基于散斑场退化补偿的水下鬼成像
For the problem of poor reconstruction quality and resolution degradation of underwater ghost imaging, an underwater ghost imaging method based on speckle degradation compensation was proposed to recover the target image degraded by the water body. Compared with ghost imaging in air medium, underwater ghost imaging has been studied by scholars in many aspects, such as the absorption effect of the water body, signal-to-noise ratio detection of the system, backward scattering noise, underwater illumination spot, and underwater turbulence. Image degradation and recovery methods based on underwater optical transmission models have been used in array detector optical imaging. However, there is no relevant study to analyze and solve the problem of degradation of underwater ghost imaging starting from the inherent optical properties of the water body. The scattering effect of the water body on the beam reduces the contrast of the speckles shining on the surface of the target and degrades the resolution, which deteriorates the intensity fluctuation characteristics of the target obtained by bucket detection, thus affecting the reconstruction quality of ghost imaging. Therefore, we hope to recover the underwater ghost imaging results affected by the water body through a method similar to deconvolution by introducing a point spread function (PSF) related to the intrinsic optical parameters of the water body.
In this research, the water body scattering degradation model was introduced into the ghost imaging image reconstruction to improve the image quality. First, the S-S (Sahu-Shanmugam) scattering phase function was linearly approximated in logarithmic coordinates in a small angular range (
In this study, the matrix form of the speckle degradation compensation method is derived theoretically. The mathematical nature of the correction compensation of the reference arm speckle before reconstruction by the second-order correlation algorithm or the pseudo-inverse algorithm is analyzed. Equation (16) shows that in the second-order correlation calculation, the reference arm speckle is convolved with the PSF of the water body equivalent to the image convolved with the PSF of the water body obtained by second-order correlation for the original non-degradation compensation. Therefore, this method, for second-order correlation reconstruction, will make the reconstruction effect doubly degrade. As shown in Eq. (17), the pseudo-inverse ghost imaging with speckle degradation compensation is essentially a method of deconvolution by obtaining the convolution kernel of the PSF of the water body from the optical parameters of the water body. If the correction compensation of reference arm speckles is consistent with the actual degradation of the object arm, the degradation of the water body can be better removed. The simulation results and experimental validation results are shown in Fig. (4) and Fig. (6), respectively. The reconstruction results of the second-order correlation algorithm with speckle degradation compensation deteriorate the image quality compared with the original second-order correlation algorithm. The image quality and resolution of the reconstruction results of the pseudo-inverse algorithm with speckle degradation compensation are significantly improved compared with the original pseudo-inverse algorithm.
In this study, an MTF of the water body that can describe the underwater speckle transmission is derived, and the reference arm speckle is corrected with the same degree of degradation as the object arm speckle, so as to compensate for the degradation of the object arm speckle. The method restores the congruence between the object arm speckle and the reference arm speckle and then performs the reconstruction calculation of the target image. Through theoretical analysis, simulation, and experiments, it is proved that the spot degradation compensation will aggravate the image degradation for the second-order correlated image reconstruction, while it can improve the image resolution and imaging quality for the pseudo-inverse reconstruction. The method has some degradation removal effect for pseudo-inverse algorithm and greedy algorithm based on least squares in underwater target image reconstruction. Unlike blind deconvolution, the accuracy of the method depends on the accuracy of the MTF or PSF of the water body, and the improvement of the image reconstruction quality characterizes the correctness of the derived MTF. The method is essentially a deconvolution method based on the scattering model of the water body, which generates ringing artifacts and noise amplification in the case of the low signal-to-noise ratio of bucket detection, making the reconstruction quality worse, which is also an important direction for subsequent research.
1 引言
水下探测技术对海洋资源开发、水下救灾救援等具有重大意义。水下光学成像是一种高分辨、携带信息丰富、应用前景广阔的水下探测技术。近年来,鬼成像雷达作为一种非常规主动成像方式,以其探测距离远、抗干扰、高分辨的特点[1-4],在弱光和复杂水下环境中的应用被很多学者关注。
一些文章已经从不同角度验证了鬼成像在散射介质成像中所具有的独特优势,Gong等[5]较早地将鬼成像应用于浑浊介质中的目标成像,并验证了鬼成像可以减小回程光路中散射介质对成像质量的影响;Le等[6]研究了不同浑浊度、不同视场角度条件下的计算鬼成像,验证了在经典光学成像方法完全失效的情况下鬼成像仍能获得令人满意的成像结果。为了探究水下湍流对鬼成像的影响,Zhang等[7]研究了忽略水体散射和吸收的海洋湍流鬼成像物理模型,吴泳波等[8]从实验角度研究了水下鬼成像的抗扰动能力特性。为解决水体的后向散射对鬼成像的影响,Chen等[9]根据水下激光雷达方程分析了脉冲激光鬼成像中影响成像性能的系统参数和水体参数,以控制后向散射的影响;Wu等[10]将偏振光应用于水下鬼成像以滤除后向散射。在水下照明散斑场的研究中,哈达玛散斑鬼成像在水下探测中具有明显优势[11-12]。在水下图像复原算法方面,DGI(Differential Ghost Imaging)[13]、直方图拉伸偏振差分鬼成像[14]、小波增强的水下压缩感知计算鬼成像[15]、基于深度学习的卷积神经网络和对抗网络水下鬼成像重构方法[16-17]都能有效提高重构质量。
与上述水下鬼成像研究不同,本文将水体散射退化模型引入鬼成像图像恢复重构以提高图像质量。本文根据近似的S-S(Sahu-Shanmugam)散射相函数[18]和Wells模型[19]推导得到水体的调制传递函数(MTF),将其用于散斑场退化描述,并将得到的MTF用于参考臂散斑的校正以补偿目标面物臂散斑退化;同时,从理论上研究了参考臂散斑校正补偿在二阶关联方法和伪逆方法重构图像过程中的作用。在保证探测能量和探测信噪比的前提下,实验与理论结果很好地匹配。
2 基本原理
2.1 鬼成像原理及影响其水下成像能力的主要因素
经典的鬼成像由物臂和参考臂两个光路组成,通过将物臂所记录的桶探测值和参考臂记录的散斑强度分布进行二阶关联运算,得到目标图像。
图 1. 赝热光鬼成像光路示意图
Fig. 1. Schematic diagram of optical path for pseudo thermal light ghost imaging
通过二阶关联计算得到的目标图像
式中:
将上述传感器记录的数据用矩阵或者向量来表示,可以得到采样矩阵
采样矩阵
由
式中:
与空气介质中的鬼成像相比,影响水下鬼成像的因素主要包括回波能量、探测信噪比、水体散射作用、水下湍流影响等。水分子和有色可溶性有机物对不同波长光的吸收作用使得传输的信道光能量大大衰减;水中的颗粒物使得光束在传播过程中偏离原来的传播方向,产生前向散射和后向散射。前向散射使得目标表面散斑退化,进而使得成像结果退化;后向散射返回桶探测器中,使得探测信噪比变差,图像噪声增加。相比于受温度起伏影响的大气湍流,海洋湍流同时受到温度、盐度等起伏的综合影响,这使得水体的折射率产生复杂变化,产生光束扩展、光束漂移、光强闪烁,将严重影响真实环境下远距离水下鬼成像的成像性能。
水体的散射吸收效应使得光在水下传播时能量以指数量级衰减,系统探测端接收到的后向散射噪声随着距离的增加先增大后减小,目标回波能量则呈单调指数衰减,探测能量和信噪比是影响水下鬼成像的重要因素。西安交通大学[6]从实验上研究了距离、水体浑浊度、视角对鬼成像的影响,散斑浑浊度和距离对成像的影响显著,随着视角的增加,传统成像成像质量的下降程度比鬼成像大,说明水下鬼成像具有更大的视角优势。长春理工大学[9]根据双向反射分布函数和水下激光雷达方程分析影响水下鬼成像的因素,从理论和实验上验证了系统的探测性能依赖于目标反射特性、入射角、目标距离和介质衰减引起的目标强度的变化。真实水下场景中湍流的作用不容忽视,水下湍流的强度、湍流尺度、温度盐度、传播距离等参量对成像效果同样有很大的影响。华东交通大学[22-23]在海洋湍流为Kolmogorov微流的前提下,将空间功率谱引入湍流强度模型,分析了传播距离和海洋湍流强度对鬼成像的影响。中国海洋大学[24]根据海洋湍流的功率谱和扩展的惠更斯-菲涅耳积分,建立了与散射吸收无关的海洋湍流脉冲响应函数和鬼成像可见性的理论表达式。潍坊学院[25]采用广义惠更斯-菲涅耳原理和海洋湍流Rytov近似,理论推导出反射式鬼成像在Kolmogorov海洋湍流中的脉冲响应函数表达式。上述关于水下湍流的研究都表明单位质量流体湍流的动能耗散率、均方温度耗散率、温度和盐度波动的相对强度值对鬼成像的成像质量有直接影响。
与空气中的鬼成像相比,水体对光束的散射作用使得照在目标表面的散斑的对比度和分辨率降低,桶探测得到的目标强度涨落特性变差,从而影响鬼成像的重构质量。本文对水体散射引起的鬼成像结果退化进行了分析,提出在重构计算前通过将参考臂散斑进行校正实现对物臂散斑场退化补偿的方法,同时以二阶关联和伪逆算法为例,将所推导的水下光学传输模型用于对物臂目标表面散斑退化的描述,该方法在不同重构算法下进行了理论推导分析和实验验证。
2.2 散斑场水下传输模型
当物臂散斑光束在水中传输时,水体的散射吸收作用会将光束扩展并使其强度减弱。为了描述光在稳定均匀水体中的传播,许多学者将传播过程描述为与固有光学特性(IOPs)(如水体散射相函数,水体的吸收系数、散射系数)有关的点扩散函数(PSF)或MTF[26]。Duntley[27]、Wells[19]和Voss[28]先后通过实验或理论推导得到PSF或MTF模型,Wells模型能够在较大的水体参数范围内精确描述水体中的光束传输过程。本文将文献[18]提出的S-S散射相函数进行小角度范围(0.1°~5.0°)内的对数函数线性近似,进行Hankel变换,将Wells的MTF模型代入,即可得到图像在水中传播的MTF。MTF是PSF的频域表达形式,能够较好地描述真实海水对散斑的退化作用。Wells[19]从第一性原理出发推导得到MTF模型,将散射相函数、传输距离、散射系数、衰减系数用于水体对光束传输的描述,文献[29]验证了该模型的正确性。通过水体传播到目标表面的散斑光强分布
式中:
水体PSF被描述为由IOPs(体散射函数、散射系数、衰减系数等)和传输距离建模得到。由于鬼成像视场角较小(本文实验装置的成像视场角为0.43°),本文只关心小角度范围的散射相函数拟合精度,S-S散射相函数[18]是一个分段函数,其在0.1°~5.0°角度范围内的散射相函数
式中:
式中:
式中:B为归一化常数。将归一化后的散射相函数进行Hankel变换,得到
式中:ψ为空间角频率;
本文将推导得到的
式中:
式中:
2.3 参考臂散斑退化补偿的水下伪逆鬼成像
在传统水下成像及其图像复原方法中,通过反卷积图像复原方法能够很好地提高图像质量[31]。这类解卷积优化水下图像质量的方法大致包括两种:一种是基于物理模型的解卷积方法,即采用水下固有光学参数构建退化模型,以反卷积方式恢复目标图像[31-32];另一种是在没有图像退化的必要先验知识的前提下,利用原始模糊退化图像,以某种方式提取退化信息估计,即同时估计PSF和清晰图像的盲解卷积方法[33-37]。本文所提出的散斑退化补偿的水下鬼成像方法可以看作是基于物理模型的图像恢复方案。
通过所推导的PSF或MTF获得水体对物臂散斑退化效果的近似描述,提出在重构计算前将参考臂散斑进行校正以实现对物臂散斑场退化补偿的方法,不同重构方法的补偿效果不同。对于二阶关联类方法,该补偿方法会使得重构图像变得更模糊;而基于伪逆算法或基于最小二乘的贪婪算法更关注照射在目标表面的光场强度分布与用于重构计算的参考臂散斑强度分布之间的一致性,因此对于该类算法,可以在重构之前先对参考臂散斑进行校正补偿,以提高图像分辨率和成像质量。
式中:
参考臂散斑校正补偿的二阶关联运算可表示为
式中:
先对参考臂散斑进行校正补偿再进行伪逆重构的过程可表示为
式中:
3 仿真与实验验证
本文分别采用仿真计算和实验来验证所提算法的有效性。
实验环境为室内某船舶拖曳水池,实验时水体为近似恒温静止水体。采用高光谱衰减测量仪(ACS)测得水体条件为衰减系数
图 2. 水下赝热光鬼成像实验原理及装置图。(a)水下鬼成像原理图;(b)水下鬼成像装置图
Fig. 2. Experimental principle and device diagram of underwater pseudo-thermal light ghost imaging. (a) Underwater ghost imaging principle diagram; (b) underwater ghost imaging device diagram
表 1. 不同算法针对仿真数据的重构图像峰值信噪比
Table 1. Peak signal-to-noise ratio of reconstructed images of simulation data obtained by different algorithms
|
本文仿真过程中是通过将赝热光相干传输到目标表面再将其与非相干PSF卷积实现的。相干光通过动态散射介质时,散射光相位被打乱,这使其传输到目标表面时不能形成散斑,只有未被散射的信道光相干传输到目标表面才能形成散斑场。信道光在水体中的传播等价于在折射率为纯水的无散射介质空间中的相干传输,而散射光在散射介质中的相位变成随机的,其散射传输过程和非相干光在水体介质中的传播类似,仿真中采用本文引入的PSF进行模拟。仿真的水体条件与实验水体条件相同,
如
图 4. 仿真数据的重构结果对比图。(a)原始GI结果;(b)散斑DCGI结果;(c)原始PGI结果;(d)散斑DCPGI结果;(e)框图标记的三缝归一化分辨率曲线图
Fig. 4. Comparison of reconstruction results for simulated data. (a) Raw GI results; (b) pattern DCGI results; (c) raw PGI results; (d) pattern DCPGI results; (e) three-slit normalized resolution plot marked in frames
为了验证该方法在真实场景中的有效性,同时为了保证探测信噪比,将三缝目标放置于水下距离光学系统38 m处,图像尺寸为256 pixel×256 pixel。实验过程中,激光器、CCD、PMT在同步控制系统的控制下完成数据的采集,采用脉冲光对目标进行探测,实验装置如
图 5. PMT采样回波曲线图。(a)单次采样回波曲线;(b)多次采样叠加的回波曲线
Fig. 5. Graphs of PMT sampling echo curves. (a) Single sampling echo curve; (b) multi-sampling superimposed echo curve
对3种尺寸的三缝靶标进行重构的计算结果如
图 6. 实验得到的重构结果图。(a)原始GI结果;(b)散斑DCGI结果;(c)PGI结果;(d)散斑DCPGI结果;(e)三缝分辨率曲线图
Fig. 6. Experimental reconstruction results. (a) Raw GI results; (b) pattern DCGI results; (c) PGI results; (d) pattern DCPGI results; (e) three-slit resolution curves
4 结论
推导得到了可对散斑场在水下传输进行描述的水体MTF,通过对参考臂散斑进行与物臂散斑相同退化程度的校正,实现对物臂散斑退化的补偿,使得物臂散斑和参考臂散斑恢复一致性,再进行重构计算。通过理论分析、仿真和实验,发现基于散斑退化补偿进行二阶关联图像重构会加剧图像退化,而伪逆重构则可以提高图像分辨率和成像质量。该方法对于伪逆方法及基于最小二乘法的其他贪婪算法的水下目标图像重构具有一定去退化的作用;区别于盲解卷积,该方法的精度依赖于水体MTF或PSF的精确程度,图像重构质量的改善也表明推导得到的MTF的正确性;该方法本质上是一种基于水体散射模型的解卷积方法,在桶探测信噪比较低的情况下会产生振铃效应和噪声放大,使得重构质量变差,这也是后续研究需要解决的问题。
[1] Shapiro J H, Boyd R W. The physics of ghost imaging[J]. Quantum Information Processing, 2012, 11(4): 949-993.
[2] Moreau P A, Toninelli E, Gregory T, et al. Ghost imaging using optical correlations[J]. Laser & Photonics Reviews, 2018, 12(1): 1700143.
[3] Han S S, Yu H, Shen X A, et al. A review of ghost imaging via sparsity constraints[J]. Applied Sciences, 2018, 8(8): 1379.
[4] 杨莫愁, 吴仪, 冯国英. 水下鬼成像的研究进展[J]. 光学学报, 2022, 42(17): 1701003.
[5] Gong W L, Han S S. Correlated imaging in scattering media[J]. Optics Letters, 2011, 36(3): 394-396.
[6] Le M N, Wang G, Zheng H B, et al. Underwater computational ghost imaging[J]. Optics Express, 2017, 25(19): 22859-22868.
[7] Zhang Q W, Li W D, Liu K, et al. Effect of oceanic turbulence on the visibility of underwater ghost imaging[J]. Journal of the Optical Society of America A, 2019, 36(3): 397-402.
[8] 吴泳波, 杨志慧, 唐志列. 水下鬼成像抗扰动能力的实验研究[J]. 激光与光电子学进展, 2021, 58(10): 1011031.
[9] Chen Q, Mathai A, Xu X P, et al. A study into the effects of factors influencing an underwater, single-pixel imaging system’s performance[J]. Photonics, 2019, 6(4): 123.
[10] Wu H D, Zhao M, Li F Q, et al. Underwater polarization-based single pixel imaging[J]. Journal of the Society for Information Display, 2020, 28(2): 157-163.
[11] Yang X, Liu Y, Mou X Y, et al. Imaging in turbid water based on a Hadamard single-pixel imaging system[J]. Optics Express, 2021, 29(8): 12010-12023.
[12] Wu H, Zhao G P, He C H, et al. Sub-Nyquist underwater denoising ghost imaging with a Coiflet-wavelet-order-based Hadamard matrix[J]. Physical Review A, 2022, 106(5): 053522.
[13] Bina M, Magatti D, Molteni M, et al. Backscattering differential ghost imaging in turbid media[J]. Physical Review Letters, 2013, 110(8): 083901.
[14] 李宜泽, 邓陈进, 龚文林, 等. 浑浊介质下的偏振差分关联成像研究[J]. 光学学报, 2021, 41(15): 1511004.
[15] Wang T, Chen M Y, Wu H, et al. Underwater compressive computational ghost imaging with wavelet enhancement[J]. Applied Optics, 2021, 60(23): 6950-6957.
[16] Li M D, Mathai A, Lau S L H, et al. Underwater object detection and reconstruction based on active single-pixel imaging and super-resolution convolutional neural network[J]. Sensors, 2021, 21(1): 313.
[17] Yang X, Yu Z Y, Xu L, et al. Underwater ghost imaging based on generative adversarial networks with high imaging quality[J]. Optics Express, 2021, 29(18): 28388-28405.
[18] Sahu S K, Shanmugam P. Semi-analytical modeling and parameterization of particulates-in-water phase function for forward angles[J]. Optics Express, 2015, 23(17): 22291-22307.
[19] Wells W H. Loss of resolution in water as a result of multiple small-angle scattering[J]. Journal of the Optical Society of America, 1969, 59(6): 686-691.
[21] Zhang C, Guo S X, Cao J S, et al. Object reconstitution using pseudo-inverse for ghost imaging[J]. Optics Express, 2014, 22(24): 30063-30073.
[22] Luo C L, Li Z L, Xu J H, et al. Computational ghost imaging and ghost diffraction in turbulent ocean[J]. Laser Physics Letters, 2018, 15(12): 125205.
[23] Luo C L, Wan W X, Chen S Y, et al. High-quality underwater computational ghost imaging with shaped Lorentz sources[J]. Laser Physics Letters, 2020, 17(10): 105209.
[24] Zhang Y, Li W D, Wu H Z, et al. High-visibility underwater ghost imaging in low illumination[J]. Optics Communications, 2019, 441: 45-48.
[25] 张钦伟, 曹连振, 刘霞, 等. 反射式鬼成像在海洋湍流中的成像分析[J]. 光子学报, 2020, 49(9): 0901002.
[26] Hou W L, Gray D J, Weidemann A D, et al. Comparison and validation of point spread models for imaging in natural waters[J]. Optics Express, 2008, 16(13): 9958-9965.
[27] DuntleyS Q. Underwater lighting by submerged lasers and incandescent sources, instituition of oceanography[D]. San Diego: University of California, 1971.
[28] Voss K J. Simple empirical model of the oceanic point spread function[J]. Applied Optics, 1991, 30(18): 2647-2651.
[29] Jaffe J S. Monte Carlo modeling of underwater-image formation: validity of the linear and small-angle approximations[J]. Applied Optics, 1995, 34(24): 5413-5421.
[30] PetzoldT J. Volume scattering functions for selected ocean waters[R]. San Diego: Scripps Institute of Oceanography, 1972: 72-78.
[31] 郭继昌, 李重仪, 郭春乐, 等. 水下图像增强和复原方法研究进展[J]. 中国图象图形学报, 2017, 22(3): 273-287.
Guo J C, Li C Y, Guo C L, et al. Research progress of underwater image enhancement and restoration methods[J]. Journal of Image and Graphics, 2017, 22(3): 273-287.
[32] Chen Y Z, Xia M, Li W, et al. Comparison of point spread models for underwater image restoration[J]. Optik, 2012, 123(9): 753-757.
[33] FanF, YangK C, FuB, et al. Application of blind deconvolution approach with image quality metric in underwater image restoration[C]//2010 International Conference on Image Analysis and Signal Processing, April 9-11, 2010, Zhejiang, China. New York: IEEE Press, 2010: 236-239.
[34] Fan F, Yang K C, Xia M, et al. Comparative study on several blind deconvolution algorithms applied to underwater image restoration[J]. Optical Review, 2010, 17(3): 123-129.
[35] Zhang J L, Zhang Q H, He G M. Blind deconvolution: multiplicative iterative algorithm[J]. Optics Letters, 2007, 33(1): 25-27.
[36] Lane R G. Blind deconvolution of speckle images[J]. Journal of the Optical Society of America A, 1992, 9(9): 1508-1514.
[37] ChengC Y, SungC C, ChangH H. Underwater image restoration by red-dark channel prior and point spread function deconvolution[C]//2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA), October 19-21, 2015, Kuala Lumpur, Malaysia. New York: IEEE Press, 2016: 110-115.
李育亮, 齐金泉, 陈明亮, 邓陈进, 邵学辉, 陶邦一, 韩申生. 基于散斑场退化补偿的水下鬼成像[J]. 光学学报, 2024, 44(6): 0601003. Yuliang Li, Jinquan Qi, Mingliang Chen, Chenjin Deng, Xuehui Shao, Bangyi Tao, Shensheng Han. Underwater Ghost Imaging Based on Speckle Degradation Compensation[J]. Acta Optica Sinica, 2024, 44(6): 0601003.