基于神经网络的少光子探测信号处理
Few Photon Detection Signal Processing Based on Neural Network
宋贺良 1李少波 1李华贵 1刘旭超1
作者信息
- 1. 中国电子科技集团公司第五十四研究所,河北 石家庄 050081;河北省光子信息技术与应用重点实验室,河北 石家庄 050081
- 折叠
摘要
盖革模式雪崩光子二极管(GM-APD)阵列探测器可提高少光子探测信号的信噪比,在激光通信和光子雷达信号处理领域受到广泛关注.针对由于发射功率低和背景噪声强等因素探测器无法探测到回波光子信号的问题,建立基于时域空域卷积神经网络信号处理的数学模型,时域上叠加相邻4帧的回波信号,空间域上先使用矩阵维度拓展算法拓展卷积核维度,再经卷积神经网络提取回波光子信号.仿真结果表明,该方法能够有效从噪声信号中提取回波光子信号,并将信噪比提升4.5倍,在百公里级的少光子探测信号处理领域具有重要参考意义.
Abstract
Geiger-mode avalanche photon diode(GM-APD)array improving the signal-to-noise ratio(SNR)has been widely concerned in laser communication and photonic radar.However,due to low transmitting power and strong background noise,the SNR is also low in low-photon detection signal processing.In order to solve the problem,we establish a mathematical model of signal processing based on time-domain and spatial-domain convolutional neural network.The model superimposes echo signals of adjacent four frames in the time domain.In the spatial domain,matrix dimension expansion algorithm is used to expand the convolution kernel dimension,and then echo photon signals are extracted through convolutional neural network.The results show that method can effectively extract the echo photon signal from the noise signal and improve the SNR by 4.5 times.This article can provide references for the hundred-kilometer low-photon detection signal processing.
关键词
少光子探测/GM-APD/矩阵维度拓展/卷积神经网络Key words
few photon detection/GM-APD/matrix dimension expansion/convolutional neural network引用本文复制引用
基金项目
河北省省级科技计划(22310701D)
"十四五"装备预先研究专用技术(0722-227FJ544YPF)
出版年
2024
NETL
NSTL
万方数据