Institute of Optics and Electronics

Abstract Programmable optical torque is essential for advanced optical manipulation, yet existing approaches rely on feedback-driven, time-varying optical fields, introducing complexity and limiting scalability. Here, we introduce a new concept of freeform optical torques enabled by metasurfaces, and experimentally demonstrate their feedback-free, programmable control with flexibly shaped rotational-angle dependence, overcoming the long-standing limitation of torque functions confined to constant or sinusoidal forms. This capability arises from converting the temporal degree of control into a spatial one, realized through the interaction between temporally static vector optical fields and spin-orbit-engineered metasurfaces. To characterize these optical torque functions, we developed a torsion-balance system operating near the thermal noise limit. The system achieves high sensitivity of 0.22 fNm·Hz^-0.5 for macroscopic samples and enables direct measurement of fNm-level optical angle-torque functions that were previously unresolvable. This strategy provides unprecedented freedom in optical torque control, enabling passive, all-optical attitude and rotational motion control with broad implications from microscale robotics to macroscopic lightsails.

ABSTRACT Revolutionary optical imaging enabled by vectorial and flat optics marks a transformative paradigm that unites subwavelength photonic engineering, full‐vector light‐field control (amplitude, phase, and polarization), and planar optical architectures. Conventional imaging systems, constrained by curved‐optics geometry and scalar diffraction theory, face intrinsic trade‐offs among spatial resolution, field of view (FOV), and system compactness. By harnessing flat optical elements such as metasurfaces and catenary‐phase devices together with vectorial manipulation, this emerging field transcends those limitations. The review highlights three fundamental breakthroughs: (1) Abrupt phase shifts in subwavelength scale enabling ultrathin, wide‐FOV planar imaging; (2) Evanescent–propagating wave conversion under appropriate near‐field coupling conditions, and vectorial optical synthetic apertures achieving far‐field super‐resolution beyond the diffraction limit; (3) Computational optics that integrates sensing and computing in a unified physical platform. These advances enable intelligent, programmable imaging systems that transcend the limitations of traditional optics, with applications spanning biomedical nanoscopy, remote sensing, and quantum information.

Multi-plane phase retrieval reconstructs complex wavefronts from axially scanned diffraction patterns. However, laser instability or calibration errors can cause the experimental wavelength to deviate from the nominal value, resulting in cumulative phase errors and resolution loss. Here, we propose an adaptive wavelength calibration (AWC) algorithm to compensate for wavelength deviations by adjusting axial propagation distances during iteration. Utilizing the wavelength-distance equivalence in Fresnel diffraction, the algorithm dynamically adjusts axial spacing by minimizing structural similarity differences between simulated and actual diffraction patterns. The enhanced particle swarm optimization automatically determines the optimal axial correction for each plane, achieving sub-nanometer wavelength accuracy even under large deviations. Experiments on resolution targets, phase objects, and biological samples confirm its robustness against coupled errors from wavelength drift and stage misalignment. This approach relaxes strict wavelength stability requirements, enabling high-fidelity phase imaging with low-cost laser sources.

To endow a star sensor with strong robustness, low algorithm complexity, and a small database, this paper proposes an all-sky star identification algorithm based on spherical polar pattern matching. The proposed algorithm consists of three main steps. First, the guide star is rotated to be a polar star, and the polar and azimuth angles of neighboring stars are used as polar pattern elements of the guide star. Then, the relative azimuth histogram is applied to the spherical polar pattern matching, and a star pair after spherical polar pattern matching is identified through angular distance cross-verification. Finally, a reference star image is generated from the identified star pair to complete the matching process of all guide stars in the field of view. The proposed algorithm is verified by simulation experiments. The simulation results show that for a star sensor with a medium field of view (15° × 15°, 1024 × 1024 pixel) and a limiting magnitude of 6.0 Mv, the required database size is 161 KB. When false and missing star spots account for 50% of the guide stars and the star spot extraction error is 1.0 pixel, the average star identification time is 0.35 ms (@i7-4790), and the identification probability is 99.9%. However, when false and missing star spots account for 100% of the guide stars and the star spot extraction error is 5.0 pixel, the average star identification time is less than 2.0 ms, and the identification probability is 97.1%.

This paper proposes a high-type control loop design method for LQR-LMI based on Lyapunov and polyhedral model theory. The high-type control loop design problem is simplified into a convex constraint problem, which achieves superior tracking performance. In this framework, the input amplitude of the control signal, the poles of the closed-loop system, the suppression of external interference and the perturbation of internal parameters are considered, and the linear matrix inequality (LMI) method is effectively used to solve the problems. In this paper, the polyhedral model control theory is introduced to characterize the uncertainty of the system for the change of model parameters of the controlled plant. Aiming at the problem of external disturbance suppression, the H2/H∞ control method is introduced into the system. These control methods provide the basis for the design of the high-type control loop. Compared with the simulation results of other optimization algorithms, the effectiveness and superiority of the controller parameter tuning rules in the proposed high-type control loop are verified.

Real-time object detection is critical for unmanned aerial vehicles (UAVs) performing various tasks. However, efficiently deploying detection models on UAV platforms with limited storage and computational resources remains a significant challenge. To address this issue, we propose ELNet, an efficient and lightweight object detection model based on YOLOv12n. First, based on an analysis of UAV image characteristics, we strategically remove two A2C2f modules from YOLOv12n and adjust the size and number of detection heads. Second, we propose a novel lightweight detection head, EPGHead, to alleviate the computational burden introduced by adding the large-scale detection head. In addition, since YOLOv12n employs standard convolution for downsampling, which is inefficient for extracting UAV image features, we design a novel downsampling module, EDown, to further reduce model size and enable more efficient feature extraction. Finally, to improve detection in UAV imagery with dense, small, and scale-varying objects, we propose DIMB-C3k2, an enhanced module built upon C3k2, which boosts feature extraction under complex conditions. Compared with YOLOv12n, ELNet achieves an 88.5% reduction in parameter count and a 52.3% decrease in FLOPs, while increasing mAP50 by 1.2% on the VisDrone dataset and 0.8% on the HIT-UAV dataset, reaching 94.7% mAP50 on HIT-UAV. Furthermore, the model achieves a frame rate of 682 FPS, highlighting its superior computational efficiency without sacrificing detection accuracy.

Abstract Polarimetric imaging has emerged as a powerful technique for capturing critical surface morphology and material anisotropy information, offering unique advantages including biomedical diagnostics and industrial inspection. However, conventional division-of-focal-plane polarization cameras suffer from low optical efficiency due to filtering losses, and reduced spatial resolution caused by beam splitting redundancy and limited pixel miniaturization. Here, we propose a polarization router based on a field-driven inverse-designed metasurface with non-orthogonal polarization manipulation capability, which can focus four different elliptical polarization states onto designated detector pixels. Leveraging global inverse design, this design achieves high-efficiency, high-resolution and high-precision full-Stokes polarization detection at a wavelength of 1064 nm. Two super-pixel sizes, 10 × 10 μm and 4.8 × 4.8 μm (2 × 2 pixel layout), were designed, achieving average optical efficiencies of 78.18% and 80.32%, with polarization reconstruction relative errors of 0.7% and 1.27%, respectively. The comparative analysis of size-dependent performance reveals that despite a significant reduction in dimensions, the optimized metasurface maintains stable performance. This advance paves the way for compact, high-precision real-time polarization imaging systems.

Abstract Symmetry-protected bound states in the continuum (SP-BICs) have attracted considerable attention in the realm of nanophotonics, primarily due to their infinite quality (Q) factor and remarkable ability to confine fields. However, while most of the current research realizes multiple high-Q quasi-SP-BIC resonances, these lack independent tunability, thereby significantly limiting their utility in multi-wavelength lasers with independent control. In this work, we introduce symmetry-broken in an all-dielectric metasurface to achieve dual quasi-SP-BICs with independent tunability. An analysis conducted through near-field distributions and multipole decomposition reveals that the excited dual quasi-SP-BICs are predominantly influenced by the magnetic dipole. Furthermore, the dual quasi-SP-BICs can be independently controlled by varying the magnitude of the periods in both the x- and y-directions, respectively. Our research lays the groundwork for the development of independently adjustable multiple SP-BIC metasurfaces, thereby enabling their application in a variety of fields, such as lasers, nonlinear optics, and biosensing.

Flow characteristic monitoring includes parameters such as flow regime, fluid characteristic frequency, and flow rate, which are crucial for optimizing production and ensuring the safety of oil and gas transportation systems. Existing fluid monitoring technologies, such as various flow meters, often face limitations in providing distributed and real-time monitoring data. In contrast, distributed acoustic sensing offers a spatial resolution of 1 m with high frequency sampling capability, allowing for long-term, multi-point dynamic monitoring of fluid migration characteristics. We developed an indoor physical simulation pipeline loop to assess the feasibility of using distributed acoustic sensing for monitoring flow migration characteristics. The experiment collected signal characteristics under different conditions, including background noise, single gas-phase flow, single liquid-phase flow, and gas-liquid two-phase flow. In the frequency-power spectral density analysis, single gas-phase flow signals are concentrated at lower frequencies, single liquid-phase flow displays noticeable spikes over a broader frequency range, and gas-liquid two-phase flow covers the widest frequency range with stronger amplitude signals. Autocorrelation analysis shows larger oscillations for gas-liquid two-phase flow, smoother signals for gas-phase flow, and more turbulent signals for liquid-phase flow. By examining root mean square energy changes, flow rates can be qualitatively estimated, revealing a positive correlation between energy and flow velocity. Finally, the study discussed the limitations of the experimental setup and proposed improvements and advanced research directions of distributed acoustic sensing in fluid monitoring.

Enhancing the detection capabilities of rotated objects in aerial imagery is a vital aspect of the burgeoning field of remote sensing technology. The objective is to identify and localize objects oriented in arbitrary directions within the image. In recent years, the capacity for rotated object detection has seen continuous improvement. However, existing methods largely employ traditional backbone networks, where static convolutions excel at extracting features from objects oriented at a specific angle. In contrast, most objects in aerial imagery are oriented in various directions. This poses a challenge for backbone networks to extract high-quality features from objects of different orientations. In response to the challenge above, we propose the Dynamic Rotational Convolution (DRC) module. By integrating it into the ResNet backbone network, we form the backbone network presented in this paper, DRC-ResNet. Within the proposed DRC module, rotation parameters are predicted by the Adaptive Routing Unit (ARU), employing a data-driven approach to adaptively rotate convolutional kernels to extract features from objects oriented in various directions within different images. Building upon this foundation, we introduce a conditional computation mechanism that enables convolutional kernels to more flexibly and efficiently adapt to the dramatic angular changes of objects within images. To better integrate key information within images after obtaining features rich in angular details, we propose the Multi-Order Spatial-Channel Aggregation Block (MOSCAB) module, which is aimed at enhancing the integration capacity of key information in images through selective focusing and global information aggregation. Meanwhile, considering the significant semantic gap between features at different levels during the feature pyramid fusion process, we propose a new multi-scale fusion network named AugFPN+. This network reduces the semantic gap between different levels before feature fusion, achieves more effective feature integration, and minimizes the spatial information loss of small objects to the greatest extent possible. Experiments conducted on popular benchmark datasets DOTA-V1.0 and HRSC2016 demonstrate that our proposed model has achieved mAP scores of 77.56% and 90.4%, respectively, significantly outperforming current rotated detection models.