide application in military,national defense,aviation and other high-tech fields,multi-sensor data fusion has become a hot research field.Multisensor data fusion technology refers to the combination,association and combination of data measured by multiple sensors at different times in the case of uncertain original information,so as to obtain more comprehensive and reliable data information.Its main feature is to fuse the complementary information from different sources to improve the fusion performance,so it plays an important role in image denoising.With the development of artificial neural network,neural network has shown great advantages in parallel computing and large-scale data processing,so the multi-sensor image information fusion technology has been developed rapidly.This paper mainly studies a neurodynamics method based on linear constrained least square(LCLS),which can improve the quality of image fusion and the ability of image denoising.The neurodynamics method based on LCLS model can not only solve the problem of image fusion with unknown noise covariance information,but also overcome the singularity of noise covariance matrix.The basic idea of image fusion based on LCLS model is to obtain the optimal weight coefficients in channels R,G and B of color images by minimizing the error square function between the observed data and the actual data.Then the image information acquired by the sensor is multiplied by the weight coefficient,and the fusion image under each channel is obtained by linear weighting.Finally,the fusion information of three channels is synthesized to obtain the de-noised image.In order to obtain the optimal weight coefficient of each channel,this paper introduces a recursive neural network algorithm,which is a continuous time neural dynamic fusion algorithm.It can overcome the singularity of the sample variance matrix in LCLS model,and can deal with the linear conditional constraints simply and effectively.Finally,it is proved that the recursive neural network can converge to the global optimal fusion weight.The simulation results verify the effectiveness and reliability of the neurodynamic method.