3PEMs-AOTF型偏振光谱成像多维信息获取关键问题研究

Spectropolarimetric imaging (SPI) technology is a novel optical detection approach which can collect multiple dimensional such as image, spectrum and polarization information simultaneously. Rapid and accurate obtaining these data can significantly improve the identification performance of SPI systems. Hence, it has become a research hotspot in recent several years. This project aims to solve the key theoretical and technical problems of multi-dimensional information rapid and accurate obtaining by using digital signal processing method, on the basis of three Photoelastic modulators based on differential frequency modulation (3PEMs) polarization measuring theory and acousto-optic tunable filter (AOTF) spectral imaging theory. Firstly, the mechanism of image blurring which is caused by chromatic aberration, the diffraction angle spread and the wavelength spread during the image obtaining process will be investigated, and then regularization method will be adopted to establish the image restoration model. Secondly, the coupling relationship between wavelength of diffraction direction, incident angle, ultrasound driving frequency and temperature during the spectrum obtaining process will be studied, and then spectrum modification model will be established based on artificial intelligent algorithms. Thirdly, the fast computation model of polarization information will be explored by combining sparse fast Fourier transform (SFFT) algorithm and heuristic Monte Carlo method. The research results and proposed methods can not only effectively solve the rapid and accurate obtaining problem of multi-dimensional information of 3PEMs-AOTF-SPI system, but also can promote the development of information obtaining theory and techniques of other type SPI systems.

偏振光谱成像（SPI）技术是集图像、光谱、偏振等多维信息于一体的新型光学探测技术。多维信息准确、快速获取可大幅提高偏振光谱成像系统的探测识别能力，是目前国内外研究热点。本项目基于三弹光差频调制(3PEMs)偏振测量理论、声光滤波(AOTF)光谱成像理论，结合信号处理方法，旨在解决多维信息准确、快速获取关键理论和技术问题。项目重点研究图像信息获取时由系统色差、衍射方向波长与角度展宽等因素引起的图像退化机理，利用正则化方法建立图像复原模型；研究光谱信息获取时衍射光波长与入射角、超声驱动频率、温度等因素间多变量耦合关系，利用人工智能算法建立光谱修正模型；研究偏振信息快速反演问题，结合稀疏FFT与启发式蒙特卡洛方法建立偏振快速反演模型。研究结果及分析方法不但可以有效解决弹光调制型AOTF-SPI系统多维信息的准确、快速获取问题，而且将有助于SPI多维信息准确、快速获取理论及技术的进一步发展。

本项目紧紧围绕3PEMs-AOTF型偏振光谱成像系统的多维信息的准确、快速获取问题展开研究，具体研究内容包括：（1）图像退化机理及复原研究方面，通过深入分析3PEMs-AOTF-SPI系统的成像原理，得出导致图像退化的因素主要包括两个：AOTF声光晶体色差导致的失焦和衍射光方向与波长发生展宽。本项目提出了一种基于卷积神经网络的图像模糊核函数参数估计模型。同时，考虑到多光谱成像各个通道的图像间存在相关性的特点，本项目提出了基于残差U-Net深度神经网络的多光谱图像预测方法；（2）光谱修正研究方面，为了克服现有技术中AOTF成像中同一幅图光谱不均匀的问题，本项目提出了一种采用二次拟合来修正光谱的方法，首先通过实验建立了环境温度与AOTF衍射光中心波长间的关系，其次利用最小二乘法建立了环境变量、入射角、超声驱动频率与衍射光中心波长间的非线性映射模型，最后采用修正超声驱动频率的方式对衍射光波长的温漂问题进行补偿；（3）偏振信息快速反演研究方面，基于现实图像目标物体和背景的像元像素值差异等先验知识，提出了基于SFFT与启发式蒙特卡洛方法的偏振信息快速反演算法；（4）平台搭建研究方面，基于本项目组自行设计的光学系统及弹光调制器搭建了3PEMs-AOTF-SPI实验样机，并在此基础上对多光谱图像复原算法、图像预测算法、光谱修正模型和偏振信息快速反演算法等进行了参数优化。（5）除了项目计划任务书中设定的研究内容外，本项目在光谱分析建模方面也展开了研究，提出了基于深度神经网络的红外光谱定性/定量分析模型及跨组分模型的移植方法。项目执行期间，共发表学术论文12篇，其中SCI收录10篇，EI收录1篇，中文核心1篇；第一作者论文6篇，通信作者论文10篇。共申请发明专利5项，其中已授权1项，实质审查阶段4项。共培养了5名与本项目相关的硕士研究生，其中1人获得校级优秀硕士论文奖。