Search

Effective preservation of high-frequency information such as edges and textures remains an open challenge in image denoising, with existing methods susceptible to noisy interference and poor consistency with human visual quality. To address this limitation, we propose a novel enhanced edge information network for image denoising via feature multi-modulation attention, namely ${\text {M}}^{2}$ A ${\text {E}}^{2}$ Net, for complex denoising tasks on both real and synthetic images. This method divides the conventional maximum likelihood estimation framework into three sub-problems solved collaboratively: (i) image reconstruction, (ii) gradient estimation, and (iii) detail refinement. These objectives are implemented through the integration of two sub-networks: Reconstruction and gradient estimation networks. First, we propose an end-to-end gradient loss function imposed on gradient estimation network to acquire noise-free gradient maps, which guides ${\text {M}}^{2}$ A ${\text {E}}^{2}$ Net to focus on regions abundant with rich edges and contours. Furthermore, to achieve effective feature representation of noisy images synchronously and dynamically in the reconstruction network, we propose a novel feature multi-modulation attention module to adaptively integrate gradient images obtained by the gradient estimation network into input images as an edge prior information. The Laplacian of Gaussian sharpening loss function is also proposed to enhance the preservation of edges and textures, as demonstrated by higher SSIM metric and perceptual quality of denoised images. Our approach leverages the advantages of deep learning while benefiting from classical filter-based methods. Experimental results demonstrate the promising performance of ${\text {M}}^{2}$ A ${\text {E}}^{2}$ Net in terms of PSNR and SSIM metrics and better visual quality of denoised images. The code is available at https://github.com/Wangzt1121/M2AE2Net.

The forest is an important part of carbon resources. Forest growing stock volume (GSV) is an important parameter of forest. The Water Cloud Model (WCM) is a simple equation that describes the interaction between ground objects and electromagnetic waves. It has also been applied in the estimation of forest GSV. When estimating GSV, the WCM equation parameters are usually calculated using least squares, but the least squares method relies on field reference data. The subsequent WCM development algorithm BIOMASAR uses a sliding window method that does not rely on measured data. However, the sliding window method is inefficient and can easily lead to missing pixels. We designed the backscatter/fractional vegetation coverage (FVC) feature space based on WCM and BIOMASAR to estimate forest GSV. Comparing with the national forest inventory (NFI) reference dataset and the BIOMASAR algorithm results in the study area, the method is evaluated from three aspects: accuracy, efficiency, and texture. The results show that this method does not rely on actual reference data, and the efficiency is increased from 1661s in the sliding window to 663s. The correlation with the NFI reference data is 0.45, the RMSE is 116.2327 m3/ha, and the RRMSE is 64.86%. The accuracy is better than the BIOMASAR sliding window GSV results in this study area, and compared with Google Earth images of the same period, it is also more consistent with the field texture. In short, the backscatter/FVC feature space can efficiently obtain forest GSV estimates more consistent with field conditions without relying on measured data.

Anomaly detection has become one of the crucial tasks in hyperspectral images processing. However, most deep learning-based anomaly detection methods often suffer from the incapability of utilizing spatial–spectral information, which decreases the detection accuracy. To address this problem, we propose a novel hyperspectral anomaly detection method with a spatial–spectral dual-window mask transformer, termed as S2DWMTrans, which can fully extract features from global and local perspectives, and suppress the reconstruction of anomaly targets adaptively. Specifically, the dual-window mask transformer aggregates background information of the entire image from a global perspective to neutralize anomalies, and uses neighboring pixels in a dual-window to suppress anomaly reconstruction. An adaptive-weighted loss function is designed to further suppress anomaly reconstruction adaptively during network training process. According to our investigation, this is the first work to apply transformer to hyperspectral anomaly detection. Comparative experiments and ablation studies demonstrate that the proposed S2DWMTrans achieves competitive performance.

Optical metasurfaces, i.e., thin planar structures with subwavelength metal or dielectric unit cells, have attracted great interest due to their intriguing capabilities of shaping light in both linear and nonlinear regimes. However, their saturable absorption properties and corresponding applications have so far been rarely exploited. Here we report on passively mode-locked ytterbium-doped fiber lasers utilizing saturable metasurfaces made of periodically arranged gold nanorods. Three 400-nm-period metasurfaces with gold nanorods of different lengths (from 210 to 240 nm) and plasmonic resonances (from 968 to 1044 nm) are fabricated and found to exhibit nonlinear absorption with decent modulation depths, facilitating the formation of mode-locking states at 1060 nm. The corresponding lasers generate typically mode-locked pulses with the duration of ∼ 62.3 ps, repetition rate of 10.33 MHz, and signal-to-noise ratio of ∼74 dB. Our experimental results demonstrate that the gold nanorod-based metasurfaces can be used as relatively broadband mode-lockers in the 1-μm-wavelength range.

We propose a novel optical spatial differentiator to perform the differentiation computation in terahertz region based on the graphene metalines, which consist of graphene layers with different widths and chemical potentials. The numerical simulation results show that when beam waist size w>1.9λ, the metalines perform the first-order differentiation in the reflection spectrum with efficiency>97%, which can be theoretically demonstrated by using transfer matrix method. In order to further improve the performance of the differentiator, evolutionary algorithm, such as genetic algorithm, is used to inversely design the structure parameters and chemical potentials of graphene metalines. The optimization results show that some performance metrics of the differentiator, for example normalized root-mean-square deviation, are better than the previous structures. Obviously, the proposed graphene metalines combined with inverse design technology can achieve a high-performance optical spatial differentiator in terahertz region and provide a new way to design the photonics devices.

In this paper, a triple-layer hybrid optical orthogonal frequency division multiplexing (THO-OFDM) for intensity modulation with direct detection (IM/DD) systems with a high spectral efficiency is proposed. We combine N -point asymmetrically clipped optical orthogonal frequency division multiplexing (ACO-OFDM), N /2-point ACO-OFDM, and N /2-point pulse amplitude modulated discrete multitoned (PAM-DMT) in a single frame for simultaneous transmission. The time- and frequency-domain demodulation methods are introduced by fully exploiting the special structure of the proposed THO-OFDM. Theoretical analysis show that, the proposed THO-OFDM can reach the spectral efficiency limit of the conventional layered ACO-OFDM (LACO-OFDM). Simulation results demonstrate that, the time-domain receiver offers improved bit error rate (BER) performance compared with the frequency-domain with ~40% reduced computation complexity when using 512 subcarriers. Furthermore, we show a 3 dB improvement in the peak-to-average power ratio (PAPR) compared with LACO-OFDM for the same three layers.

Structural thermal deformation is an important factor that affects the performance of MEMS devices. The mismatch of thermal expansion coefficient (CTE) between different materials is a major source. For single center supported MEMS devices, expansion difference between device and the substrate leads to the out of plane thermal deformation of the structure, resulting in the performance deterioration. This paper presents a thermal deformation suppression chip (TDSC) for single center supporting MEMS devices. It is fabricated by the MEMS process and consists of an upper plate and a lower plate. The materials of the upper and lower plates are the same as those of the device substrate and structure respectively. The principle of TDSC to suppress thermal deformation is material symmetry design. Single center anchor is adapted to connect the upper and lower plates. Besides, the center anchor can also isolate the packaging stress generated between TDSC and package shell. In this paper, center supported quadruple mass gyroscope (CSQMG) is used to verify the effect of TDSC. Finite element simulation shows that the thermal deformation suppression effect is determined by radius of TDSC anchor, and the deformation can be suppressed to 0. The conclusion is confirmed by White light interference experiment. In addition, the experiment results also show that the TDSC effectively reduce the thermal out-of-plane deformation after packaging, which is even better than that of the gyroscope without any packaging. It is further proved that TDSC can also significantly improve the temperature performance of the CSQMG. Moreover, the TDSC has simple process steps, small chip area, low cost, and is suitable for various MEMS devices with single center anchor.

The classical SURF algorithm has many disadvantages, such as high dimension of feature descriptor, large amount of computation, and low matching accuracy when the angle of rotation and angle of view is too large. To solve the above problems, an improved algorithm is proposed. Firstly, image preprocessing is carried out by image binarization, feature points are extracted by Hes-sian matrix, and then feature description is carried out by using circular neighborhood of feature points. Har wavelet response is used to establish descriptors for each feature point, and the normalized gray-level difference and second-order gradient in the neighborhood are calculated simultaneously to form a new feature descriptor. Finally, the RANSAC algorithm is used to eliminate the mismatch points. The algorithm does not Compared with the classical SURF algorithm, it has the advantage of speed, and makes full use of the gray level information and the detail information, so it has higher accuracy. Experimental results show that the algorithm has good robustness and stability to image blur, illumination difference, angle rotation, field of view transformation and so on. The algorithm is applied to remote sensing image stitching to obtain the stitched image with no obvious geometric shift and good edge connection. This algorithm is a kind of image registration algorithm with short time and high precision, which can meet the registration requirements of remote sensing image stitching.

With the proliferation of WiFi devices and infrastructures, the ubiquitous WiFi signals are used to transmit user data. Besides it is also capable of sensing and identifying human gestures. In this paper, we propose a WiFi-based gesture recognition system, namely WiGrus, which solves the problems of user privacy and energy consumption compared with the approaches using wearable sensors and depth cameras. WiGrus leverages the fine-grained Channel State Information (CSI) extracted from WiFi signals to recognize a set of hand gestures. First of all, we utilize timestamps attached to the extracted CSI values to split continuously received WiFi packets into gesture instances. Second, a Principal Component Analysis (PCA)-based method and the first order difference are employed to reduce the noise and mitigate multipath effects caused by the environment changes. Then, massive features are extracted from the processed CSI values to present the intrinsic characteristics of each gesture. Finally, a 2-stage-RF algorithm is proposed to classify the gestures. Our experiments are implemented with a wireless router and a Software Defined Radio (SDR) device, more specifically Universal Software Radio Peripheral (USRP), which are used as WiFi signal transmitter and receiver respectively. The experimental results demonstrate that WiGrus can achieve an average accuracy of 96% in Line-of-sight (LOS) scenario and 92% in Non-Line-of-Sight (NLOS) scenario in the office environment and is robust to the environment changes.

Mobile edge computing (MEC) that affords service to the vicinity of mobile devices (MDs) has become a key technology for future network. Offloading big data to the MEC server for preprocessing is an attractive approach of MDs. In this paper, we investigate data offloading from MDs to MEC servers, and a coalitional game-based pricing scheme is proposed. We apply coalitions to arrange data offloading of MDs as well as depict the offloading relationship between MDs and MEC servers. Furthermore, pricing is utilized as the stimuli for the offloading. In the proposed scheme, a scheduled MD in a coalition chooses an MEC server within the same coalition for offloading, and pays the selected MEC server for the MEC service. By analyzing the formulated coalitional non-transferable utility game, theoretical properties are derived and sufficient conditions for the core existence are given. Finally, the utility performances of the proposed scheme are demonstrated by numerical results.

The performance and parameters of neural networks have a positive correlation, and there are a lot of parameter redundancies in the existing neural network architectures. By exploring the channels relationship of the whole and part of the neural network, the architectures of the convolution network with the tradeoff between the parameters and the performance are obtained. Two network architectures are implemented by dividing the convolution kernels of one layer into multiple groups, thus ensuring that the network has more connections and fewer parameters. In these two network architectures, the information of one network flows from the whole to the part, which is called whole-to-part connected networks (WPNets), and the information of the other network flows from the part to the whole, which is called part-to-whole connected networks (PWNets). WPNets use the whole channel information to enhance partial channel information, and the PWNets use partial channel information to generate or enhance the whole channel information. We evaluate the proposed architectures on three competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet), and our models obtain comparable results even with far fewer parameters compared to many state of the arts. Our network architecture code is available at github.

Finding meaningful clustering patterns in data can be very challenging when the clusters are of arbitrary shapes, different sizes, or densities, and especially when the data set contains high percentage (e.g., 80%) of noise. Unfortunately, most existing clustering techniques cannot properly handle this tough situation and often result in dramatically deteriorating performance. In this paper, a purposefully designed clustering algorithm called Density-Based Multiscale Analysis for Clustering (DBMAC)-II is proposed, which is an improved version of the latest strong-noise clustering algorithm DBMAC. DBMAC is proposed under the assumption that all clusters are homogeneous and cannot work well when the data set contains clusters of varying densities. DBMAC-II overcomes the limitation of DBMAC by executing the multiscale analysis iteratively and can conduct strong noise-robust clustering without any strict assumption on the shapes and densities of clusters. In DBMAC-II, each data point or object is mapped into a feature space using its r-neighborhood statistics with different r (radius) values, which is similar to DBMAC. In general, the higher the value of r-neighborhood statistics, the more likely the object is considered as a “clustered”object. Instead of trying to find a single optimal r value, a set of radius values appropriate for separating “clustered”objects and “noisy”objects is identified, using a formal statistical method for multimodality test, referred to as multiscale analysis. For clusters with varying densities, multiscale analysis is applied to extract the clusters with the highest density from the current data set iteratively. Moreover, a statistical uniformity test for measuring clustering tendency is used as the self-adaptive stopping criterion of the iteration. Comprehensive experimental studies on a series of challenging benchmark data sets demonstrate that DBMAC-II is not only superior to classical density-based clustering approaches, including DBSCAN, OPTICS, and HDBSCAN, but also can consistently outperform the latest strong-noise robust clustering techniques, such as Skinny-dip.

Mixup is a neural network training method that generates new samples by linear interpolation of multiple samples and their labels. The mixup training method has better generalization ability than the traditional empirical risk minimization method (ERM). But there is a lack of a more intuitive understanding of why mixup will perform better. In this paper, several different sample mixing methods are used to test how neural networks learn and infer from mixed samples to illustrate how mixups work as a data augmentation method and how it regularizes neural networks. Then, a method of weighting noise perturbation was designed to visualize the loss functions of mixup and ERM training methods to analyze the properties of their high-dimensional decision surfaces. Finally, by analyzing the mixture of samples and their labels, a spatial mixup approach was proposed that achieved the state-of-the-art performance on the CIFAR and ImageNet data sets. This method also enables the generative adversarial nets to have more stable training process and more diverse sample generation ability.

A dynamically tunable electromagnetically induced transparency (EIT) system consisting of two coupled micro-ring resonators, one of which is embedded with a graphene layer, is proposed and numerically demonstrated. The effective refractive index of the graphene-based micro-ring resonator can be significantly tuned by varying the gate voltage applied on graphene, inducing significant modulation of the resonant wavelength of EIT transparency window over a wide spectral bandwidth. Typical tunability of the EIT resonance is approximately 1.62 nm/V around 1550 nm, which is much better than that based on a nanoelectromechanical EIT system. Such a configuration implies the possibility of constructing various optical devices toward realization of photon pulse trapping, optical modulation, and filtering on a chip.

Multiuser transmission with shared spectrum is investigated in the presence of energy harvesting. For each user, the data generation, the harvested energy arrival, and the channel state variation are, respectively, considered as stochastic processes. Each user aims to maximize its own average throughput. Accordingly, a stochastic game is formulated at first. Next, we analyze the stochastic game in infinite-stage and finite-stage scenarios, respectively. On the basis of theoretical studies, an iterated distributive algorithm with solving a linear problem in each iteration is designed for the infinite case. For the finite case, two distributed algorithms named CVPBI and GoPGA are proposed. Finally, the effectiveness of the proposed algorithms is demonstrated by simulations.