Search

As grasping behaviors in real packaging scenarios are apt to be influenced by various disturbances, visual grasping prediction systems have suffered from the poor robustness and low detection accuracy. In this study, an intelligent robotic grasp framework (RTnet) underpinned by a linear global attention mechanism has been proposed to achieve the highly robust robot grasp prediction in real packaging factory scenarios. First, to reduce the computational resources, an optimized linear attention mechanism has been developed in the robotic grasping process. Then, a local window shifting algorithm has been adapted to collect feature information and then integrate global features through the hierarchical design of up and down sampling. To further improve the developed framework with the capability of mitigating noise interference, a self-normalizing feature architecture has been established to empower its robust learning capabilities. Moreover, a grasping dataset in the real operational environment (RealCornell) has been generated to realize a transition to real grasping scenarios. To evaluate the performance of the proposed model, its grasp prediction has been experimentally examined on the Cornell dataset, the RealCornell dataset, and the real scenarios. Results have shown that RTnet has achieved a maximum accuracy of 98.31% on the Cornell dataset and 93.87% on complex RealCornell dataset. Under the consideration of real packaging situations, the proposed model have also demonstrated the high levels of accuracy and robustness in terms of grasping detection. Summarily, RTnet has provided a valuable insight into the advanced deployment and implementation of robotic grasping in the packaging industry.

Graph neural networks(GNNs) in heterogeneous graph has shown superior performance and attracted considerable research interest. However, many applications require GNNs to make predictions on test examples that are distributionally different from training ones, while task-specific labeled data is often arduously expensive to obtain. An effective approach to this challenge is to pre-train an expressive GNN model on unlabeled data, and then fine-tune it on a downstream task of interest. While pre-training has been demonstrated effectively in homogeneous graph, it remains an open question to pre-train a GNN in heterogeneous graph as it contains different types of nodes and edges, which leads to new challenges on structure heterogeneity for graph pre-training. To capture the structural and semantic properties of heterogeneous graphs simultaneously, in this paper, we develop a new strategy for Pre-training Heterogeneous Graph Neural Networks(PHGNN). The key to the success of PHGNN is that PHGNN innovatively proposed to use two different tasks to capture two kinds of similarities in heterogeneous graph: the similarities between nodes with the same type and the similarities between nodes with the different type. In addition, PHGNN proposed an attribute type prediction task to preserve node attributes information. We systematically study pre-training on two real-world heterogeneous graphs. The results demonstrate that PHGNN improves generalization significantly across downstream tasks.

Large electromechanical equipment typically operates under low-speed, heavy-duty conditions, significantly increasing the likelihood of bearing failures. The reduced speed diminishes the frequency of fault impacts per unit time, rendering them prone to being obscured within intricate vibration signals. To address this challenge, a novel fault weak feature extraction method is proposed for low-speed, heavy-duty operations. Initially, Improved Variational Modal Decomposition (IVMD) is employed to efficiently isolate the primary signal components from diverse signals and reconstruct them. Subsequently, a Stacking Autoencoder (SAE) is leveraged to extract higher-order features from each reconstructed signal spectrum, markedly enhancing the discriminative power between distinct signals. Lastly, Random Forest (RF) is utilized to establish a precise mapping between these higher-order features and fault category labels, enabling effective diagnosis of bearing health status. This integrated method effectively tackles the difficulties of weak fault feature extraction and classification under low-speed, heavy-duty conditions. Experimental data validates the method’s accuracy, achieving a remarkable 97% success rate.

The rapid development of micro synthetic aperture radar (SAR) satellites provides a new means of all-day, all-weather, and high-frequency Earth observation. However, the geometric stability of micro SAR satellites is weak, and the geometric positioning error of SAR images is large. The traditional method improves the geometric positioning accuracy (GPA) of satellite images by adding well-distributed ground control points (GCPs), but it is time-consuming and laborious. This study proposed a method for improving the GPA of micro SAR satellites with large positioning errors. The proposed method designs a multilevel and multiscale SAR image matching framework, utilizes high-precision spaceborne SAR digital orthophoto image (DOM), and realizes high-precision virtual GCPs (VGCPs) extraction between micro SAR satellite images and DOM through image matching. Finally, using the matched VGCPs directly participate in the geometric correction to improve the GPA of the micro SAR images. The experiment adopted Gaofen-3 (GF-3) 10-m resolution DOM product as the control data. VGCPs were extracted and GPA was evaluated for five Luojia-2 (LJ-2) SAR images (3-m resolution, stripmap mode). The results show that the GPA of the LJ-2 image is at the kilometer level, and it can be improved to better than 5 m (relative to GF-3 DOM) after processing, which proves the validity and reliability of the proposed method. The proposed method can provide support for the cooperative processing of multisource SAR satellite images.

Electroencephalogram (EEG) signals are electrical signals generated by the activity between neurons in the brain and are already extensively applied for seizure prediction. Semi-supervised learning (SSL) has been applied in seizure prediction studies, as acquiring labeled EEG data can be time-consuming and costly. Current available SSL methods use far fewer labeled EEG samples than unlabeled EEG samples to train models. However, these methods have two limitations. First, the scarcity of EEG data limits the generalizability of the model. Second, the imbalance in the numbers of labeled and unlabeled EEG samples during model training leads to a potential distribution mismatch between the two sample types. Therefore, a novel semi-supervised seizure prediction method is designed in this study. First, an EEG data augmentation network is used to augment EEG signals at the signal level. Then, the labels of the original and augmented unlabeled EEG data are acquired from the spectrograms of EEG signals. Subsequently, the acquired labels are sharpened and the convex combination of labeled and unlabeled data is computed. Finally, deep pairwise representation alignment is performed to predict seizures. Several experiments are conducted using the proposed method on the Children’s Hospital Boston-Massachusetts Institute of Technology EEG database and the Kaggle Epilepsy Prediction Challenge EEG database. The results show that the proposed semi-supervised seizure prediction method is effective and reliable, given that it provides satisfactory results with as few as 25 labeled data per class when trained on these two datasets.

For fault diagnosis, it is important to effectively leverage the inherent characteristics of small datasets, but it is rarely considered in many the existing deep learning approaches. To this end, a dual-path network model based on multiple attention mechanisms is proposed in this work. The proposed model enriches the features of small sample by combining one-dimensional (1-D) frequency signals with two-dimensional (2-D) time-frequency images. A 1-D attention mechanism is applied in the 1-D frequency extraction path to focus classification on sensitive frequency information, while an improved global attention mechanism is added to the 2-D feature extraction path, which refines the key features and reduces interference resulting from noise in the image data. Moreover, the stability of the learning process is enhanced through the application of a combinatorial loss function composed of label smoothing regularization and gradient harmonizing mechanism loss functions. Finally, the diagnostic performance of the proposed method is validated using two different public fault diagnosis datasets in comparison with the state-of-the-art methods.

Bohai Sea is China's inland sea, its complex marine and atmospheric optical properties pose a challenge to the application of satellite data to retrieve chlorophyll-a (CHLA) concentration with high accuracy. The high accuracy retrieval results of CHLA require simultaneous consideration of the adaptability of atmospheric correction algorithms and CHLA retrieval models. In this study, four atmospheric correction methods [the standard atmospheric correction algorithms of GDPS1.4.1 and GDPS2.0, the standard near-infrared atmospheric correction algorithm of NASA (Seadas_Default), and management unit of the North Sea mathematical models (Seadas_MUMM)] and four CHLA retrieval models (OC2, YOC, OC3G, and OC2M-HI) were selected in the process of applying geostationary ocean color imager (GOCI) data to retrieve CHLA in Bohai Sea. Based on the in situ data, the adaptability of their pairwise combinations in retrieval of CHLA in Bohai Sea was evaluated. The results indicate that the OC2 and OC3G models significantly overestimated the CHLA. The combination of the Seadas_Default atmospheric correction algorithm with the YOC CHLA retrieval model, or the combination of the Seadas_MUMM atmospheric correction algorithm with the YOC CHLA retrieval model, is more suitable for the retrieval of CHLA using GOCI data in Bohai Sea. In addition, this study shows that the CHLA obtained based on the data from eight-scene GOCI data were different to the data obtained based on single-scene GOCI data (approximating traditional polar-orbiting satellite sensor data) in daily, monthly, and yearly average results. The monthly mean difference between the two is the most significant, ranging from -0.66 to 1.49 μg/l.

The t-rail electrode is an effective method to enhance the silicon optoelectronic modulator's performance. To design and optimize T-rail electrodes, engineers often rely on finite-element numerical simulations that require complex device modeling and enormous computing resources. In this paper, we present an equivalent circuit model for carrier-depletion-based push-pull silicon modulators with T-rail electrodes. The analytical solution for the bandwidth of the modulator can be derived from the equivalent circuit. The utilization of the analytical solution offers advantages in terms of memory conservation and flexibility. The values calculated by the equivalent circuit model are in excellent agreement with the numerical full-wave HFSS simulations. Hence, the proposed model can accurately and efficiently develop silicon optical modulators.

The electromagnetic radiation signal generated by loaded coal and rock is widely used to predict coal and rock dynamic disasters. However, due to the presence of significant electromagnetic interference at the coal mine site, the accuracy of the collected signals is insufficient. For the collected noisy charge induction signals, the wavelet threshold function of traditional denoising methods has problems such as non-progressiveness and discontinuity at the threshold. In order to achieve a better signal noise reduction effect, this paper proposes a collection based on an improved wavelet threshold (IWT) function and an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) combined denoising algorithm. It overcomes mode aliasing and optimizes signal smoothness. Firstly, the algorithm is used to decompose the noisy signal and calculate the intrinsic mode function (IMF) and correlation coefficient of each order to distinguish the noise from the correlated signal. Then, the IMF component dominated by the signal is reconstructed to complete the denoising. The simulation and experimental results show that this algorithm can effectively remove noise in charge induction signals, and its signal-to-noise ratio (SNR) is improved by 2.3482 and 0.095 compared to six algorithms such as IWT and VMD, respectively. Compared with four algorithms, including the improved threshold function and the improved threshold function combined with Ensemble Empirical Mode Decomposition (EEMD), its noise-to-noise ratio (Rnn) decreased by 3.112, showing good noise reduction performance. The results presented in this paper provide a new method for collecting real charge induction signals.

Synthetic aperture radar (SAR) is an active remote sensing system that utilizes radar to produce images of the Earth's surface. Due to its ability to operate under diverse weather conditions and throughout the day, SAR has gained significant attention in both civilian and military domains. The utilization of multiview SAR sequences enables the acquisition of a more comprehensive range of information than a single image, and facilitates adaptation to diverse scenarios, thereby enhancing the ability to accommodate variations in samples. Drawing inspiration from the Transformer architecture, this article proposes a multiview SAR target recognition method, called Res-Xformer, that not only deconstructs the deep learning procedure into single image feature extraction and sequence feature fusion but also divides the task of sequence feature extraction into sequence information fusion and feature channel fusion. Different from the Transformers focusing on the attention mechanism to fuse sequence information, alternative fusion methods such as multilayer perceptron (MLP) and pooling are also proposed in this study. Experimental results using the Moving and Stationary Target Acquisition and Recognition dataset demonstrate that the proposed method performs well across various operational conditions, with MLP and pooling as sequence token mixers yielding comparable performance to the attention mechanism.

Phase unwrapping (PU) is one of the core procedures of interferometric synthetic aperture radar. The traditional PU methods estimate the absolute phase gradient based on the assumption that phase is spatially continuous. However, this assumption is often not true due to the large-phase gradient. In this article, a local enhanced U-net (LEU-Net) was developed to determine the phase discontinuity of synthetic aperture radar (SAR) differential interferograms in areas with substantial elevation change. The phase discontinuity in interferograms predicted by this network was used as a priori information for the phase unwrapping maximum flow\minimum cut algorithm (PUMA). Within the LEU-Net, an interactive compression module was used to reduce the loss of detail information and ensure the ability of the model to segment discrete phase discontinuities; and a cross-fusion module was used to fuse multilevel features and highlight the location information while suppressing noise. The phase discontinuity information predicted by the LEU-Net guided the PUMA to obtain the unwrapped phase through iterative optimization; and in the meantime, the information can suppress the propagation of PU errors from areas of low signal-to-noise areas to areas of high quality. Experiments on simulated and real SAR interferograms with large-phase gradients showed that our method has a more robust PU capability.

Leakage current is one of the important parameters reflecting the operation status of distribution network-type surge arresters. At this stage, the polymagnetic current sensor has the advantages of miniaturization and high accuracy for leakage current measurement, but the complexity of electromagnetic interference in the field easily introduces more noise interference signals, which limits the performance of the polymagnetic current sensor for field application. To this end, this paper proposes a leakage current measurement method for distribution network-type surge arresters based on EEMD-SVD and low-rank RBF neural network methods. Firstly, the measured leakage current is decomposed by Ensemble Empirical Mode Decomposition (EEMD) to decompose the modal components containing eigenfrequencies, and then Singular Value Decomposition (SVD) is used to extract the non-eigenfrequency signals. Finally, a low-rank Radical Basis Function (RBF) neural network is used to approximate the leakage current signal after de-interference and combined with the Gaussian window function to remove the white noise interference. The experimental results show that the signal-to-noise ratio of the polymagnetic current sensor is improved by about 20dB and the maximum average absolute error is only 2.62%, which can truly reflect the leakage current of the network-type lightning arrester.

Transcranial ultrasound stimulation (TUS) is a noninvasive brain neuromodulation technique. The application of TUS for Alzheimer’s disease (AD) therapy has not been widely studied. In this study, a long-term course (28 days) of TUS was used to stimulate the hippocampus of APP/PS1 mice. We examined the modulatory effect of TUS on behavior and neural oscillation in AD mice. We found that TUS can 1) improve the learning and memory abilities of AD mice; 2) reduce the phase-amplitude coupling of delta-epsilon, delta-gamma and theta-gamma frequency bands of local field potential, and increase the relative power of epsilon frequency bands in AD mice; 3) reduce the spike firing rate of interneurons and inhibit the phase-locked angle deflection between the theta frequency bands and the spikes of the two types of neurons that develops with the progression of the disease in AD mice. In summary, we demonstrate that TUS could effectively improve cognitive behavior and modulate neural oscillation with AD.

Predicting outage areas for affected customers in a disaster requires real-time meteorological data, which can be challenging to obtain and process with high quality. In particular, missing or inaccurate data can occur in certain regions or during extreme weather events. Therefore, a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) based algorithm is proposed to improve the accuracy and reduce the prediction time in the outage region. Ground-based automated weather observatories are used to obtain real-time weather data, including extreme weather data. These data are pre-processed to enhance data quality and accuracy by removing duplicates, filling in missing values, handling anomalies, normalizing input variables, and reducing dimensionality. Based on the results of the data pre-processing, the outage rate was calculated for different types of meteorological disasters and the geographical characteristics of the outages were analyzed. This analysis provides insights into the impact of different types of meteorological disasters on power outages and helps improve the accuracy of predictive models. The proposed algorithm employs a CNN neural network to capture spatial and temporal information from raw meteorological data by stacking convolution and pooling layers. The extracted features are then organized and output by fully connected layers, laying the foundation for subsequent time series modeling. An LSTM network is further utilized to construct a prediction model for the outage area, which takes as input the feature extraction results of the meteorological data. By integrating the temporal dimension information of meteorological data, the model outputs accurate predictions for the outage area. The experimental results demonstrate a consistent outcome with the actual test results, achieving high prediction accuracy with a short prediction time of 4.3 s and a maximum non-outage detection rate of 2%. Therefore, the proposed algorithm proves to be significant for accurate and fast prediction of outage areas for affected customers in real world applications.

Remote sensing sensor platforms are typically located at a significant distance from the ground, ranging from several hundred meters to hundreds of kilometers. This means that, compared to natural images, remote sensing images (RSI) have larger coverage areas and more complex information. The larger size and data volume of RSI presents challenges for computer vision matching algorithms (MAs), making it difficult to apply them directly to RSI matching. Moreover, a matching framework for multisource RSIs capable of large-scale processing by integrating multiple MAs with the entire RSI as input is presently lacking. This study proposes a tie points (TPs) matching framework of multisource remote sensing images based on the geometric and radiation characteristics of RSI. First, RSI is divided into different grids and undergoes local geometry correction. Next, matching between slice images is performed by MAs. Finally, TPs are generated by mapping matched points in multiple slice images to the whole RSI using a geometric processing model. Six representative MAs including artificial feature MAs and deep learning algorithms are integrated into the framework to match TPs from different RSI. Results demonstrate the extraction of TPs for multisource RSI, validating the framework's efficacy. In addition, a large-scale TPs matching test for deep learning MA is performed by using 13 synthetic aperture radar images (10-m resolution) with TPs root mean square error of 0.368 pixels, further confirming the framework's reliability.

Hyperspectral images are widely use due to their rich spectral information. Meanwhile, the difficult acquisition of data labels makes unsupervised classification attracts attention. Subspace clustering as an unsupervised classification method is widely used for hyperspectral image analysis because of its excellent performance and robustness. However, conventional subspace clustering does not consider the nonlinear structure of hyperspectral data, and deep subspace clustering tends to ignore the intrinsic structure of hyperspectral data. To address these problems, we developed a self-expressive learning network, ResSENet, for hyperspectral data; we then proposed the application of ResSENet under graph constraints (GC-ResSENet), considering the intrinsic graph structure of the data. Unlike conventional deep subspace clustering, our model discards the self-expressive layer; self-expressive coefficients between datasets are directly solved by the data using network parameters. Hyperparameters are used in the joint loss to balance the self-expressive properties of the data and the graph constraint terms. We evaluated GC-ResSENet by applying it to four well-known datasets, and our network achieved optimal performance. In addition, because of its abandonment of the self-expressive layer, ResSENet is theoretically capable of clustering with large datasets; thus, we evaluated it using two large datasets.

In order to realize the automatic dispensing and remote monitoring of emulsions in the coal mine production process and improve the efficiency and safety of coal mine production, an automatic dispensing and remote monitoring system of emulsions based on the IoT platform was built. The APRUS X industrial network management adapter in this system is the upper computer of Siemens S7-200SMART PLC and the data transmission hub between the dispensing system and the MixIOT cloud platform, while the MixIOT cloud platform realizes the functions of data visualization and remote control terminal; the PLC acts as the control core and monitors the weight sensors, solenoid valves, DC brushless booster pumps, agitators and other equipment at the same time. The data collected by the sensors are uploaded to the cloud through IOT technology and combined with real-time data to realize automatic dispensing and remote control of emulsion. Meanwhile, remote monitoring is realized through Web page and apps, which is convenient for managers to monitor the production site in real time and realize intelligent production. There are 27 groups of liquid dispensing experiments, and the average relative error of emulsion concentration ratio obtained after 27 groups of tests is 4.8%, and the accuracy requirement meets the actual engineering needs. The research results are of great significance to promote the intelligent production of coal mines and the application of industrial IOT technology, and provide specific reference and references for the research and application in related fields.

In this paper, we investigate the performance of hybrid free-space optical (FSO) and radio frequency (RF) relay systems in satellite terrestrial integrated network where the effect of weather conditions is taking into consideration. Specifically, we design three different relay schemes in different weather conditions based on high-altitude platform station (HAPS) and reconfigurable intelligent surface (RIS) assisted unmanned aerial vehicle (UAV). The RF link is modeled by shadowed Rician model while the FSO link is assumed to follow the Gamma-Gamma distribution in the presence of path loss, pointing errors and angle-of-arrival (AOA) fluctuations where it operates under either intensity modulation with direct detection (IM/DD) or heterodyne detection (HD). The performance of our proposed scheme is evaluated in terms of outage probability, average bit-error rate (BER) and ergodic capacity in the form of bivariate Fox's H function and bivariate Meijer's G function. Moreover, asymptotic results and diversity order for the outage and average BER at high signal-to-noise ratio (SNR) regime are also obtained to get more insights. Finally, numerical results and comparison scheme are provided to verify the accuracy and superiority of our scheme.

Change detection is a basic remote sensing image processing task in agriculture, urban planning, disaster assessment, and other fields. Fully convolutional Siamese network is currently employed as the primary solution for remote sensing change detection. Fully convolutional networks lead to loss of accurate spatial location information during downsampling. This directly affects the edge accuracy of remote sensing change detection. In the prior method, researchers introduce attention mechanisms and dense networks to improve the accuracy of change detection methods. These approaches often result in a proliferation of network parameters and increased computational requirements, undermining their suitability for real-world, time-sensitive applications, such as disaster relief efforts. With the objective of addressing computational and parameter inefficiencies of existing change detection models, we propose the lightweight composite convolution network (LWCCNet). This involves the construction of a downsampling network based on a composite convolution operator, in conjunction with a novel spatial attention mechanism-edge attention module. In addition, we construct a lightweight feature pyramid network to fuse multiscale features. We conduct experiments on the change detection dataset (CDD) and the WHU building change detection dataset. On the CDD dataset, LWCCNet reduces the number of parameters by 86.3${\%}$ while the accuracy is reduced by only 1.9${\%}$, indicating that LWCCNet achieves better lightweighting while ensuring accuracy.

The position accuracy of the third mirror’s Z axis will directly affect the measurement accuracy of the optical telescope, so it is necessary to improve the position accuracy. In order to improve the control performances of the third mirror’s Z-axis of a large optical telescope, this paper proposes a novel control algorithm based on hyperbolic tangent function (tanh). The proposed algorithm is used to replace the traditional integral separated proportional-integral (IS-PI) controller which is widely used in industry servo control system. Firstly, a mathematical model of the third mirror’s Z-axis is established. Secondly, a controller based on tanh is proposed, and the characteristics and stability of the algorithm are analyzed. Thirdly, the parameters tuning method of the proposed algorithm is given. The simulation results show that the proposed algorithm has higher tracking accuracy than IS-PI controller. Finally, experiments are carried out on a large optical telescope’s Z axis. In the experiments, IS-PI controller and tanh controller are compared, and the results show that the control performances are greatly improved.

Demand forecasting is an important method for dealing with the supply-demand relationship in social resource management. The demands of daily life discussed in this study are mainly about hotels, restaurants, gas stations, drugstores, shopping malls, etc. Accurate demand forecasting can help enterprises meet people’s needs properly, especially since COVID-19 has spread worldwide and significantly changed people’s lives. Previous studies have proven the effectiveness of statistical and deep learning models in demand forecasting, including the Autoregressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), and their variants. These methods focus only on one-step prediction of one demand and do not perform well in long-term or multi-step predictions. To address this issue, we propose a framework called the Deep Spatial-Temporal Neural Network (DSTNN) for multi-demand forecasting, which performs well in long short-term forecasting. The DSTNN adopts a transformer encoder-based structure and establishes a direct correlation between the demand distribution and spatiotemporal features. The transformer encoder is designed to learn high-order feature interactions. An auxiliary task of pandemic trend forecasting was used to help the main task learn about pandemic trends. As regional mobility affects the spread of COVID-19, a population mobility graph was designed to achieve a better spatial feature representation. The DSTNN eliminates the dependency on historical sequential data, which helps the model to perform well in long-term forecasting, and its performance does not obviously decline as the prediction period increases. The experimental results show that the DSTNN performs much better than conventional models, particularly in terms of long-term forecasting.

Generally, the position closed-loop control of stepper motors requires high-precision models, however stepper motors are complex nonlinear systems, and high-precision models are difficult to obtain. In order to solve this problem, a variable frequency controller based on hyperbolic tangent function is proposed. Similar to the proportional integral (PI) control method, this method does not require an exact model of the plant. The experimental results show that when tracking the sinusoidal signal ${180^ \circ }\sin (0.628t)$ , the tracking error of the Tanh control algorithm is reduced by nearly 60% compared with the PI control algorithm, which provides a new idea for the position closed-loop control of stepper motor.

Major depressive disorder (MDD) is a prevalent mental health condition and has become a pressing societal challenge. Early prediction of treatment response may aid in the rehabilitation engineering of depression, which is of great practical significance for the relief of suffering and burden of MDD. In this paper, we present a sequence modeling approach that uses data collected by passive sensing techniques to predict patients with an outcome of treatment responded defined by the reduction in clinical administrated scales. Hundreds of patients with MDD have been recruited from outpatient clinics at 4 psychiatric sites. Each has been delivered with a self-developed app to passively record their daily phone usage and physical data with minimal human action. An unavoidable dilemma in passive sensing is missing values. To overcome that, the proposed approach combined feature extraction and sequence modeling methods to fully utilize the pattern of missing values from longitudinal data. With no treatment constraints, it enables us to predict the treatment response of MDD 8–10 weeks before the completion of the treatment course, leaving time for preventative measures. Our work explored the feasibility of treatment response prediction using longitudinal passive sensing data and sparse ground truth, and also has the potential for preventing depression by forecasting treatment outcomes weeks in advance.

Previous scientific literature retrieval methods, which are based on mathematical expression, ignore the literature attributes and the association between the literature, and the retrieval accuracy was affected. In this study, literature retrieval model based on Graph Convolutional Network (GCN) is proposed. By extracting document attributes from a structured document dataset, an Attribute Relation Graph (ARG) is constructed. Using GCN to capture the dependencies among literature nodes and generate literature representations by information aggregation to realize graph-based literature modeling; Introducing the advantages of Hesitant Fuzzy Set (HFS) theory in multi-attribute decision-making to realize the similarity evaluation between mathematical query expressions and mathematical retrieval result expressions. Finally, the similarity between literature features and mathematical expressions is integrated to obtain the ordered output of scientific literature retrieval results. Experiments were conducted on the arXiv public dataset, and the average precision of the top 10 retrieval results was 0.892, and the average NDCG value of the top 10 rankings was 0.875.

Laser altimeters are capable of achieving fine mapping of the permanently shadowed regions (PSRs) of the Moon, which can provide fundamental topographic data for planetary missions. However, various factors can cause uncertainty in the geolocation of laser spots, which in turn causes terrain artifacts. In this article, we present an iterative self-constrained adjustment method to reduce the uncertainty of laser spot positioning. First, grid search was conducted for each altimetric profile from the lunar orbiter laser altimeter (LOLA), to minimize the weighted root-mean-square error (RMSE), constrained by the other altimetric profiles. Second, the updated profiles were iteratively adjusted until the adjustment value for the plane position converged. In addition, statistics from the standardized de-trended slope and residual were created to eliminate outliers, which were indeed some pseudo-topographic observations. In order to validate the results, the deviation of the elevation by projecting the adjusted laser profiles onto the improved LOLA digital elevation model (DEM) were calculated. The mean absolute error between the two is 0.25 m and the RMSE is 0.46 m. For the local terrain features with large differences, high resolution optical images were used for visual interpretation. The analysis shows that the obtained results appear to be more reasonable. Finally, using the corrected LOLA altimetric data, we made a new DEM of the PSRs within 89°S of the lunar south pole, which can provide a refined and reliable topographic dataset for follow-up research.

We report a 2.1 μm band mode-locked holmium-doped fiber (HDF) laser with tunable wavelength and pulse width in dissipative soliton resonance (DSR) region for the first time to our knowledge. A piece of HDF is pumped by a home-made 1.9 μm band fiber source, and a nonlinear-amplifying-loop-mirror structure is used to realize mode-locking operation with net anomalous dispersion. The repetition rate of pulse is 8.97 MHz. The pulse duration can be adjusted from 1.16 to 3.93 ns with the increase of pump power, while maintains a constant peak power. At the maximum pump power of 2.1 W, the pulse energy is up to 2 nJ. Moreover, the central wavelength is in the tunable range from 2054.4 to 2076 nm by appropriately adjusting the polarization controllers. We also explore the nonlinear correlation characteristic of DSR pulse by changing the length of high nonlinear fiber.

Accurate registration between synthetic aperture radar (SAR) images is the basis for high-precision geometric correction of SAR images. The feature points extracted by conventional feature extraction methods are unsatisfactory, which are affected by imaging geometric characteristics and speckle noise of SAR images. This article innovatively proposes a spaceborne SAR image feature learning framework to realize automatic sample generation and model training. It mainly includes two modules: The feature sample generation module based on the initial geometric information of spaceborne SAR. The initial rational polynomial coefficient (RPC) parameters of the spaceborne SAR are adopted to realize the initial positioning of the SAR image, and a variety of feature extraction operators are used to match the overlapping areas to obtain high-precision matching points, which are employed as training samples for image pairs; pseudo-Siamese feature learning network SARPointNet for SAR image feature learning. The pseudo-Siamese network is used to extract the feature points and descriptors of the sample image pairs. The feature optimization process is realized through the descriptor constraints between image pairs, which promotes the network to improve the accuracy of feature extraction. The proposed method has been tested in mountain, hilly, flatland, and urban scenarios, respectively. The results demonstrate that the correspondence points extracted by SARPointNet are evenly distributed, are in large quantities (at least ten times that of other methods), and achieve high precision (the root mean square error is less than 1 pixels), which shows great advantages over other methods.

A noise-like pulse (NLP) thulium-doped fiber laser based on single-wall carbon nanotubes (SWCNTs) has been proposed and demonstrated in this paper, which the microfiber decorated with SWCNTs work as a saturable absorber to generate an ultrafast pulse. The length of the cavity is 12 m, corresponding to the fundamental frequency of 17.2 MHz. NLP with a duration of 643 fs can be obtained at 1965 nm under the signal-to-noise ratio of 56 dB. This work may assist us to better understand the dynamic characteristic of NLP with the help of materials and offer a fine laser source to the subsequent applications just like the generation of supercontinuum.

Representativeness error caused by scale transformation (REST) is an intrinsic property of data assimilation, as assimilating new observations likely involves the fusion of multisource and multiscale data. Earlier studies focused on specific cases and failed to obtain a general concept. This study attempts to achieve a further understanding of REST in both theory and practice. Based on scale-related definitions and formulations, the statistical RESTs of observation errors and analysis errors are deduced in stochastic ensemble data assimilation. Experiments based on ensemble Kalman filter are conducted to validate the interpretability of the proposed formulations. A synthetic experiment uses the stochastic Lorenz model as the forecasting operator, and a real-world experiment employs a simple biosphere model as the forecasting operator and uses a series of mixed ground-based and remote sensing soil moisture observations. The results confirm that REST should be proportional to the scale difference when assimilating direct observations and both system states and observations are homogeneous processes. Due to the nonlinearity in modeling, assimilation, and scale transformation, increasing RESTs are found if the scale of the observation is much larger than that of the state space, or multiscale observations are added into the assimilation system. Quantifying REST improves the understanding of uncertainty in data assimilation, but further studies on REST are required in both theory and practice, for example, REST correlates with other errors in forcing, parameters, and models, and introduces an observation operator to assimilate indirect observations.

Synthetic aperture radar (SAR) is widely used in terrain classification, object detection, and other fields. Compared with anchor-based detectors, anchor-free detectors remove the anchor mechanism and implement detection box encoding in a more elegant form. However, anchor-free detectors are limited by complex scenes caused by geometric transformations, such as overlaying, shadow, vertex displacement during SAR imaging. And the scattered power distribution of noise is similar to the edge of the object, making it difficult for the detector to locate the edge of the SAR object accurately. In order to alleviate these problems, we propose a high-speed and high-performance SAR image anchor-free detector. First, we propose a shallow feature refinement (SFR) module to effectively extract and retain the detailed information of objects, while coping with deformed complex scenes. Second, we analyze the optimization focus of the detector at different training iterations and propose iteration-aware loss to guide the detector, making the detector more accurately locate the edge of the object disturbed by the noise scattered power distribution. Third, number estimation helps to detect objects with more flexible criteria in box selection without manual labor. Compared with mainstream optical object detectors and SAR dedicated detectors, our method achieves the best speed-accuracy tradeoff on the SAR-ship dataset, with 96.4% average precision when the value of intersection over union is 50% ($AP_{50}$) at 64.9 frames per second. The experimental results prove the effectiveness of our method.

Correcting the relative geometric errors is an essential step in the generation of digital ortho maps and digital surface models from GaoFen-7 (GF-7) regional images, in which the block adjustment (BA) technology is always adopted. However, the traditional BA methods are typically ineffective and not universal due to their dependence on sufficient control data. Although recently developed BA methods using a large number of observation images can substantially eliminate the additional control data, they are not suitable for the BA with few images in the absence of sufficient control data, which is a situation that is often encountered in practical applications. In such situation, the inevitable elevation error caused by the initial positioning error of images will transfer and accumulate in the horizontal direction of the block, resulting in the inconsistent geometric accuracies among images. Aiming at this problem, this article proposed a practical method of combining a stable free BA and a posteriori compensation. Through the stepwise correction and integration of geometric errors, the elevation error and its cumulative effect in free BA could be eliminated at a small control cost, and a robust result with consistent accuracy could be achieved, even with only a few observation images. This method was validated through experiments performed on GF-7 stereo images, and the satisfactory results indicated that it effectively restrained error accumulation and improved the geo-positioning consistency of regional GF-7 images.

It has been a long-standing topic to detect the formation of urban agglomerations; however, current studies remain constrained by the definition of criterion or the usage of remote sensing technique, ignoring the impact of spatial interactions. This study aims to develop a methodological framework to detect and evaluate urban agglomerations by fusing human activities and their spatial interactions. To verify our methods, experiments were conducted in Mainland China, where nighttime light imageries and human movement data were used. 1) Urban boundaries of all cities were extracted from nighttime light imageries using the Lorenz curve and the watershed method; 2) they were combined with human movement data to construct the spatial interaction network, and urban agglomerations were then derived using the network community detection method; 3) the development levels of urban agglomerations were evaluated quantitatively according to the National Thirteenth Five-Year Plan (NTFP). Experimental results suggest that a total of 17 urban agglomerations were derived, and compared with those planned in the NTFP, they can be classified into four categories with different levels of development. Our methods can enrich studies on detection of urban agglomerations, while experimental results can provide decision-making support for regional management and development of planning policies.

In this article authors present a prototype of a dichroic scanning nonlinear optical microscope with a photodetecting polymer Scanner and a customized objective lens with an external back focal plane. Two types of two-axis electromagnetic actuated polymer scanners are manufactured by a standard printed circuit board (PCB) fabrication process. Both scanners share the same overall dimensions of 45 mm by 45 mm. The maximum optical scan angle of a type I scanner is 13 degrees at 420 Hz by the fast axis in resonance and 4 degrees by the slow axis in the linear mode. The type II scanner shows similar performance. A scan lens as well as a tube lens in a conventional nonlinear optical microscope are replaced by placing such a polymer scanner on the external back focal plane of a customized water immersion objective lens. The objective lens has the back working distance of 6 mm, the numerical aperture of 0.7, the front working distance of 1.16 mm, a field of view of 1.67 mm in the diameter and the angle of view of +/−15 degrees. Preliminary imaging results are also presented.

This paper presents a correlation-based all-digital background estimation method for timing skew mismatches in time-interleaved analog-to-digital converters (TI-ADCs). Exploiting the first-order approximation of an autocorrelation function of the digital output of each Sub-ADC, the timing skew mismatch between adjacent channels is identified, which can be used for analog or digital correction methods. The proposed estimation method has a low computation complexity and a simple implementation structure, because it basically only requires multiply-accumulate and average operations. In addition, the proposed method is not limited in the number of channels, does not need any pilot injection, is basically suitable for any bandlimited analog input signal, possess a large estimation range of mismatches and is immune against gain and offset mismatches to a certain extent. Simulation results of a four-channel TI-ADC behavioral model and measurement results from a commercial 12-bit 3.6-GS/s TI-ADC show the effectiveness and superiority of the proposed estimation method.

Accurate estimation of subjective assessment plays an essential role in not only speech quality perception, but also communication environment assessment. Traditional speech quality perception is almost always with an environment, in which the content of speech can be heard clearly. Unlike speech quality assessment, in extreme communication environment with ultralow SNR, like short wave channel with active jamming, speech intelligibility is impaired. Under this condition, subjective assessment as absolute category ranking (ACR) whose scores are made by experienced staffs, and many objective measurements cannot handle this situation. In this paper, we propose information damage level (IDL) to replace subjective ACR as subjective assessment. IDL is the average of scores which is marked on the subjective recognition rate. Under extreme communication environment, it can effectively avoid excessive differences from people to people to some extent. We also provide a new dataset collected in an environment with active jamming, whose speech files are recoded under three different environments, named as indoor simulation environment (EN1), outdoor simulation environment (EN2), and outdoor real environment (EN3). We also benchmark a novel open framework on random forest for direct predicting subjective assessment by combining all possible objective measurements. Experiment results prove the effectiveness of our open framework together with our dataset.

Recently, Internet finance is increasingly popular. However, bad debt has become a serious threat to Internet financial companies. The fraud detection models commonly used in conventional financial companies is logistic regression. Although it is interpretable, the accuracy of the logistic regression still remains to be improved. This paper takes a large public loan dataset, e.g. Lending club, for example, to explore the potential of applying deep neural network for fraud detection. We first fill the missing values by a random forest. Then, an XGBoost algorithm is employed to select the most discriminate features. After that, we propose to use a synthetic minority oversampling technique to deal with the sample imbalance. With the preprocessed data, we design a deep neural network for Internet loan fraud detection. Extensive experiments have been conducted to demonstrate the outperformance of the deep neural network compared with the commonly-used models. Such a simple yet effective model may brighten the application of deep learning in anti-fraud for Internet loans, which would benefit the financial engineers in small and medium Internet financial companies.

The submodule capacitor voltage fluctuation of the modular multilevel converter (MMC) leads to the increase of the size of the submodule capacitor, the increase of system cost, and the decrease of system reliability and stability. In order to solve the problem, the mathematical model of circulating current voltage drop is derived, and the mechanism of circulating current on the submodule capacitor voltage fluctuation is studied through the submodule switching function. A control strategy for suppressing the submodule capacitor voltage fluctuation is proposed, which is based on circulating current voltage drop balance (CCVDB). The simulation model is built to compare three control strategies for suppressing submodule capacitor voltage fluctuation under the step changes of DC bus voltage and active power, namely, the circulating current is suppressed to zero, the second-order component of the submodule capacitor voltage is suppressed to zero, and the CCVDB strategy. The research results show that the CCVDB strategy can effectively suppress the submodule capacitor voltage fluctuation, reduce the capacitance demand, and reduce the total harmonic distortion (THD) of the submodule capacitor voltage, which verifies the feasibility and effectiveness of the CCVDB strategy.

Since the limitations of the user equipment, they do not have enough computing power to process large amounts of data. At the same time, it’s unbearable for users to spend a long time uploading data to the remote cloud center. In order to solve these problems above, the concept of mobile edge computing(MEC) is proposed. The computing and storage resources are placed close to the end equipment, reducing the transmission delay. MEC can meet the high real-time requirements of the user equipment. In the real-time face recognition application scenario, a three-layer hierarchy MS-CE is proposed for the shortcoming of the traditional centralized cloud center. And the distributed MEC servers are utilized at the MEC layer to provide parallel computing capabilities. Aiming at the problem of how to perform task scheduling in a geographically distributed MEC server, we propose a task scheduling based queue algorithm(TSBQ), which considers the data transmission delay and server load, and carries out a reasonable task allocation policy. We evaluate the MS-CE and TSBQ through simulation experiments. We can find the MS-CE architecture are better than others and TSBQ is more effective than Corral and Greedy.

With the development of renewable energy technology, distributed power supply mode with multi energy and multi-directional power flow including utility grid, renewable energy and energy storage unit has gradually become a research hotspot. An AC/DC hybrid multi-port power routing (MPPR) system which based on partial power conversion (PPC) of dual DC buses is proposed in this paper. The photovoltaic (PV) port, the battery port and two DC voltage buses form a power router. PV maximum power point tracking (MPPT) and high-voltage ride through (HVRT) of the grid-tied inverter are implemented by the same auxiliary port voltage modulation. The PPC based PV conversion features that only the power determined by voltage difference between PV panel and the series connected DC bus is dealt with, which significantly reduces the loss compared to the full power conversion (FPC) for PV. The detailed control schemes of all converters and energy transmit are given. The simulation and experimental results verify the effectiveness of the proposed scheme.

Synthesizing intended programs from user-specified input-output examples, also known as Programming by Examples (PBE), is a challenging problem in program synthesis, and has been applied to a wide range of domains. A key challenge in PBE is to efficiently discover a user-intended program in the search space that can be exponentially large. In this work, we propose a method for automatic synthesis of functional programs on list manipulation, by using offline-trained Recurrent Neural Network (RNN) models to guide the program search. We adopt miniKanren, an embedded domain-specific language for flexible relational programming, as an underlying top-down deductive search engine of candidate programs that are consistent with input-output examples. Our approach targets an easy and effective integration of deep learning techniques in making better PBE systems and combines two technical ideas on generating diverse training dataset and designing rich feature embeddings of probable subproblems for synthesis generated by deductive search. The offline-learned model is then used in PBE to guide the top-down deductive search with specific strategies. To practically manipulate data structures of lists, our method synthesizes functional programs with popular higher-order combinators including $\texttt {map}$ , $\texttt {foldl}$ and $\texttt {foldr}$ . We have implemented our method and evaluated it with challenging program synthesis tasks on list manipulation. The experiments show promising results on the performance of our method compared to related state-of-the-art inductive synthesizers.

Post-stroke care encounters challenges, including high cost, lack of professionals, and insufficient rehabilitation state evaluation. Computer technology can alleviate these issues, as it allows health care professionals (HCP) to quantify the workload and thus enhance rehabilitation care quality. In this paper, a novel multi-model fusion method, terms as pose dual-stream network (PDSN), is devised, aiming to test the feasibility of monitoring the training actions of rehabilitating stroke patients in care management. In particular, this deep-learning-based algorithm combines human pose estimation and dual-stream networks in an innovative way. We utilize an improved OpenPose to estimate human pose from videos obtained by the low-cost monocular camera. In dual-stream networks, the spatial and motion streams are flexibly integrated. The spatial stream network combines the Gated Recurrent Unit (GRU) and attention mechanism to extract spatiotemporal data, while the motion stream network is composed of improved multi-layer 1D Convolutional Neural Networks (CNN), which enhanced by causal and dilated convolution skillfully. Additionally, an adaptive weight fusion strategy is used to fuse the two networks for the final action classification. Results show high accuracy on two public datasets and a dataset created by us, which validate the superiority and feasibility of our method.

Data fusion can effectively improve the accuracy of remotely sensed (RS) soil moisture (SM) products. Understanding the error structures of RS SM products is beneficial for formulating a data fusion scheme. In this article, a data fusion scheme is examined on the Tibetan Plateau, and the Soil Moisture Active Passive mission, Soil Moisture and Ocean Salinity mission, and Advanced Microwave Scanning Radiometer 2 products are used as the experimental input datasets. The RS apparent thermal inertia (ATI) is transformed into SM values as the reference data with reliable systemic variability. The ATI-based SM, along with three RS SM products, is introduced into the triple collocation (TC) method to decompose the errors of the three RS SM products into systemic and random errors at each RS pixel. Due to the presence of systemic errors, the temporal mean values and amplitudes of the three RS SM products were calibrated by those of the ATI-based SM. The rescaled anomalies (including amplitude and random error) were merged according to their random errors estimated by the TC method, and then the merged anomalies were added to the temporal mean values of the ATI-based SM to obtain the final merged results. Compared with the merged European Space Agency Climate Change Initiative passive SM product and input SM datasets, the merged results in this article exhibit optimal accuracy. The scheme for merging RS SM products shows high data fusion performance and can be further considered a reliable way to obtain a high-quality merged RS SM dataset.

Nitrogen dioxide (NO2) as one of the key pollutants in the atmosphere, is an important trace gas of traffic emissions or industrial activities. The coronavirus disease (COVID-19) epidemic preventions have significantly reduced the emissions of NO2, which provided an opportunity to disentangling the contributions to global emissions in major epidemic regions. Here, we quantified the influences of lockdown measures on the NO2 concentration at a global scale during the two waves of the COVID-19 epidemic, and disentangled the relative contributions of the major epidemic regions (i.e., China, Europe, and USA) on the global NO2 densities due to the lockdown measures. Our results revealed that there was an evident decrease of global NO2 in 2020, especially in China, Europe, and USA. The global mean NO2 densities decreased approximately 2.1 and 3.9 μmol/m2 in the first and second waves of the epidemic. While in the Northern Hemisphere, NO2 densities decreased 6.6 μmol/m2 on average during the first stage of the epidemic, and a slight increase (i.e., 1.7 μmol/m2) in the second wave of the epidemic. The magnitudes and durations of the second wave were much smaller than the first wave of the coronavirus infections. The strict lockdown measures implemented in China significantly decreased the NO2 concentration, which therefore the largest contributor in the first wave decreases of global average NO2 densities. Intervention at an early stage would be beneficial to the preventions of epidemic situation. Our study could provide references for economic loss assessments and economic recovery strategies during the stages ofpost-pandemic.

The Tatun Volcano Group (TVG) is located at the northern coast of Taiwan Island with only 15 km to the Taipei metropolis. Recent geothermal and geochemical studies suggest that the TVG is dormant-active rather than extinct, implying the eruption potential to devastate the nearby area of 7 million inhabitants. Although some geodetic tools (e.g., global positioning system and precise leveling) have been utilized to analyze the activity associated with volcanism, it is still challenging for conventional interferometric synthetic aperture radar (InSAR) technique to obtain clear ground surface movement at the TVG due to decorrelation and complex atmospheric delay. In this study, we present an improved multi-temporal InSAR (MTInSAR) approach to explore the temporal evolution and spatial extent of displacement at the TVG from 19 L-band ALOS/PALSAR images acquired between 2007 and 2011. Stratified atmospheric delays and orbit error are corrected with a patch-based joint model as different phase components can be separated according to their distinct spatiotemporal characteristics. The spatial deformation patterns indicate an uplift of 10 mm/year at the SW-NE ridge and subsidence with a rate of 10 mm/year at the E-W ridge. The temporal deformation variations at Mt. Chihshin and Huangzuei have high similarity with correlation coefficients of 0.9 above. This indicates that the subsidence at Mts. Chihshin and Huangzuei might be caused by released hydrothermal fluids related to tectono-magmatic activity within the hydrothermal system beneath the TVG.

Semiconductor optical amplifiers (SOAs) offer direct electrical injection, power consumption, integration, and anti-radiation advantages over optical fiber amplifiers. However, saturation output power and gain bandwidth have been limited in traditional structure SOAs. We demonstrate a monolithic integrated SOA with broad spectrum, high power, high gain, and small spectral linewidth expansion. The device adopts a two-stage amplified large optical cavity structure, and a lower optical field confinement factor is obtained by adjusting the thickness of the waveguide layer. The lower optical field confinement factor is conducive to improving the coupling efficiency and the maximum output power. Our device, fabricated only by standard i-line lithography with micron-scale precision, obtains excellent and stable performance. When the input power is set to 1 mW, the output power is 419 mW with a gain of 26.23 dB. When the input power is set to 25 mW at 25 °C, the output power increases to 600 mW with a gain of 13.8 dB. The corresponding gain bandwidth of 3 dB measures at least 70 nm. The spectral linewidth after the SOA is 1.15 times wider than that of the seed laser.

Semantic segmentation of remote-sensing imagery strives to assign a pixel-wise semantic label. Since encoder-decoder networks have demonstrated tremendous success in natural image semantic segmentation, the adoption and extension of this kind of method are transferring such superior performance for the problems in remote-sensing. Facing the high-altitude angle of imaging and complex and diverse ground objects of remote-sensing data, it is necessary to strengthen the features’ distinguishability by enhancing the network’s capability. Nevertheless, the existing methods suffer from the structural stereotype, leveraging the short-range and long-range contextual information insufficiently. Attempting to address the problems mentioned above, a hierarchical self-attention embedded neural network with dense connection for remote sensing image semantic segmentation (HSDCN) is proposed. In the encoder stage, multiple self-attention modules (SAM) are embedded to model pixel-wise and channel-wise relationships at various scales hierarchically, making the representations more refined and discriminative. Then the dense connections are used to fuse the heterogeneous features. Thus, the network could produce logical and reasonable clues for labeling pixels. The extensive experiments are conducted on ISPRS Vaihingen and Potsdam benchmarks. And the results reveal significant improvements in comparison with other state-of-the-art methods.

The personalized recommendation has become increasingly prevalent in real-world applications, to help users in discovering items of interest. Graph Convolutional Network (GCN) has achieved great success and become a new state-of-the-art for collaborative filtering. However, most of the existing GCN based methods can only capture information about the user’s purchase (or click) history, reflecting only one aspect of the user preferences and item characteristics. To provide more accurate recommendations, we need to go beyond modeling user-item interactions and take auxiliary information into consideration. In this paper, we propose a Light GCN based Aspect-level Collaborative Filtering model (LGC-ACF) to exploit multi-aspect user-item interaction information. First, we construct aspect-level user-item interaction graphs according to the interaction history and the knowledge information of items, then feed them to a delicately designed Light GCN based model to learn aspect-level representations of users and items. Finally, the representations of all aspects and all propagation layers are fused for recommendation. We apply LGC-ACF to three datasets: Movielens, Amazon, and Taobao. The experiment results show that LGC-ACF achieves average NDCG improvements of 5.31%, 4.06%, 14.9% in Movielens, Amazon, and Taobao datasets, respectively, compared with state-of-the-art baselines for recommendation.

In this paper, three-dimensional (3D) beamforming characteristics and applications in fifth generation (5G) mobile communications have been studied by considering the physical structure of array antennas, and the properties of the 3D beam pattern formed by planar, rectangular array antennas. Array beam gains are formulated according to rectangular array antennas. The effect of array antenna configuration on 3D beamforming is studied especially according to the building height. The field trial and measurement results have been presented for single and multiple mobile users. The field trial results show that (1) The total sum rate from all users can be increased multiple times (i.e., 3 to 4 times) as large as that of a single user. When the number of users is larger than 8, the sum rate becomes saturated; (2) Users with uniform angular distribution can achieve larger sum rate than users with centralized distribution due to space separation; (3) The performance of the multi-antenna system is best under static-user conditions, with it dropping considerably for mobile conditions, even by more than 50% due to poor channel state information estimation; (4) In case of 3D beamforming, good coverage performance can be achieved in medium or high buildings.

One of the main reasons that affecting the stability of power systems is low frequency oscillation (LFO). The existence of noise influences the accuracy of LFO mode identification extracted from wide-area measurement system (WAMS). The wavelet threshold de-noising is widely used in signal processing. In this paper, wavelet soft threshold is illustrated to attenuate the noise of LFO signal, the optimal wavelet basis and decomposition level for de-noising LFO signal with noise are obtained and verified by experiments. Following the signal de-noising, an improved Matrix Pencil (MP) algorithm is used for mode identification of LFO. This improvement particularly lies in the ratio of adjacent singular entropy increment difference (RASEID) designed as an adaptive order determination method in the MP algorithm proposed in the paper. RASEID not only makes the MP algorithm adaptive, but also enhances the stability of the order determination in the mode identification process. The proposed method ensures the accuracy of mode identification with lower sensitivity to noise interference. Finally, the validity of the proposed method is verified by three cases studies. The first study is on the analysis of synthetic signal typically performed in many literatures. The second study is to identify the mode of active power LFO signal which is generated by IEEE four-generator and two-area system given with disturbance on RT-LAB experimental platform. The third study is the oscillation analysis of the actual LFO data in the North American power grid. The results validate the feasibility of the proposed method for mode identification of noisy LFO.

Research into rolling bearing fault diagnosis methods is of great significance because rolling bearings are a key part of mechanical equipment. The effect of iterative generalized demodulation (IGD) on the demodulation of the fundamental frequency component is obvious in the fault diagnosis of rolling bearings at variable speeds. However, there is a problem; the frequency curve of the demodulation octave frequency component overlaps, and multiple determinations of the bandpass filter parameters produce an artificial error that leads to the misdiagnosis of faults. Therefore, a method for rolling bearing fault diagnosis based on adaptive generalized demodulation (AGD) is proposed. First, the resonance band is intercepted by the fast kurtogram and its envelope results. Second, the adaptive chirp mode decomposition (ACMD) algorithm is used to decompose the envelope signal, the relationship between the time and frequency of the signal is clearly characterized by the form of multimedia pictures, and the instantaneous frequency of each signal component is calculated. Third, the instantaneous frequency is used as the phase function to perform generalized demodulation for each signal component. Last, all the demodulated signals are accumulated, and a fast Fourier transform (FFT) is used to extract the fault's characteristic frequency. The proposed method is compared with IGD by using simulation signals and actual bearing signals collected by sensors under the Internet of Things (IoT). An adaptive diagnosis function is realized through this proposed method at variable speeds. Moreover, the average frequency spectrum identification rate of rolling bearing faults is improved by more than 2.6 times compared with that of the IGD in the simulation signal verification and by more than 1.7 times compared with that of the IGD in the real signal verification. This method is strongly immune to noise.