基于神经网络的红外图像非均匀性校正  

Non-uniformity Correction for Infrared Image Using Neural Networks

作　　者：张龙[1,2] 董峰 傅雨田[1] 

机构地区：[1]上海技术物理研究所中国科学院红外探测与成像技术重点实验室,上海200083 [2]中国科学院大学,北京100049

出　　处：《红外技术》2018年第2期164-169,共6页Infrared Technology

摘　　要：由于红外焦平面材料、制造工艺的差异以及多路模拟信号输出的电路设计等因素,红外探测器在面源黑体目标下像元的输出不均匀,针对同一辐射目标得到的响应也不一致。红外探测器像元间的非均匀性影响目标辐射探测的质量,也使得获得的红外图像不能很好地反应目标辐射特性。先对红外图像进行盲元检测和补偿,通过双边滤波方法获得像元期望输出值,利用随机梯度下降法的神经网络模型对红外图像进行非均匀性校正。实验验证该方法较基于标定的校正方法具有适应场景变化、效果好的优点。For the difference of infrared focal plane array （IRFPA） materials, manufacturing process, and multi-channel circuit design, every pixel on the IRFPA has different responses under surface blackbody, thus the output is not uniform. The non-uniformity of infrared detector pixels （non-uniformity） affects the quality of target radiation detection. The characteristic of the infrared target cannot be determined due to infrared focal plane non-uniformity. In this study, the blind element is detected and compensation is performed, the label output is achieved by bilateral filter, and then the non-uniformity correction of the infrared image is carried out by the neural network. Experimental results show that this method is effective and easy to implement, and it has the advantage of adapting to scene changes.

关 键 词：红外图像 神经网络 非均匀性校正 随机梯度下降 

分 类 号：TN215[电子电信—物理电子学]