为解决传统基于神经网络的红外图像非均匀校正算法存在的目标模糊、拖影等问题, 提出了一种增强型神经网络方法。该方法首先采用边缘保护滤波器得到期望值, 以达到利用景像的边缘信息来指导校正系数更新的目的, 在此基础上, 通过自适应学习率以稳定和加速学习过程, 实验结果表明, 该方法解决了目标模糊和拖影问题, 同时有效改善了非均匀性校正的效果和效率。

The traditional non-uniformity correction algorithm of infrared image based on neural network exists problems of the ghosting artifact and the target fade-out. To overcome these problems, the enhancement neural network method is proposed. It firstly obtains expected values by the edge-preserving filters, in order to guide correction coefficient updating by using the edge of the picture information, and stabilize and accelerate the learning process by using self-adaptive learning rate. The simulating experiment indicates that the new algorithm not only overcomes the problems of the ghosting artifact and the target fade-out, but also fairly reduces the non-uniformity.