红外焦平面阵列非均匀性校正的改进神经网络算法  

An Improved Neural Network Algorithm for Nonuniformity Correction of IRFPA

作　　者：陈宝国[1,2] 樊养余[1] 张学峰[2] 王巍[1,2] 

机构地区：[1]西北工业大学,陕西西安710072 [2]中国空空导弹研究院,河南洛阳471009

出　　处：《红外技术》2012年第12期690-694,共5页Infrared Technology

摘　　要：非均匀性校正是红外焦平面阵列应用中的关键技术之一。神经网络算法是比较传统的非均匀性校正算法,由于该算法采用像元四邻域的平均值作为此像元的真值,所以这种估计方法具有较大的误差。在传统的神经网络算法基础上对焦平面阵列像元响应的真值估计进行了改进:基于图像匹配算法,采用了相邻多帧图像中不同像元对同一场景点的响应的均值作为真值,因而具有更高的准确性。对比仿真试验的结果表明,该改进算法比传统的神经网络算法具有更好的效果,在有效去除各种非均匀性的同时,保持了图像细节,改善了图像的视觉效果。Nonuniformity correction is one of the key technologies in the application of infrared focal plane array （IRFPA）. Neural network algorithm is a traditional nonuniformity correction algorithm. The average of four neighbor pixels is adopted as the real response of their centre pixel in the algorithm. Thus error is induced by this method. An improved neural network algorithm is presented in this paper. The algorithm is based on image registration and the average of different pixels responses to the same field point in multi frames is adopted as the real response, which can reduce error. The effect of this advanced algorithm is proved in simulation. Results show that the improved algorithm could eliminate nonuniformity effectively. In the meantime, image details are retained and imae quality is then improved.

关 键 词：红外焦平面阵列 非均匀性校正 神经网络算法 图像配准 固定图形噪声 随机噪声 

分 类 号：TN215[电子电信—物理电子学]