基于FPGA的红外图像非均匀校正实现方法

发布时间：2018-05-26 16:17

本文选题：非均匀性校正 + 神经网络算法　；参考：《激光与红外》2016年08期

【摘要】：针对某国产探测器成像特点,对传统两点校正和神经网络非均匀性校正算法进行了改进和定点化处理。对算法实现时的存储和数据流需求进行分析后,利用存储控制器对DDR2高速读写的优势,在以FPGA为核心的红外成像装置预处理平台上实现了校正系数的在线标定和自适应迭代。在系数更新时,引入运动判断环节,以防止神经网络校正算法带来的目标退化和鬼影现象。成像系统仅采用一片FPGA芯片,使得系统小型化成为可能,充足的资源余量使其具有功能可扩充性。实验证明该实现方法明显改善了红外成像装置的非均匀性,在抑制时间漂移上也取得了满意的效果。
[Abstract]:According to the imaging characteristics of a home-made detector, the traditional two-point correction and neural network nonuniformity correction algorithm are improved and fixed-point processing is carried out. After analyzing the requirement of storage and data flow when the algorithm is implemented, the on-line calibration and adaptive iteration of correction coefficients are realized on the pre-processing platform of infrared imaging device based on FPGA, using the advantage of memory controller to read and write DDR2 at high speed. In order to prevent the target degradation and ghost phenomenon caused by the neural network correction algorithm, the motion judgment is introduced when the coefficients are updated. Only one FPGA chip is used in the imaging system, which makes the system miniaturized and has sufficient resource margin to make it functional and extensible. The experimental results show that the proposed method can improve the nonuniformity of infrared imaging device and achieve satisfactory results in suppressing the time drift.
【作者单位】：中国空空导弹研究院;
【分类号】：TP391.41;TN219

【相似文献】