改进的红外图像神经网络非均匀性校正算法

Improved infrared image neural network non-uniformity correction algorithm

红外焦平面阵列(IRFPA)像元响应存在不一致性,会严重影响红外成像系统成像的质量,实际应用中需要采用响应的非均匀性校正(NUC)技术。传统的神经网络校正算法在校正结果中存在图像模糊和伪像的问题,影响人们对于目标的观察。在分析了传统的神经网络性校正算法所出现问题原因的基础上,提出了有效的改进算法:用非线性滤波器代替传统算法中使用的均值滤波器。算法改进之后所得到的校正图像,不仅在清晰度方面有明显的改善,而且有效的消除了传统算法中存在伪像的问题。

The responsive of infrared focal plane arrays（IRFPA） is different; it will affect the quality of imaging system seriously. Non-uniformity correction technology will need in practical application. The calibrated images have the problems of blurring and existing ghost artifacts when use the traditional neural network correction algorithm. And it is bad for the observation of the target. After analysis the reasons for the problems in the traditional neural network correction algorithm,proposed the improved algorithm. Replace the mean filter, which used in the traditional algorithm, by the nonlinear filter. The corrected image by the improved algorithm not only a marked improvement in clarity, but also effectively eliminate the problem of artifacts in traditional algorithms.