Li Jingwei

Huawei Technologies · Department of R&D

The framework provides a novel pipeline of action recognition. The action recognition task classifies the action label of the scene. High-speed cameras are commonly used to generate high frame-rate videos for capturing sufficient motion information. However, the data volume would be the bottleneck of the system. With the insight that the discrete c...

Compressive imaging senses optically encoded high-dimensional scene data with far fewer measurements and then performs reconstruction via appropriate algorithms. In this paper, we present a novel noniterative end-to-end deep learning-based framework for compressive imaging, dubbed CoCoCs. In comparison to existing approaches, we extend the pipeline...

We propose a DCT-Cam to capture the video spectrum directly from the scene in a single shot, which can reduce 75% data requirement and achieve the same accuracy for the action recognition.

Optical diffractive deep neural network (OD²NN) is an optical machine learning framework that utilizes diffraction on the cascaded diffractive surfaces to perform an arbitrary function. Compared with the deep neural networks (DNNs) implemented in electronic domain, proof-of-principle demonstrations of OD²NNs show promising advantages in terms of sp...

In recent years, the explosive development of artificial intelligence implementing by artificial neural networks (ANNs) creates inconceivable demands for computing hardware. However, conventional computing hardware based on electronic transistor and von Neumann architecture cannot satisfy such an inconceivable demand due to the unsustainability of...

We propose a novel optical training framework for Optical Diffractive Deep Neu-ral Network based on Direct Feedback Alignment. The proposed method can accelerate the training process enormously without significant loss of accuracy.

Reservoir computing (RC) attracts much attention in many time-­series tasks due to its nonlinear mapping capability from the lower space to the higher space and the dynamic memory ability. With conceptual simplicity and computation cheapness, RC is also suitable for hardware implementation. In this paper, the RC is applied to perform the signal equ...

The combining of reflected endoscopic imaging with spectral data has recently attracted much attention. In this study, we used an optical fiber bundle probe, a galvo-scanning module and an imaging spectrometer module to realize a dual-channel endoscope that is capable of simultaneously acquiring real-time video data and high-throughput hyperspectra...

Environmental and internal factors can degrade the performance of a traditional hyperspectral imager. Here, we combine a self-developed hyperspectral imager with a calibration camera to improve its performance with our algorithm. The calibration camera is utilized to reduce the effect of the camera shake and provide high-definition splicing image i...

Light focusing inside live tissue by digital optical phase conjugation (DOPC) has drawn increasing interest due to its potential biomedical applications in optogenetics, microsurgery, phototherapy, and deep-tissue imaging. However, fast physiological motions in a live animal, including blood flow and respiratory motions, produce undesired photon pe...

Focusing light deep by engineering wavefronts toward guide stars inside scattering media has potential biomedical applications in imaging, manipulation, stimulation, and therapy. However, the lack of endogenous guide stars in biological tissue hinders its translations to in vivo applications. Here, we use a reversibly switchable bacterial phyto-chr...

Focusing light inside scattering media by optical phase conjugation has been intensively investigated due to its potential applications, such as in deep tissue imaging. However, no existing physical models explain the impact of the various factors on the focusing performance inside a dynamic scattering medium. Here, we establish an angular-spectrum...

The fast response and analysis of oil spill accidents is important but remains challenging. Here, a compact fluorescence hyperspectral system based on a grating-prism structure able to perform component analysis of oil as well as make a quantitative estimation of oil film thickness is developed. The spectrometer spectral range is 366-814 nm with a...

We present a fast hyperspectral line-scan fluorescence imaging system to verify the feasibility of quantitative fluorescence authentication of powders. Vanillin, which is restricted for use in milk powder, especially in milk powder for infants, is mixed with milk powder in different mass concentrations (5%, 10%, 30%, 50%, 70%, and 90%). Mixed powde...

An inelastic hyperspectral Scheimpflug lidar system is developed for range-resolved oil pollution detection and discrimination. A theory of system parametric design is built for aquatic circumstances, and laser-induced fluorescence spectra with an excitation wavelength of 446 nm are employed to detect oil pollution. Seven kinds of typical oil sampl...

In this study, a portable confocal hyperspectral microscope is developed. In traditional confocal laser scanning microscopes, scan lens and tube lens are utilized to achieve a conjugate relationship between the galvanometer and the back focal plane of the objective, in order to achieve a better resolution. However, these lenses make it difficult to...