一种ICMOS相机关键组合器件响应标校及降噪方法  

A Response Calibration and Noise Reduction Method for Key Combination Devices in ICMOS Camera

作　　者：刘皓挺 葛建跃 杨少华 李刚 严明 刘璐 LIU Haoting;GE Jianyue;YANG Shaohua;LI Gang;YAN Ming;LIU Lu(National Key Laboratory of Intense Pulsed Radiation Simulation and Effect,Xi’an 710024,China;School of Automation and Electrical Engineering,University of Science and Technology Beijing;Beijing Engineering Research Center of Industrial Spectrum Imaging:Beijing 100083,China)

机构地区：[1]强脉冲辐射环境模拟与效应全国重点实验室,西安710024 [2]北京科技大学自动化学院 [3]北京市工业波谱成像工程技术研究中心,北京100083

出　　处：《现代应用物理》2024年第1期44-53,共10页Modern Applied Physics

基　　金：国家自然科学基金资助项目(61975011);强脉冲辐射环境模拟与效应全国重点实验室基金资助项目(SKLIPR2024);广东省基础与应用基础研究基金(2023A1515010275)。

摘　　要：针对增强型互补金属氧化物半导体(intensified complementary metal oxide semiconductor,ICMOS)相机中光纤光锥和CMOS传感器关键组合器件响应标校及其在实际应用中的降噪问题,采用K均值异常点剔除、最小二乘理想线性响应拟合、多项式响应误差补偿拟合的成像系统响应标校方法,有效降低了系统固定模式噪声、单个像素响应非线性、整场像素响应非均匀性对ICMOS相机响应输出的影响。采用金字塔真实图像降噪网络(pyramid real image denoising network,PRIDNet)的成像系统图像降噪方法,进一步抑制了ICMOS相机应用过程随机噪声对相机成像效果的影响。研究结果表明:与传统方法中仅做固定模式噪声处理的标校方法相比,应用本文标校方法,像素平均线性度指标提高了约7%,整场响应均匀度指标提高为原来的2.2倍。与其他类型网络相比,采用PRIDNet降噪方法,其峰值信噪比在不同环境光输入条件下均提高了3%~5%。To address the response calibration of key combination devices of fiber optic tapers and CMOS sensors in intensified complementary metal oxide semiconductor(ICMOS)cameras and its noise reduction in practical applications,a response calibration method for imaging system is proposed.The method includes K-means clustering with outlier removal pattern recognit,linear least-squares fitting,and polynomial fitting of response error compensation.The effects of fixed pattern noise,single pixel response nonlinearity,and non-uniformity of the entire field pixel response on the response output of ICMOS camera are effectively reduced.An image denoising method for imaging system using the pyramid real image denoising network(PRIDNet)is also developed.The impact of random noise on imaging effect during ICMOS camera application is further suppressed.The research results show that compared with the traditional calibration method that only uses fixed pattern noise reduction,the calibration method proposed in this paper improves the average pixel linearity index by about 7%,and the overall response uniformity index by 2.2 times.Compared with other types of networks,the peak signal-to-noise ratio(PSNR)has increased by 3%5%under different ambient light input conditions by using the PRIDNet denoising method.

关 键 词：ICMOS 响应校正 图像降噪 多项式拟合 K均值 

分 类 号：TL81[核科学技术—核技术及应用] TB85[一般工业技术—摄影技术]