Single-pixel imaging can achieve high-dimensional imaging with only a single-pixel detector, in which the high-dimensional information is encoded into a one-dimensional detection sequence through time-varying structured patterns. Due to its extremely low requirement for detector’s resolution, single-pixel imaging has great research value in special waveband imaging, extremely low light imaging, and lidar detection. Compared with traditional intensity imaging, complex optical field imaging can not only obtain the change of light field intensity, but also obtain the phase information, therefore, it performs a great value in label-free microscopy imaging, surface measurement, crystal analysis and so on. Combining the advantages of both single-pixel imaging and complex optical imaging, single-pixel complex optical imaging has drawn widely attentions and can be applied in low-light microscopy imaging and measurement. Starting from the basic principles of single-pixel imaging and complex light field detection, the research status and some recent advances of single-pixel optical field imaging are reviewed, and the challenging problems as well as future research directions are also discussed.

Ghost imaging lidar is a computational imaging technology that obtains the target’s information by actively modulating the fluctuation of light field and receiving the target’s echo signal with a single pixel detector in the condition of staring detection. It has important application prospects in the fields of long-distance target recognition, three-dimensional imaging and the defense of important place, and has become a research hotspot in recent years. The basic principles and characteristics of the two kinds of ghost imaging lidar based on narrow pulse direct detection system and long pulse heterodyne detection system respectively are briefly reviewed, and their recent main research progress are introduced. Furthermore, with respect to the application of long-distance, high-resolution and high-speed moving target detection and recognition, the key problems and development trend of ghost imaging lidar are discussed and prospected.

Diffraction imaging via intensity correlation is a novel diffraction imaging method based on the high-order correlation characteristics of optical field. In this method, the Fourier diffraction spectrum information of a sample can be obtained nonlocally by calculating the correlation between the intensity fluctuation of the reference beam and the test beam. As a new imaging method, diffraction imaging via intensity correlation has the advantages of nonlocal imaging, low radiation imaging, high resolution and so on, thus it can solve problems that are difficult to handle by conventional imaging. This technology has attracted much attention recently, and has great potential in biomedicine, material science and many other fields in the future. In this paper, the latest progress of diffraction imaging via intensity correlation is briefly reviewed, and some related image reconstruction algorithms are introduced. Finally, the existing problems of this method are pointed out and its future development is prospected.

Optical imaging is one of the most important techniques for information acquisition. As a novel technique based on high-order correlation of light field, ghost imaging can reconstruct the two-dimension image of objects by using only a single pixel detector. It has the advantages of object-image separation, high imaging sensitivity and strong robustness, which promises new opportunities and prospects for the development of optical imaging. However, the imaging mode based on multi-frame cumulation takes toomuch time to acquire images, which severely limits the efficiency of image acquisition. In order to solve this problem, research on efficient reconstruction algorithm is one of the most practical methods to improve the imaging quality and imaging efficiency besides the optimization designs of system structure, light source and detection. A good reconstruction algorithm can not only greatly reduce the number of measurements required for imaging, improve the efficiency of information extraction and quality of reconstructed image, but also reduce the requirements of hardware. In recent years, many different kinds of reconstruction algorithms have been developed. In this paper, the principle and mechanism of ghost imaging are briefly reviewed, and then the basic principles of several typical algorithms of ghost imaging are introduced systematically, as well as their advantages, disadvantages and applicable scenarios.

Scattering media will change the propagation direction and path of photons, resulting in the degradation of image quality and even the formation of speckle. Theoretically, information of targets can be recovered by using the transmission matrix of the scattering medium, but the process to solve the transmission matrix is too complicated. The rapid development of deep learning provides a new method to solve the problem of imaging through scattering media. As a typical method for solving inverse problems, deep learning can recover the target information accurately, improve the imaging quality and so on, and it has achieved many important research results in the field of scattering imaging. The existing deep learning methods could be divided into two categories: supervised learning and unsupervised learning. Herein, we summarize the research progress of imaging through scattering media based on deep learning from these two aspects, and compare the performance of some intelligentalgorithmic imaging techniques in terms of network structure, imaging quality and generalization of deep learning. Finally, the advantages and challenges of deep learning-based imaging technology are analyzed, and the future development of this field is prospected.

In recent years, the rapid development of image acquisition equipment has greatly enriched the types and quantities of images. As the key of image analysis and processing, image registration technology has become increasingly important in the fields of image fusion, pattern recognition and computer vision, and how to register images with high accuracy and in real time has become the focus of research in related fields. At the same time, with the rapid development of deep learning techniques, convolutional neural networks show unique advantages in image representation and feature extraction. The aim of this work is to provide a systematic review of research on image registration using deep learning techniques. By discussing typical deep learning-based image registration methods from deep iterative registration, fully supervised image registration, weak/dually supervised image registration, and unsupervised image registration, we highlight common challenges faced by related researchers, and explore possible future research directions to address these challenges.

In the optical synthetic aperture imaging system, multiple small aperture telescopes are arranged into a sparse aperture array to increase the equivalent aperture of the system, so as to achieve the high-resolution imaging effect of the large aperture optical system. The detection of co-phasing error between subapertures is an important prerequisite for realizing high-resolution imaging of synthetic aperture systems, and this technology has always been one of the focuses of researchers in this field. The emerging artificial intelligence and big data technology provide a new idea and open up a new direction for the detection of co-phasing error of synthetic aperture imaging system. On the basis of a brief review of the co-phasing error detection methods of synthetic aperture imaging system, the research progress of deep learning technology in co-phasing error detection of synthetic aperture imaging system in recent years is introduced and analyzed, and the future development direction is finally summarized and prospected.

Nowadays, optoelectronic devices are developing towards miniaturization and integration. However, traditional optoelectronic devices are usually based on silicon wafer technology or waveguide technology, which makes the chip need to be coupled with optical fiber pigtail through waveguide mode converter. Therefore, it is of great practical significance to develop optoelectronic devices compatible with optical fiber system. The development of “optical fiber laboratory” technology promotes the combination of low-dimensional materials and optical fibers, as well as the integration and development of photonic chips on optical fibers, and is helpful for developing a new generation of miniaturized, integrated, lightweight, low-cost, and multi-functional all fiber photonic integration platform. According to the different interaction modes of light and matter, optical fiber integrated photodetectors can be divided into two types: along waveguide integration and optical fiber end integration. This paper mainly reviews the fabrication methods and research progress of these two types of optical fiber integrated photodetectors in recent years, and prospects the future development of optical fiber as an integrated platform for optoelectronic devices.

The orbital angular momentum (OAM) of vortex beams can be used for information coding, which has important applications in free space optical communications. Nevertheless, the propagation of beams may encounter various random media in actual transmission space, which will distort the wavefront of vortex beam, rendering the ineffectiveness of traditional methods in retrieving the OAM. To tackle this issue, by taking ground glass as a random medium, recognition of OAM of vortex beam from speckle pattern is realized based on deep learning. Moreover, to further upgrade the information coding and transmission capability, recognition of multiple OAMs is examined as well. The results show that the designed network is capable of simultaneously recognizing five OAMs from a single-shot speckle pattern generated by vortex beams through ground glass.

To speed up single-pixel imaging at a low cost, solutions are proposed from both hardware and algorithm aspects, so that a single-pixel imaging system with software and hardware cooperation is built. On algorithm, DQN-Hadamard single-pixel imaging is proposed by introducing DQN algorithm into Hadamard single-pixel imaging, to obtain the relatively optimal sampling sequence for specific target objects. On hardware, based on the high-speed and low-cost pattern projection scheme using LED array, ZYNQ 7020, which combines the characteristics of ARM and FPGA, is selected as the controller to build a Hadamard single-pixel imaging system with the cooperation of ARM and FPGA. Finally, combining the algorithm and hardware, the transplantation of DQN-Hadamard single-pixel imaging to the built Hadamard single-pixel imaging system is completed, and then DQN-Hadamard single-pixel imaging system is compared with the single-pixel imaging system using adaptive zig-zag sampling in detail. The imaging results show that as for the proposed imaging system, the target object can be reconstructed at the sampling rate of 30%, so that the feasibility of low-cost single-pixel imaging with software and hardware cooperation is verified.

A high spatial resolution ghost imaging technology based on cosine encoded multiplexing is demonstrated. Firstly, many low spatial resolution cosine encoded patterns are constructed and multiplexed to produce high spatial resolution modulation patterns, and then the high spatial resolution modulation patterns is used to illuminate the object. A single-pixel detector collects the backscattered light generated by the interaction between the modulated light and the object, and the mixed image of the object can be recovered by applying an iterative algorithm. Given the certain deterministic spectrum structure of the employed cosine encoded matrices, the low spatial resolution images of the object can be decoded and reconstructed efficiently using the digital image processing method, and finally spliced into high spatial resolution object images. The implementation method of the proposed cosine encoded multiplexing high spatial resolution ghost imaging is analyzed theoretically, and the effectiveness of the proposed method is verified by numerical simulation. It is shown that the proposed method greatly reduces the modulation patterns required for traditional high spatial resolution ghost imaging, reduces the online sampling time and the offline reconstruction time, and improves the imaging efficiency of high spatial resolution ghost imaging.