Research On The Adaptive Correction Algorithm Of Infrared Detector

Due to the material and manufacturing process problems,the infrared detector will appear the phenomenon that the responses of different detector elements to the same radiation are inconsistent.A more complex problem is that the parameters of detector elements response drift slowly over time,making it impossible to completely solve the problem of nonuniformity by a single calibration method.Scene-based nonuniformity adaptive correction method can not completely avoid the "ghosting" artifacts,and the algorithm has the problems of low real-time performance and poor robustness.This paper mainly aims at the problem of poor correction effect such as ghost in scene-based correction methods,the nonuniformity correction method of single-frame infrared detector is studied.The main research contents are as follows:(1)The nonuniformity response distribution characteristics of infrared detectors are analyzed statistically,according to the spatial distribution characteristics of nonuniformity response of infrared detectors,an adaptive correction method based on image block prior single-frame infrared image is proposed.Firstly,the probability density statistics and the solution of the probability density function are carried out respectively for the overall distribution characteristics of the nonuniform response of the detector and the randomly selected distribution characteristics in the region of interest.The analysis results show that the mixed Gaussian model can describe the non-uniform response of the detector well.Based on the analysis results,the mixed Gaussian model is introduced into the infrared detector response model as a statistical prior regularization function,and the parameters of the mixed Gaussian model are obtained by the maximum likelihood estimation method.Then,the real thermal radiation of the scene is estimated by the optimization solution,and the nonuniformity response correction result of the infrared detector is obtained.Through the non-uniformity correction test of simulated infrared data and real infrared data,it is verified that the method can complete the non-uniformity correction in a single frame.(2)Aiming at the problem that the nonuniformity correction method of middle infrared histogram equalization doesn't fully consider the scatter distribution in the nonuniformity response of infrared detectors,an adaptive middle equilibrium autoregressive prediction nonuniformity correction method is proposed by introducing a piecewise autoregressive prediction model.Firstly,the algorithm completes the approximate original thermal radiation estimation of the single pixel of the infrareddetector,and then completes the nonuniformity response correction of the mesh distribution by combining the middle infrared histogram equalization nonuniformity correction method,so as to realize the correction of the nonuniformity response in the single-frame infrared detector response.The experimental results show that the proposed algorithm can correct the nonuniformity response within a single frame and the nonuniformity response of scatter distribution is corrected by comparing the nonuniformity correction method of middle infrared histogram equalization with the correction results of the proposed algorithm.(3)An adaptive correction method for single frame infrared image based on convolution neural network is proposed.By assuming that the two-point correction result is an estimate of the real thermal radiation response,a training sample set is produced using the two-point correction result and the corresponding real infrared detector response.The depth convolution neural network is introduced and the loss function of the network is designed according to the similarity of the data set elements.the nonuniform response model of the detector is obtained by learning the " real" thermal radiation response and the detector's original response.Compared with the traditional scene-based correction method,this method does not need to rely on a priori knowledge,and experimental results show that the details in the correction results are relatively complete.