基于水下湍流关联成像的方法对比及神经网络去模糊研究
Comparison of Underwater Turbulence Correlation Imaging Methods and Neural Network Deblurring Research

水下湍流对光学成像的影响主要表现为光波在传播过程中因湍流的相位扰动而导致的散斑、模糊及失真现象，这严重制约了传统成像方法在水下环境中的应用。通过相位屏模型可以有效模拟水下湍流对光传播的影响，为研究水下湍流条件下的成像技术提供仿真基础。关联成像技术凭借其无需直接探测目标、对弱光灵敏以及抗干扰能力强的特点，展现出在湍流环境下的应用潜力。本研究将湍流相位屏模型与三种基于哈达玛矩阵的关联成像方法(传统方法、小波变换方法、切蛋糕方法)相结合，分别对比三种方法在不同湍流强度下的成像性能。随后，为进一步提升成像质量，将经过湍流关联成像后的模糊图像输入条件生成对抗网络中，进行去噪与复原处理，探索人工智能技术在水下湍流成像优化中的可行性与优势。实验结果表明，结合深度学习的关联成像方法可显著提高湍流环境下的成像质量，为水下光学成像的实际应用提供了一种高效的解决方案。
Underwater turbulence affects optical imaging primarily through speckle, blurring, and distortion caused by phase disturbances in light waves during propagation, which severely limits the application of traditional imaging methods in underwater environments. The phase screen model can effectively simulate the impact of underwater turbulence on light propagation, providing a simulation foundation for the study of imaging technology under turbulence conditions. Correlation imaging technology, with its ability to detect targets indirectly, high sensitivity to weak light, and strong anti-interference capabilities, shows extensive potential in turbulent environments. This study combines the turbulence phase screen model with three correlation imaging methods based on Hadamard matrices (traditional method, wavelet transform method, and cake-cutting method), systematically comparing their imaging performance under different turbulence intensities. Subsequently, to further improve image quality, the blurred images obtained through turbulence correlation imaging are input into a conditional generative adversarial network for denoising and restoration, exploring the feasibility and advantages of artificial intelligence in optimizing underwater turbulence imaging. The experimental results demonstrate that the combination of deep learning and correlation imaging significantly improves imaging quality in turbulent environments, providing an efficient solution for practical underwater optical imaging applications.