Search

Recording and sharing of educational or lecture videos has increased in recent years. Within these recordings, we find a large number of math-oriented lectures and tutorials which attract students of all levels. Many of the topics covered by these recordings are better explained using handwritten content on whiteboards or chalkboards. Hence, we find large numbers of lecture videos that feature the instructor writing on a surface. In this work, we propose a novel method for extraction and summarization of the handwritten content found in such videos. Our method is based on a fully convolutional network, FCN-LectureNet, which can extract the handwritten content from the video as binary images. These are further analyzed to identify the unique and stable units of content to produce a spatial-temporal index of handwritten content. A signal which approximates content deletion events is then built using information from the spatial-temporal index. The peaks of this signal are used to create temporal segments of the lecture based on the notion that sub-topics change when large portions of content are deleted. Finally, we use these segments to create an extractive summary of the handwritten content based on key-frames. This will facilitate content-based search and retrieval of these lecture videos. In this work, we also extend the AccessMath dataset to create a novel dataset for benchmarking of lecture video summarization called LectureMath. Our experiments on both datasets show that our novel method can outperform existing methods especially on the larger and more challenging dataset. Our code and data are publicly available.

Social relation, as the basic relation in our daily life, is vital for social action analysis. However, how to learn the social feature between people is still not tackled. In this work, we propose a gaze-aware graph convolutional network (GA-GCN) for social relation recognition, which targets discovering the context-aware social relation inference with gaze-aware attention. To predict the gaze direction, we apply a convolutional network trained with gaze direction loss. Then, we build a graph convolutional inference module, which is a two-stream graph inference with both gaze-aware attention and distance-aware attention. The attention can pick up relevant context objects for context-aware representation. We further introduce additional scene features and construct a multiple feature fusion module, which can adaptively learn social relation representation from both scene feature and context-aware feature. Extensive experiments on the PISC and the PIPA datasets demonstrate that our GA-GCN can find interesting contextual objects and achieves state-of-the-art performances.

In single-pixel imaging (SPI) systems, the traditional free-space photodetector can be easily interfered. Its size and configuration limit its applications in complex and restricted scenarios. In this study, we develop an SPI based on optical fibers. We use a single inclined face optical fiber as an SPI detection device to realize multidimensional remote imaging in slim channels. This system is immune to interference from external ambient lights, and it is flexible to be moved and rotated to implement endoscopic observation inside various narrow spaces, such as fluid tubes and electronic device gaps. Furthermore, it can be used in medical endoscopic imaging and medical intervention therapy.

Transient stability assessment (TSA) is of great importance in power system operation and control. One of the usual tasks in TSA is to estimate the critical clearing time (CCT) of a given fault under the given network topology and pre-fault power flow. Data-driven methods try to obtain models describing the mapping between these factors and the CCT from a large number of samples. However, the influence of network topology on CCT is hard to be analyzed and is often ignored, which makes the models inaccurate and unpractical. In this paper, a novel data-driven TSA model combining Mahalanobis kernel regression and ensemble learning is proposed to deal with the problem. The model is a weighted sum of several sub-models. Each sub-model only uses the data of one topology to construct a kernel regressor. The weights are determined by both the topological similarity and numerical similarity between the samples. The similarities are decided by the parameters in Mahalanobis distance, and the parameters are to be trained. To reduce the model complexity, sub-models within the same topology category share the same parameters. When estimating CCT, the model uses not only the sub-model which the sample topology belongs to, but also other sub-models. Thus, it avoids the problem that there may be too few data under some topologies. It also efficiently utilizes information of data under all the topologies. Moreover, its decision-making process is clear and understandable, and an effective training algorithm is also designed. Test results on both the IEEE 10-machine 39-bus and a real system verify the effectiveness of the proposed model.

The basic reproduction number plays an important role in exploring the dynamics of the epidemic models. Such value has been extensively used in the estimation of how severe an epidemic outbreak. Although several methods for calculating the basic reproduction number has been proposed, there isn't an effectively universal method to estimate such value. In this paper, we propose a general approach to calculate the explicit formulation of the basic reproduction number by the renewal equation. We apply such a method to estimate the basic reproduction number of many epidemic models on complex networks consisting of mean-field models, pairwise models, and edge-based compartmental models.

In this paper, a destination-assisted-jamming-based secure transmission is devised for an untrusted and energy harvesting relay network. In the proposed scheme, both source and destination split their energies into two parts for energy harvesting at the relay and transmissions of confidential and cooperative jamming signals, respectively. We formulate an optimization problem to maximize the achievable secrecy rate by jointly optimizing the energy splitting for the source and destination. The existence of the solution for the optimization is carefully analyzed. Due to high complexity of the joint optimization, an iterative optimization algorithm to obtain the near-optimal energy splitting policy is proposed. The numerical results validate that the optimized energy splitting policy leads to a higher achievable secrecy rate, and the iterative algorithm has a quick convergence rate.

We report a novel low-loss splicing approach for hollow-core photonic crystal fiber (HC-PCF) and conventional single mode fiber (SMF) by inserting an etched SMF tip into the hollow core. If the divergence solid angle (DSA) of the inserted fiber tip is smaller than the guiding solid angle (GSA) of HC-PCF, all the radiated optical power from the tip will be collected by PCF with an extremely high coupling efficiency and guided by the PCF based on photonic bandgap effect. A DSA-GSA matching model is proposed to calculate the efficiency for different splicing situations. In the optimal condition of GSA > DSA, the transmission loss could approach zero, only including the scattering loss of etched fiber. Then, we experimentally demonstrate the splicing between NKT HC19-1550-01 and SMF-28 with coupling efficiency of 84.5%, agreeing well with the theoretical result of 85%. In addition, our proposed method is coaxiality error free, angle deviation free, and insertion length insensitive. This new technique is expected to provide new possibilities in some special system designs for HC-PCF-based applications.

Reconstruction of highly accelerated dynamic magnetic resonance imaging (MRI) is of crucial importance for the medical diagnosis. The application of general robust principal component analysis (RPCA) to MRI can increase imaging speed and efficiency. However, conventional RPCA makes use of nuclear norm as convex surrogate of the rank function, whose drawbacks have been mentioned in plenty of literature. Recently, nonconvex surrogates of the rank function in RPCA have been widely investigated and proved to be tighter rank approximation than nuclear norm by the massive experimental results. Motivated by this, we propose a nonconvex alternating direction method based on nonconvex rank approximation to reconstruct dynamic MRI data from undersampled $k-t$ space data. We solve the associated nonconvex model by the alternating direction method and difference of convex programming. The convergence analysis provided guarantees the effectiveness of our algorithm. Experimental results on cardiac perfusion and cardiac cine MRI data demonstrate that our method outperforms the state-of-the-art MRI reconstruction methods in both image clarity and computation efficiency.

In this paper, we propose a coherent optical interleaved single-carrier frequency-division multiplexing (CO-I-SC-FDM) uplink scheme for the long-reach passive optical network (LR-PON). I-SC-FDM is a modified form of orthogonal frequency-division multiplexing (OFDM) with low computational complexity and low peak-to-average power ratio (PAPR). By designing a simple encoding structure, the computational complexity of the transmitter can be reduced from O(Nlog2(N)) to O(N). At the probability of 10-3, the PAPR of the I-SC-FDM is about 10.7 dB lower than that of conventional OFDM for quadrature phase-shift keying (QPSK) modulation. We experimentally demonstrate a 10-Gb/s CO-I-SC-FDM uplink scheme over 100-km standard single-mode fiber with 1:128-way passive splitter for LR-PON. In the conventional CO-OFDM uplink scheme, the splitting ratio is only 1:64. The proposed scheme shows great potential for application in access and metro networks.

We demonstrate a state variable fiber pulse laser based on a graphene-integrated microfiber device. A 3-D microfiber coil integrated with graphene acts as a polarization-sensitive saturable absorber. By adjusting the polarization controllers, stable Q-switched pulses and mode-locked rectangular pulses are observed. To our knowledge, this is the first study conducted to achieve switching between rectangular pulses and Q-switched pulses while keeping the pump power constant. The rectangular pulses have a pulsewidth and repetition rate of ~10 ns and ~940 kHz, respectively, and the Q-switched pulses have a pulsewidth and repetition rate of ~40 μs and ~4.8 kHz, respectively.

A heterodyne optical fiber current sensor based on a nanowire-grid (NWG) in-line polarizer is proposed. The NWG is a one-dimension Au grating fabricated on a fiber tip by focused ion beam machining. Because the grating pitch is far smaller than the wavelength of the incident light, it works like an in-line transflective fiber polarizer with both the transmission and reflection light power being measured simultaneously. Based on this unique feature, an all-fiber optical current sensor is demonstrated. The heterodyne measurement makes it withstand light power variations showing good sensing stability. As an all-fiber sensor, discrete and bulky optical components are saved, making it more reliable in practical applications.

We propose a surface plasmon interferometer based on a wedge metal waveguide. Working principles of the device are introduced through theoretical analyses. The interference performances are investigated by finite-difference time-domain (FDTD) method. Because of the special wedge geometry, the interferometer shows an extremely low coupling loss in comparison with the previous reports. An interesting surface plasmon polariton (SPP) coupling and splitting phenomenon is found at the wedge section. It is then split into two surface plasmon modes in two interference arms. Refractive index sensing applications of this interferometer are also discussed, which shows sensitivity as high as 6 × 103 nm/RIU.

An ~ 12-μm-long Bragg grating was engraved in an ~ 5-μm-diameter optical fiber tip by focused ion beam (FIB) milling. An ~ 10-dB extinction was achieved at 1570 nm with only 11 indentations. The grating was used for temperature sensing, and it exhibited a temperature sensitivity of ~ 22 pm°C.

We propose a photoelastic pressure sensor using a liquid crystal (LC) tunable fiber polarizer. The polarizer contains an etched single-mode fiber sealed in an LC cell. A low-index LC is used as a tunable birefringent cladding of the fiber. Any electric-field-induced LC director reorientation may thus affect the fiber's light-guiding characteristics. Different polarization states experience different LC cladding indices. The transmittance contrast between the x- and y-polarized modes reaches 31.9 dB, which is sensitive enough to monitor any polarization state change in the fiber. A pressure sensor with the sensitivity of 0.25 rad/N is thus demonstrated. The corresponding stress-optic coefficient of the fiber is measured at 1550-nm telecomm band. The advantages and applications of the LC tuning and sensing technologies are discussed.