Shansi Zhang

Bio: Shansi Zhang is an academic researcher from Beijing Institute of Technology. The author has contributed to research in topics: Computer science & Engineering. The author has an hindex of 1, co-authored 1 publications receiving 3 citations.

LRT: An Efficient Low-Light Restoration Transformer for Dark Light Field Images

Abstract: Light field (LF) images containing information for multiple views have numerous applications, which can be severely affected by low-light imaging. Recent learning-based methods for low-light enhancement have some disadvantages, such as a lack of noise suppression, complex training process and poor performance in extremely low-light conditions. To tackle these deficiencies while fully utilizing the multi-view information, we propose an efficient Low-light Restoration Transformer (LRT) for LF images, with multiple heads to perform intermediate tasks within a single network, including denoising, luminance adjustment, refinement and detail enhancement, achieving progressive restoration from small scale to full scale. Moreover, we design an angular transformer block with an efficient view-token scheme to model the global angular dependencies, and a multi-scale spatial transformer block to encode the multi-scale local and global information within each view. To address the issue of insufficient training data, we formulate a synthesis pipeline by simulating the major noise sources with the estimated noise parameters of LF camera. Experimental results demonstrate that our method achieves the state-of-the-art performance on low-light LF restoration with high efficiency.

A Deep Retinex Framework for Light Field Restoration under Low-light Conditions

Abstract: Light field (LF) images can record the scene from multiple directions and have many applications, such as refocusing and depth estimation. However, these applications can be heavily influenced by poor light condition and noise. This work aims to recover the high-quality LF images from their lowlight detection. First, a decomposition network is employed to decompose each LF image into its reflectance and illumination with the Retinex theory. Then, two enhancement networks are designed to denoise the reflectance and enhance the illumination, respectively. They adopt alternate spatial-angular feature extractions and process all the views synchronously with high efficiency. A parallel dual attention mechanism is integrated to both the spatial and angular feature extractions, to encode more important information. Moreover, a discriminator is introduced during the training to generate more realistic LF images by making judgment according to both the spatial and angular characteristics. Experimental results have demonstrated the superior performance of our method, which can restore the content, luminance, color and geometric structures of LF images effectively.

A Spectral Analysis of Light Field Signal for Plenoptic Sampling and Rendering

Abstract: Light field rendering (LFR) has been widely used for generating multiview images in Image-Based Rendering. However, to ensure the quality of novel views, this conventional rendering technology, requires a mass of manually captured light field images as input, which contains complete light field signal information and is highly time-consuming. Depth information of 3D scene and light field spectrum analysis are fundamental to light field reconstruction. The depth information was acquired with spectrum statistical analysis. This study aims to provide a broad applicability and robust depth layered scene mapping method by extending the traditional plenoptic sampling theory to reasonably determine the number of necessary captured images. A mathematical function was derived on the basis of overlapped spectrum, to decide the optimal mapping layer. Then, an optimal reconstruction filter is designed to reconstruct novel views. The proposed method exhibited a larger PSNR value in a variety of scenes. The depth layered scene mapping method showed potential for light field rendering without the need for complete multiview images requiring continuous sampling.

Denoising for Photon-limited Imaging via a Multi-level Pyramid Network

Abstract: Imaging under photon-scarce situations introduces challenges to many applications as the captured images are with low signal-to-noise ratio. Here, we target on the denoising under photon-limited imaging. We develop a multi-level pyramid denoising network (MPDNet), which employs the idea of Laplacian pyramid to learn the small-scale noise map and larger-scale high-frequency details at different levels. Feature extractions are conducted on the multi-scale input images to encode richer contextual information. The major component of MPDNet is the multi-skip attention residual block, which integrates multi-scale feature fusion and attention mechanism for better feature representation. Experimental results have demonstrated that our MPDNet can achieve superior denoising performance on the images with low photon counts.

Disturbance Observer Based Robust Sliding Mode Control of Permanent Magnet Synchronous Motor

Abstract: This paper addresses speed control of permanent magnet synchronous motor under load torque perturbations. The mathematical model is derived using park’s transformation. The load torque disturbance is considered unknown bounded, and states variables are available in feedback. In order to achieve robust speed performance, sliding mode control (SMC) is introduced. However, it is noted that conventional SMC does not provide satisfactory performance under load torque disturbances. To end this, a novel control strategy called disturbance observer SMC (DOSMC) is formulated. It includes an observer that offers a tool to vanish the effect of load torque. The DOSMC technique has two distinguished characteristics: first, the design gains are needed to be greater than the maximum limit of disturbance estimation error instead of disturbances, second; the proposed observer estimates load disturbances and provide a compensator to update sliding surface and control input. The stability analysis of overall control system is verified using Lyapunov theorem. Simulations in MATLAB/Simulink proves efficacy of the proposed scheme.

LRT: An Efficient Low-Light Restoration Transformer for Dark Light Field Images

Abstract: Light field (LF) images containing information for multiple views have numerous applications, which can be severely affected by low-light imaging. Recent learning-based methods for low-light enhancement have some disadvantages, such as a lack of noise suppression, complex training process and poor performance in extremely low-light conditions. To tackle these deficiencies while fully utilizing the multi-view information, we propose an efficient Low-light Restoration Transformer (LRT) for LF images, with multiple heads to perform intermediate tasks within a single network, including denoising, luminance adjustment, refinement and detail enhancement, achieving progressive restoration from small scale to full scale. Moreover, we design an angular transformer block with an efficient view-token scheme to model the global angular dependencies, and a multi-scale spatial transformer block to encode the multi-scale local and global information within each view. To address the issue of insufficient training data, we formulate a synthesis pipeline by simulating the major noise sources with the estimated noise parameters of LF camera. Experimental results demonstrate that our method achieves the state-of-the-art performance on low-light LF restoration with high efficiency.

Control of magnetic levitation positioning stage with an eddy current damper

Abstract: To enhance the system damping,a permanent magnet set which served as an eddy current damper was added to the magnetic levitation positioning stage which consists of a moving table,four Halbach permanent magnetic arrays,four stators and displacement sensors.The dynamics model of this stage was a complex nonlinear,strong coupling system which made the control strategy to be a focus research.The nonlinear controller of the system was proposed based on the theory of differential geometry.Both simulation and experimental results show that either the decoupling control of the movement can be realized in horizontal and vertical directions,and the control performance was improved by the damper,verifying the validity and efficiency of this method.

Dynamic simulation of a rigid-flexible coupling positioning stage base on modified floating frame of reference formulation

Abstract: A rigid-flexible coupling positioning stage (RFCPS) is proposed in this study for long-stoke-high-precision (LSHP) positioning, and its dynamic model is established. The structure of the RFCPS contains three parts: the working stage, the frame and the flexure hinge group. The flexure hinges can provide micro deformation to compensate the positioning error cause by the friction of the mechanical bearing. In order to consider the deformation of the flexure hinges, the flexible multi-body system dynamic analysis method is adopted to establish the dynamic model of the RFCPS. The investigation of the principle model demonstrates that the floating frame of reference formulation (FFRF) has higher accuracy than the absolute nodal coordinate formulation (ANCF) for the deformation analysis of the RFCPS. Sequentially, a finite element based model (FE model) is established by using the FFRF. Numerical simulation results shown that the results of the FE model is consistent to the analytical solution, and the deformation of the flexure hinges is effectively obtained. The establishment of the dynamic model lays the foundation for structural optimization and control system design of the RFCPS.