Visual tracking algorithm based on a Siamese network is an important method in the field of visual tracking in recent years, and it has good performance in tracking speed and accuracy. However, most tracking algorithms based on the Siamese network rely on an off-line training model and lack of online update to tracker. In order to solve this problem, we propose an online learning-based visual tracking algorithm for Siamese networks. The algorithm adopts the idea of double template, treats the target in the first frame as a static template, and uses the high confidence update strategy to obtain the dynamic template in the subsequent frame; in online tracking, the fast transform learning model is used to learn the apparent changes of the target from the double template, and the target likelihood probability map of the search area is calculated according to the color histogram characteristics of the current frame, and the background suppression learning is carried out. Finally, the response map obtained by the dual templates is weighted, and the final prediction result is obtained. The experimental results on OTB2015, TempleColor128, and VOT datasets show that the test results of this algorithm are improved compared with the mainstream algorithms in recent years and have better tracking performance in target deformation, similar background interference, fast motion, and other scenarios.

Stereoscopic image zoom optimization is a popular basic research problem in the field of image processing and computer vision in recent years. The zoom visual enhancement technology of 3D images has attracted more and more attention. To this end, this paper proposes a method of stereoscopic zoom vision optimization based on grid deformation from the model of camera zoom shooting, and strives to improve the experience of 3D stereoscopic vision. Firstly, use the digital zoom method to simulate the camera model to properly zoom in on the target area, and then establish the mapping relationship between the reference image and the target image according to the camera zoom distance. Secondly, extract the foreground target object and use the modified just noticeable depth difference (JNDiD) model to guide the adaptive depth adjustment of the target object. Finally, combined with the seven grid-optimized energy terms designed in this paper, the image grid is optimized to improve the visual perception of the target object and ensure a good visual experience for the entire stereoscopic image. Compared with the existing digital zoom method, the proposed method has better effects on the size control of the image target object and the depth adjustment of the target object.

Aiming at the working characteristics of the end-effector of the gantry robot with only three translational degrees of freedom, a method for calibrating the point cloud coordinate system of the 3D vision sensor and the tool coordinate system of the robot actuator is designed, on the basis of the traditional two-step method of hand-eye calibration. In this method, only three calibration target pictures and three sets of point clouds are collected by two orthogonal translation movements of the robot, the rotation matrix and translation vector of the hand-eye relationship can be calibrated by measuring the base coordinates of mark points on the target through the TCP contact of the actuator. The method is simple to operate, and the calibration target is easy to make with low cost. The XINJE gantry robot and 3D vision sensor of structured light was used to build an experimental platform for experiments. The results show that the method has good stability and is suitable for field calibration, with calibration accuracy within ?0.2 mm.

The star/airborne optical remote sensing image has a wide field of view and a complex scene. It is easy to produce a large number of false alarms that are similar to the ship's target due to the impact of the shore construction and broken cloud, causing great interference to the ship's detection. Traditional marine ship detection algorithms are difficult to be effective extracting discriminative features that are conducive to detection, results in low detection rates and high false alarm rates for ships. In view of this, this paper proposes an optical ship target detection method combining hierarchical search and visual residual network from the perspective of low false alarm and low missed detection. Firstly, the land and sea area are segmented based on the texture integral map; secondly, the target candidate area is extracted by combining the multi-scale local structural features; then, the primary false alarm is removed by the layered removal strategy based on multi-dimensional visual features; finally, the visual residuals are built the network finely removes false alarms from suspected candidate areas to obtain the final detection result. Based on the GF2 remote sensing GF2 set, the algorithm proposed in this paper is tested and verified. The comprehensive detection rate of this algorithm is 92.0%, the false alarm rate is 12.58%, the average processing time is 0.5 s, the detection effect is good, the efficiency is high, and the adaptability to various scenes is good. It can achieve accurate and efficient detection and positioning of optical ships in complex environments.

In the phase measuring profilometry, the phase measuring accuracy could be heavily affected by the nonlinearity effects of the projecting and imaging devices. Therefore, it is very important to reduce the nonlinear errors fast and efficiently. An analytic model of nonlinear errors is introduced. Then we propose a phase compensation method which is based on the accurate mathematical model of the phase error. The proportion of each harmonic component is collected by using a large-step phase-shifting algorithm to measure a reference plane. Then the phase errors of the measured object could be compensated by an iterative algorithm. The experimental results show that the proposed method can realize nonlinear error compensation effectively and improve the precision of phase measurement. Meanwhile, since all the harmonic components are pre-calibrated, there is no extra fringe needed, which can meet the requirements of fast and real-time measurement.

The mode field diameter is an important parameter of single-mode fiber, and the GB.15972.45-2008 recommends using the far-field variable aperture method to measure it. This paper analyzes the distribution of the propagating light field in a single-mode fiber. The mode behavior of the light field is the solution of the Helmholtz equation, which in theory should satisfy the Bessel function. In this regard, a method using Bessel function to fit the optical field distribution of the fiber based on the far-field variable aperture method is proposed, and the mode field diameter is calculated from the fitted mode field distribution curve. Compared with the commonly used far-field variable aperture method, when the measurement data is normal, this method has the same measurement accuracy. When there are errors in the measurement data, this method can still ensure the stability and accuracy of the measurement results.

This paper proposes an efficiency-tunable terahertz focusing lens based on the graphene metasurface. The unit cell is composed of two symmetrical circular graphene hollows and an intermediate dielectric layer, wherein the hollow circular middle is connected by a rectangular graphene sheet. This structure can realize polarization conversion, for example, when an incidence with left-hand circular polarization emitted on the metasurface the polarization of the transmitted light is right-hand circular polarization. According to the principle of geometric phase, by rotating the direction of the rectangular bar, the transmitted wave will carry an additional phase and can cover the range of 2π. An THz focusing lens can be realized by properly arranging the unit structure of the graphene metasurface. The simulation results show that the conversion amplitude of circular polarized light can be adjusted by changing the Fermi level of graphene, and the focusing efficiency of the metalens can also be dynamically adjusted. Therefore, this graphene metasurface-based efficiency-tunable focusing lens can be realized by changing the Fermi level without changing the size of the unit cell, and can be widely used in terahertz applications such as energy harvesting and imaging.

Real-time detection of small objects is always a difficult problem in image processing. Based on the target detection algorithm of deep learning, this paper proposed an end-to-end neural network for mobile phone small target detection in complex driving scenarios. Firstly, an end-to-end small target detection network (OMPDNet) was designed to extract image features by improving the YOLOv4 algorithm. Secondly, based on the K-means algorithm, a K-means-Precise clustering algorithm of more appropriate data samples distribution in the clustering center was designed, which was used to generate prior frames suitable for small target data, so as to improve the efficiency of the network model. Finally, we constructed our own data set with supervision and weak supervision, and added negative samples to the data set for training. In the complex driving scene experiments, the OMPDNet algorithm proposed in this paper can not only effectively complete the detection task of using mobile phone while driving, but also has certain advantages over the current popular algorithms in accuracy and real-time for small target detection.

Polyimide (PI) film is widely used in aerospace, microelectronics, and other fields because of its excellent thermal stability and mechanical strength. However, there are very few reports about its application in the direction of optical imaging. To use PI film for imaging, the requirements for the optical homogeneity of the PI film are extremely demanding. The optical homogeneity of the stretch-resistant PI film proposed in this paper with 100 mm diameter and low thermal expansion coefficient meets the Rayleigh criterion, which has the potential for applications in the imaging field. In addition, the tensile strength of this PI is 285 MPa, which is ~2.6 times that of the PMDA-ODA type PI; the coefficient of thermal expansion is about 3.2 ppm?K-1, which is comparable to that of the Novastrat?905 type PI and is one order of magnitude lower than that of the commercial PI films. These excellent basic properties reserve more space to further improve the space adaptability of the PI film. The solution of the optical homogeneity of the PI film will lay the foundation for its application in thin film diffractive optical elements.

Aiming at the technical difficulties in the rapid detection and reconstruction of three-dimensional micro-nano devices that are difficult to achieve both high precision and high speed, this paper proposes a structured light detection method based on time-domain phase shift technology. The measured light is modulated by a spatial light modulator, and the time-domain phase shift technology is further employed to realize the detection and reconstruction of three-dimensional micro-nano devices. Compared with the traditional structured light detection method, this technology uses the spatial light modulator to measure the phase shift while the sample is scanned axially, so as to ensure the measurement accuracy and improve the measurement efficiency. By analyzing the measurement data, this method can quickly realize three-dimensional shape detection and reconstruction, and the measurement accuracy can be better than 10 nm.