Underwater imaging plays an increasingly important role in marine military, marine engineering, marine resource development,marine environmental protection, and so on, with the advantage of providing rich information, high resolution and high visibility underwater images. However, a large number of plankton and suspended particles in water environment, especially in the marine environment, causing strong scattering and absorption effects and resulting in image degradation problems such as blurring, short imaging distance, color distortion, low contrast, etc. Therefore, a series of underwater imaging methods have been proposed to solve the above problems.The underwater image enhancement technology can be used for image denoising, contrastimprovement and color distortioncorrection. The underwater image restoration uses the physical model of water degradation to restore the real image. The underwater polarization imaging uses the polarization difference between background and target to remove noise. The underwater ghost imaging and underwater compressed sensing imaging are used for imaging in scattering media. The underwater spectral imaging is used for color restoration. The underwater laser imaging is used for long-range and three-dimensional imaging. The underwater holographic imaging is used for water microorganism imaging, and so on. However, the above methods can only solve some image degradation problems, and there are some drawbacks, such as the subjectivity of underwater image enhancement technology, the dependence of underwater image recovery technology on prior information, and the computational load of underwater image correlation.The development of deep learning together with the development of hardware technology provides new solutions to the above problems, which makes the combination of deep learning and underwater imaging technology more and more widely used. As a powerful tool, neural network can extract similar features of different images using a wide range of datasets and convert them into high-level features, which can be used to process new input data, and completes a variety of complex tasks implicitly. It performs excellently in the field of image processing, and has made some achievements in the application of underwater imaging.Deep learning-basedimage restoration uses neural network to establish image-parameter mapping to estimate model parameters, avoiding human-dominant influence. Deep learning-based polarization imaging uses a neural network to map polarized images to clear images for image denoising. Deep learning-based spectral underwater imaging technology uses neural network to fuse multispectral images and hyperspectral images to obtain images with both high spatial resolution and hyperspectral resolution. However, some problems such as lack of datasets, poor generalization, and insufficient network interpretabilitystill exist, which need to be further solved.In this review, we discuss the characteristics of water environment and the various problems existing in underwater imaging, such as image blurring, short imaging distance, severe color distortion, and so on. The causes of the problem are analyzed and the underwater IFM model proposed by Jaffe-McGlamey is introduced. The latest application progress of various classic underwater imaging methods is systematically reviewed, including underwater image enhancement, underwater image restoration, underwater polarization imaging, underwater correlation imaging, underwater spectral imaging, underwater compression sensing imaging, underwater laser imaging and underwater holographic imaging. In addition, the basic concepts of deep learning, the composition of neural network and the structure of classical CNN network are introduced, and the latest application in combination with the above underwater imaging technology is systematically reviewed. At the same time, the application characteristics, deficiencies of traditional underwater imaging and the improvement by deep learning are analyzed and compared, and the applications of deep learning in various imaging methods are summarized. CNN network structure and MSE loss function are most commonly used due to its simplicity and efficiency. Finally, the future direction of underwater imaging technology based on deep learning is prospected.