针对传统的基于神经网络的自适应非均匀性校正(Neural-Network-based Non-Uniformity Correction, NN-NUC)算法在 实际应用中存在校正能力有限和容易产生鬼影的问题，深入分析了NN-NUC算法中的鬼 影产生过程，并给出了抑制鬼影的一般性方法；然后结合实际红外成像系统的特点，提出 了一种改进型NN-NUC算法。仿真实验结果表明，该算法可以最大限 度地抑制场景鬼影的产生，并可有效减小系统输出图像的非均匀性噪声。此外，本文算法 计算量小，且易于用硬件实现，因此具有很好的工程应用价值。

In practical applications, the traditional adaptive nonuniformity correction algorithm based on Neural Network (NN-NUC) has a limited correction capability and is easy to generate ghosting artifacts. To solve this problem, the ghosting artifact generating process in the NN-NUC algorithm is analyzed in detail and the common methods for removing ghosting artifacts are given. Then, by incorporating the characteristics of actual infrared imagers, an improved NN-NUC algorithm is proposed. The simulation experimental results show that the proposed method can suppress the generation of ghosting artifacts in a scene extremely and can reduce the nonuniformity noise of the image effectively. Moreover, the proposed algorithm has a small calculation amount and is easy to be implemented by hardware. So it is of good value to practical applications.