Search

It is still a challenging task to perform the semantic segmentation with high accuracy due to the complexity of real picture scenes. Many semantic segmentation methods based on traditional deep learning insufficiently captured the semantic and appearance information of images, which put limit on their generality and robustness for various application scenes. Thus, in this paper, we proposed a novel strategy that reformulated the popularly used convolution operation to multi-layer convolutional sparse coding block in semantic segmentation method to ease the aforementioned deficiency. To prove the effectiveness of our idea, we chose the widely used U-Net model for the demonstration purpose, and we designed CSC-Unet model series based on U-Net. Through extensive analysis and experiments, we provided credible evidence showing that the multi-layer convolutional sparse coding block enables semantic segmentation model to converge faster, extract finer semantic and appearance information of images, and improve the ability to recover spatial detail information. The best CSC-Unet model significantly outperforms the results of the original U-Net on three public datasets with different scenarios, i.e., 87.14% vs. 84.71% on DeepCrack dataset, 68.91% vs. 67.09% on Nuclei dataset, and 53.68% vs. 48.82% on CamVid dataset, respectively. In addition, the proposed strategy could be possibly used to significantly improve segmentation performance of any semantic segmentation model that involves convolution operations and the corresponding code is available at https://github.com/NZWANG/CSC-Unet.

In complex environments of wireless communication, to improve the communication anti-interference performance, quasi-synchronous frequency-hopping (FH) communication system requires low-hit-zone (LHZ) FH sequence sets with optimal Hamming correlation (HC) properties and flexible parameters. In this paper, based upon cyclotomy theory, two classes of LHZ FH sequence sets with flexible parameters not included in the literatures are designed, and the periodic HC properties of which are analyzed. It turns out that the designed LHZ FH sequence sets are optimal on the maximum periodic HC. And thus the new constructions can be used to provide optimal FH sequence sets for quasi-synchronous FH communication system.

For semantic segmentation of remote sensing images (RSI), trade-off between representation power and location accuracy is quite important. How to get the trade-off effectively is an open question, where current approaches of utilizing very deep models result in complex models with large memory consumption. In contrast to previous work that utilizes dilated convolutions or deep models, we propose a novel two-stream deep neural network for semantic segmentation of RSI (RSI-Net) to obtain improved performance through modeling and propagating spatial contextual structure effectively and a decoding scheme with image-level and graph-level combination. The first component explicitly models correlations between adjacent land covers and conduct flexible convolution on arbitrarily irregular image regions by using graph convolutional network, while densely connected atrous convolution network (DenseAtrousCNet) with multi-scale atrous convolution can expand the receptive fields and obtain image global information. Extensive experiments are implemented on the Vaihingen, Potsdam and Gaofen RSI datasets, where the comparison results demonstrate the superior performance of RSI-Net in terms of overall accuracy (91.83%, 93.31% and 93.67% on three datasets, respectively), F1 score (90.3%, 91.49% and 89.35% on three datasets, respectively) and kappa coefficient (89.46%, 90.46% and 90.37% on three datasets, respectively) when compared with six state-of-the-art RSI semantic segmentation methods.

Wildfires could pose a significant danger to electrical transmission lines and cause considerable losses to the power grids and residents nearby. Previous studies of preventing wildfire damages to electrical transmission lines mostly analyze wildfire and power system security independently due to their differences in disciplines and cannot satisfy the requirement of the power grid for active and timely responses. In this paper, we have designed an integrated wildfire early warning system framework for power grids, taking prediction of wildfires and early warning of line outage probability together. First, the proposed model simulates the spatiotemporal process of wildfires via a geography cellular automata model and predicts when and where wildfires initially get into the security buffer of an electrical transmission line. It is developed in the context of electrical transmission line operating with various situations of topography, vegetation, wind and, especially, multiple ignition points. Second, we have proposed a line outage model (LOM), based on wildfire prediction and breakdown mechanisms of the air gap, to predict the breakdown probability varying with time and the most vulnerable poles at the holistic line scale. Finally, to illustrate the validation and rationality of our proposed system, a case study for a 500-kV transmission line near Miyi county, China, is presented, and the results under various wildfire situations are studied and compared. By integrating wildfire prediction into the LOM and alarming the holistic line breakdown probability along time, this paper makes a significant contribution in the early warning system to prevent transmission lines to be damaged by wildfires, illustrating the related breakdown mechanisms at the line operation level rather than laboratory experiments only. Meanwhile, the implementation of cellular automata model under comprehensive environmental conditions and simulation of the breakdown probability for the 500-kV transmission line could serve as references for other studies in the community.

The interdisciplinary research of the radio-frequency (RF) signal-based wireless power and information transfer is expected to address the energy shortage issue in the massively deployed low-power Internet of Things devices. Different from conventional wireless powered communication networks (WPCNs), the hybrid base station (H-BS) adopts the simultaneous wireless information and power transfer (SWIPT) for the sake of satisfying the downlink data and energy requests of the multiple user equipments (UEs). The energy harvested from the downlink transmissions can be depleted for supporting the UEs' uplink transmissions. Integrating SWIPT in the downlink transmission of the WPCN yields a generic integrated data and energy communication network, where the H-BS is equipped with multiple antennas and both the downlink and uplink transmissions are slotted in the time-domain. Furthermore, both the sum-throughput and the fair-throughput of the uplink transmissions are maximized by jointly optimizing the transmit beamformer of the H-BS in the spatial-domain, the time-slot allocation in the time-domain and the signal splitting strategies of the UEs in the power domain, while satisfying the UEs' minimum downlink transmission requirements. Due to the non-convexity of the problem, a low-complexity successive convex approximation-based algorithm is relied upon for obtaining the optimal resource allocation scheme in the time-domain, power-domain, and spatial-domain. The numerical results validate the efficiency of our proposed resource allocation algorithm and they also demonstrate that supporting low-rate data services during the downlink transmissions does not degrade the wireless power transfer and hence does not reduce the uplink throughput.

Imbalanced data exists commonly in machine learning classification applications. Popular classification algorithms are based on the assumption that data in different classes are roughly equally distributed; however, extremely skewed data, with instances from one class taking up most of the dataset, is not exceptional in practice. Thus, performance of algorithm often degrades significantly when encountering skewed data. Mitigating the problem caused by imbalanced data has been an open challenge for years, and previous researches mostly have proposed solutions from the perspectives of data re-sampling and algorithm improvement. In this paper, focusing on two-class imbalanced data, we have proposed a novel divergence-encouraging autoencoder (DEA) to explicitly learn features from both of the two classes and have designed an imbalanced data classification algorithm based on the proposed autoencoder. By encouraging maximization of divergence loss between different classes in the bottleneck layer, the proposed DEA can learn features for both majority and minority classes simultaneously. The training procedure of the proposed autoencoder is to alternately optimize reconstruction and divergence losses. After obtaining the features, we directly compute the cosine distances between the training and testing features and compare the median of distances between classes to perform classification. Experimental results illustrate that our algorithm outperform ordinary and loss-sensitive CNN models both in terms of performance evaluation metrics and convergence properties. To the best of our knowledge, this is the first paper proposed to solve the imbalanced data classification problem from the perspective of explicitly learning representations of different classes simultaneously. In addition, designing of the proposed DEA is also an innovative work, which could improve the performance of imbalanced data classification without data re-sampling and benefit future researches in the field.

This paper describes a new type of communication network called data and energy integrated communication networks (DEINs), which integrates the traditionally separate two processes, i.e., wireless information transfer (WIT) and wireless energy transfer (WET), fulfilling co-transmission of data and energy. In particular, the energy transmission using radio frequency is for the purpose of energy harvesting (EH) rather than information decoding. One driving force of the advent of DEINs is wireless big data, which comes from wireless sensors that produce a large amount of small piece of data. These sensors are typically powered by battery that drains sooner or later and will have to be taken out and then replaced or recharged. EH has emerged as a technology to wirelessly charge batteries in a contactless way. Recent research work has attempted to combine WET with WIT, typically under the label of simultaneous wireless information and power transfer. Such work in the literature largely focuses on the communication side of the whole wireless networks with particular emphasis on power allocation. The DEIN communication network proposed in this paper regards the convergence of WIT and WET as a full system that considers not only the physical layer but also the higher layers, such as media access control and information routing. After describing the DEIN concept and its high-level architecture/protocol stack, this paper presents two use cases focusing on the lower layer and the higher layer of a DEIN network, respectively. The lower layer use case is about a fair resource allocation algorithm, whereas the high-layer section introduces an efficient data forwarding scheme in combination with EH. The two case studies aim to give a better explanation of the DEIN concept. Some future research directions and challenges are also pointed out.

In this article, a scheme with a designed overmoded multigap circuit based on the TM13 mode and linearly distributed two electron beams is proposed for power enhancement of sub-THz extended interaction oscillators (EIOs). The analysis of the electromagnetic characteristics shows that the designed circuit can reproduce the resonant frequency, field distribution, and ohmic loss of the conventional multigap circuit with a single beam. When considering the same beam power injected into the designed and conventional circuit, the analysis of the beam loading shows the same effect on the circuit oscillation through the comparison in the electron conductance between two circuits. To examine the performance of the designed circuit, the EIOs are designed with two circuits at 0.22 THz. The analysis of the coupling coefficients between two circuits, combined with pic simulations, shows that the designed circuit produces the efficient interaction with two beams and, most importantly, two beam–wave interaction units can promote and achieve each other. The EIO with the designed circuit can deliver kW-level power at 0.22 THz. Accordingly, the proposed scheme has shown great potential in enhancing the power of sub-THz EIOs. [ABSTRACT FROM AUTHOR]

An effective overmoded circuit is proposed to support efficient interaction with two beams for the development of high power extended interaction oscillators (EIOs) toward sub-terahertz (THz) and higher frequency. The circuit design focuses on the effect of the frequency-determining parameter on mode characteristics of the $\mathrm{TM}_{{1} {n}}$ sequence. Frequency equivalence between $\mathrm{TM}_{{1} {n}}$ modes with different ${n}$ supports the mode selection for two-beam interaction. Accordingly, the TM13 mode is selected to duplicate the dispersion relation around the axial mode of $2\pi $ and ohmic loss of conventional TM11 mode used in conventional single-beam circuits. On this basis, the overmoded circuit is formed with two beams based on the TM13 mode. In order to verify its effectiveness, a 0.22-THz EIO is designed with the overmoded circuit and, most importantly, the effect of the uniformity of two beams on the EIO performance is simulated. When the current of one beam is assumed to be constant and that of another beam is increased, the power is increased linearly and then tends to be saturation. This shows good capability of the EIO in interacting with nonuniform two beams, which is critical for its stable operation. Simulation results have shown the achievement of the average power of 0.22-THz wave over 1.26 kW with double beams at 23 kV and each current of 0.2 A. [ABSTRACT FROM AUTHOR]

An improved topology of Extended interaction oscillators (EIO) is presented with enhanced efficiency at 0.35 THz. A bi-periodic ladder-type structure is used for optimizing the performance of the TM $_{{31}}- 2\pi $ mode operation, based on bi-periodic interaction gaps composed of staggered short and long slots. The bi-periodic interaction mechanism permits to operate the circuit with a standing-wave field between the $\pi $ and $2\pi $ modes, thus providing the potential to combine the advantages of both $\pi $ and $2\pi $ modes in terms of both coupling and output performance. The resulting transverse TM31 operating mechanism exhibits a good agreement with the bi-periodic structure in terms of coupling performance, with the short slot length positively correlated with the value of effective characteristic impedance ${M}^{\,{2}}{R}/{Q}$. The circuit demonstrates attractive coupling and output characteristics by optimization of ${M}^{\,{2}}{R}/{Q}$ and ${Q}_{e}$. A state-of-the-art value ${M}^{\,{2}}{R}/{Q}$ of 59.69 $\Omega $ and a moderate ${Q}_{e}$ of 2523.52 are achieved for a proposed eight-period ladder-type EIO. An RF power of 540 W is obtained with a 42.5 kV, 0.6 A sheet beam, with 2.1% electronic efficiency at 0.35 THz. [ABSTRACT FROM AUTHOR]