基于亚像素配准的神经网络非均匀性校正

摘要

红外焦平面存在严重影响成像质量的非均匀性,本文使用基于亚像素配准算法和动量项BP神经网络的非均匀性校正算法进行校正.对短波红外相机成像过程中,由于相机视轴与成像目标位置的相对偏移(由相机安装平台晃动所致),使用基于矩阵乘法的亚像素配准算法进行配准;为了加速算法收敛,采用两点法来对校正系数进行初始化;为了改善BP神经网络容易陷入局部最优值,采用增加动量项的方法来改善校正效果.通过仿真实验可以看出提出的算法消除了传统神经网络校正方法存在的鬼影和边缘模糊等问题,获得了良好的校正效果,同时提高了算法的收敛速度.为短波红外图像数据后期处理提供了良好的基础.%Non-uniformity seriously affects the imaging quality of infrared focal plane.To solve the problem,a non-uniformity correction algorithm is proposed based on the sub-pixel registration and the momentum-based BP neural network.Because of the shift of optical axis and imaging target location in the shortwave infrared camera,the sub-pixel registration algorithm based on matrix multiplication is used for registration.In order to accelerate the convergence of the algorithm,the two-point method is used to initialize the correction coefficient.As the BP neural network is easy to fall into the local optimal value,the correction effect is improved by increasing the momentum of the neural network method.The simulation results show that the proposed algorithm can eliminate the problems of ghost and edge blur in the traditional neural network correction method,obtain good correction effect and can improve the convergence speed of the algorithm,which provides a good foundation for short-wave infrared post-processing.