Search

Single-channel speech separation can be adopted in many applications. Time-frequency (T-F) masking is an effective method for single-channel speech separation. With advancements in deep learning, T-F masks have become used as a training target, achieving notable separation results. Among the numerous masks that have been proposed, the ideal binary mask (IBM), ideal ratio mask (IRM), Wiener filter (WF) and spectral magnitude mask (SMM) are commonly used and have proven effective, though their separation performance varies depending on the speech mixture and separation model. The existing approach mainly utilizes a single network to approximate the mask of the target speech. However, in mixed speech, there are segments where speech is mixed with other speech, segments where speech is mixed with silent intervals, and segments where high signal-to-noise ratio (SNR) speech is mixed due to pauses and variations in the speakers’ intonation and emphasis. In this paper, we attempt to use different networks to handle speech segments containing various mixtures. In addition to the existing network, we introduce a network (using the Rectified Linear Unit as activation functions) to specifically address segments containing a mixture of speech and silence, as well as segments with high SNR speech mixtures. We conducted evaluation experiments on the speech separation of two speakers using the four aforementioned masks as training targets. The performance improvements observed in the evaluation experiments demonstrate the effectiveness of our proposed method based on the joint network compared to the conventional method based on the single network.

Existing conversational recommender systems (CRS) use insufficient generality in incorporating external information using knowledge graphs. The recommendation module and generation module are loosely connected during model training and shallowly integrated during inference. A simple switching or copying mechanism is used to merge recommended items into generated responses. These problems significantly degrade the recommendation performance. To alleviate this problem, we propose a novel unified framework for collaboratively enhancing conversational recommendations using pre-trained language models and knowledge graphs (CollRec). We use a fine-tuned pre-trained language model to efficiently extract knowledge graphs from conversational text descriptions, perform entity-based recommendations based on the generated graph nodes and edges, and fine-tune a large-scale pre-trained language model to generate fluent and diverse responses. Experimental results on the WebNLG 2020 Challenge dataset, ReDial dataset, and Reddit-Movie dataset show that our CollRec model significantly outperforms the state-of-the-art methods.

Due to the relatively low operational reliability of the MMC valve, it is necessary to establish corresponding test methods to examine the tolerance of the valve to electrical and thermal stresses under engineering requirements through operational tests, to verify its operational reliability. To meet the operational testing requirements for higher voltage and larger current converter valves, an appropriate test topology is required. The paper first analyzes the voltage and current stress characteristics of capacitor components and MMC converter valves in ±800 kV/8 GW Flexible DC transmission system, providing a theoretical basis for the stress of operational testing valve, requiring that to be equivalent to actual operating conditions. Addressing the operational testing demands for high voltage and large current converter valves in the ±800 kV/8 GW flexible DC transmission MMC, a test topology is proposed. This topology can meet the operating conditions of higher voltage and larger current and simultaneously conduct tests on valve segments and modules, significantly improving operational testing efficiency and reducing testing costs. Additionally, a control method applicable to this test topology is proposed, enabling flexible adjustment of DC and AC in the test circuit, thereby enhancing the equivalency of operational tests. In this paper, the proposed topology and control method have enhanced testing efficiency, reduced testing costs, improved testing equivalence, and enhanced the precision of current and capacitor voltage control. Finally, simulations and experiments are constructed to validate the feasibility and effectiveness of the proposed operational test topology and its control method through simulations and experiments.

Typhoons' rapid variation makes effective hazard prevention difficult, which remains a challenging research topic requiring novel observation technology and analysis methodology. We developed an automatic methodology for mining the information on the rapid variation of typhoon cloud-top motion from high-spatial-temporal-resolution satellite cloud images, which has steps of estimating and decomposing the typhoon cloud-top motion, locating the typhoon cloud-top center, and analyzing the motion and center position variations. The Gaofen-4 satellite acquired 50-m-1-min-resolution cloud images of Typhoon Megi's (2016) landfall over mountainous Taiwan Island. The data provide an excellent chance to explore what such high-spatial-temporal-resolution successive observations can tell about the cloud-top motion variation of a typhoon landfall using the methodology. Before its landfall, Typhoon Megi's cloud-top center dramatically increased the northwestward migration speed from ∼22 to ∼58 km/h in only ∼2.5 h. It also presented small-scale oscillation along its migration path with a 1.6-to-2.4-km amplitude, a 31-to-33-km spatial period, and a 1.46-to-1.48-h temporal period. We averaged the motion speeds (velocity magnitudes) in each motion field to assess the field strength. The so-calculated cloud-top average rotation speed of Typhoon Megi decreased quickly during landfall. Meanwhile, its average divergence speed increased dramatically by ∼30% in a short period of ∼2.5 h, reaching ∼2 times the rotation speed. This result means that Megi's cloud-top motion rapidly changed from a typical rotation motion-dominated status to a divergence motion-dominated status. This rapid transformation from typical rotation motion dominance to divergence motion dominance implies intense upward warm and moist airflows that cause severe precipitation.

Goal: Cervical cancer is one of the most common cancers in women worldwide, ranking among the top four. Unfortunately, it is also the fourth leading cause of cancer-related deaths among women, particularly in developing countries where incidence and mortality rates are higher compared to developed nations. Colposcopy can aid in the early detection of cervical lesions, but its effectiveness is limited in areas with limited medical resources and a lack of specialized physicians. Consequently, many cases are diagnosed at later stages, putting patients at significant risk. Methods: This paper proposes an automated colposcopic image analysis framework to address these challenges. The framework aims to reduce the labor costs associated with cervical precancer screening in undeserved regions and assist doctors in diagnosing patients. The core of the framework is the MFEM-CIN hybrid model, which combines Convolutional Neural Networks (CNN) and Transformer to aggregate the correlation between local and global features. This combined analysis of local and global information is scientifically useful in clinical diagnosis. In the model, MSFE and MSFF are utilized to extract and fuse multi-scale semantics. This preserves important shallow feature information and allows it to interact with the deep feature, enriching the semantics to some extent. Conclusions: The experimental results demonstrate an accuracy rate of 89.2% in identifying cervical intraepithelial neoplasia while maintaining a lightweight model. This performance exceeds the average accuracy achieved by professional physicians, indicating promising potential for practical application. Utilizing automated colposcopic image analysis and the MFEM-CIN model, this research offers a practical solution to reduce the burden on healthcare providers and improve the efficiency and accuracy of cervical cancer diagnosis in resource-constrained areas.

Under non-Gauss noise condition, the performance of traditional state estimation methods based on Gauss measurement noise will be greatly reduced. In order to solve this problem, a robust state estimation method based on Mahalanobis distance under non-Gauss noise is proposed in this paper. First of all, based on the Mahalanobis distance, the calculation method of optimal buffer length for PMU measurements is used, which can unify the SCADA measurements with PMU measurements in the same snapshot. Then, Based on the two-stage processing method, in the first stage, the SCADA measurements are used for filtering by using maximum likelihood estimator to obtain the estimated values, and then the estimated values in the first-stage are combined with PMU measurements as the second-stage measurements for filtering, and finally the final results are obtained. Based on the IEEE-39 buses system and IEEE-118 buses system, under Gaussian noise and non-Gaussian noise, the AEE results of proposed method are very small, and which are all within 10−3, numerical tests under different simulation conditions verified the robustness and effectiveness.

As shipping routes and resource exploration move toward high-latitude oceans, sea ice becomes a major threat to the safety of ship navigation, posing significant challenges to the shipping industry and offshore resource development. Continuous development of satellite remote sensing and deep learning has made large-scale and wide-ranging ship detection (SD) possible, which is of great significance for ship safety. However, existing ship datasets used for deep learning only include ship images in open waters (OW), such as ports and inland rivers. Currently, remote sensing datasets suitable for SD in ice-infested waters (IIW) are lacking. SD in IIW is more difficult than SD in OW because of complex background interference from sea ice. Thus, it is infeasible to directly use the features of ships in OW for SD in IIW. Herein, we propose a remote sensing SD dataset called IceRegionShip, which includes subdatasets IceRegionShip–red, green and blue (RGB) and IceRegionShip–ice region ship index (IRSI). IceRegionShip–IRSI consists of low-resolution images processed with IRSI. IceRegionShip–RGB and IceRegionShip–IRSI contain 11 436 and 9073 ship instances, respectively. IRSI was proposed to address false alarms caused by ice interference. To the best of our knowledge, this is the first dataset designed specifically for SD in IIW. In addition, the dataset was evaluated using several advanced detection algorithms, providing a benchmark for SD in IIW and demonstrating the effectiveness of IRSI for SD in low-resolution optical remote sensing images.

Accurate segmentation of lesion region from skin lesion images can provide guidance for medical experts to make an early diagnosis of skin cancer. In this study, we construct Recurrent Attentional Convolutional Networks (O-Net), which exploits the skin lesion’s attention class feature with a recurrent O-shape structure, in an iterative refinement strategy for skin lesion image segmentation. Inspired by the recently proposed attention class feature network, we integrate the attention class feature module into the proposed networks. The O-Net, with recurrent unit to iteratively refine the segmentation result, is designed to extract attention feature information and enable coarse-to-fine feature representation by iteratively integrating attention feature maps into network. Furthermore, O-Net calculates the attentional class feature by extracting attention information from the coarse segmentation result. Two currently popular datasets ISIC-2017 and PH2 are employed to explore the validity of our proposed model. The study provides detailed comparisons of our proposed network, the attention class feature network and Recurrent U-Net. O-Net achieved Dice coefficient by 87.04% on the ISIC-2017 dataset, 92.12% on the PH2 dataset with corresponding Jaccard indices of 80.36% and 86.15%, respectively, on the same dataset, which exhibits competitive performance for skin lesion image segmentation in results. The visual results also shown that more detailed tissues are extracted by O-Net than other methods.

In order to equip more high-energy pulse loads and improve power supply reliability, the vessel integrated power system (IPS) shows an increasing demand for high-voltage and large-capacity energy storage systems. Based on this background, this paper focuses on a super capacitor energy storage system based on a cascaded DC-DC converter composed of modular multilevel converter (MMC) and dual active bridges (DAB). The cascaded converter is called MMC-DAB for short. This paper analyzes the topology and modulation strategy of the system in detail. Taking into account the shortcomings of the traditional bidirectional power control strategy, this paper proposes a control strategy where the DAB module of each branch independently controls the voltage of the sub-module capacitor. The mathematical models of the traditional and the proposed control strategy are established. The stability of the two control strategies is analyzed and compared. Finally, a MW-level engineering prototype of the MMC-DAB energy storage system is designed and manufactured, and the effectiveness of the proposed control strategy was verified through experiments.

A novel dualband high-temperature superconducting (HTS) bandpass filter is proposed with group delay equalization for 5G emergency communication receivers with N41 and N79 bands in this paper. The dual-folded stub-loaded stepped impedance resonator (DSLSIR) is applied to realize dualband performance. The proposed DSLSIR has miniaturized circuit size and more design freedom to independently adjust the two designed center frequencies, comparing with other reported DSLSIRs. Two cross-coupling transmission lines are added at the top and bottom parts of the cascaded sixth-order filter with DSLSIRs to flatten the group delay of the passband at both bands. A dualband bandpass HTS filter etched on the YBCO superconducting material is fabricated to verify the proposed design method. The measurement is actualized at the temperature of 77 K to make YBCO microstrip lines perform superconducting features. The results show that the first center frequency is 2.595 GHz with 6.2% bandwidth for N41 band and 0.3 dB insertion loss while the second center frequency is 4.85 GHz with 2% bandwidth for N79 band and 0.4 dB insertion loss. Meanwhile, the group delays at the N41 and N79 bands keep 2 ns fluctuations in the 60% passband. The agreement between the measured and simulated results indicates that the proposed dualband bandpass HTS filter with low insertion loss, sharp roll-off skirt and flat group delay is a promising candidate for emergency communication receiver.

To obtain a high-precision bearing fault diagnosis of different speeds under intense background noise, a novel method called state-space principal component tracking filtering (SPCTF) is proposed for fault diagnosis. SPCTF can recover the fault information through the sequence of “Prediction – Measurement – Correction - Optimal estimation - Principal component extraction” from the polluted raw signals with noise. Firstly, the switch Kalman filter (SKF) is utilized to establish a dynamic filter model for time-series signals. And then, a feature tracking matrix is generated from the principal component features extracted from the optimal estimated state matrix of each time in the model. Thirdly, the principal component analysis extracts the effective fault feature from redundant information generated in the feature tracking matrix. Finally, the fault characteristic frequency is analyzed with the envelope spectrum to achieve a high-precision fault bearing diagnosis. The proposed method is demonstrated on signals from simulation signals and laboratory tests (several speeds), outperforming traditional methods.

In this paper, we demonstrate high-performance quasi-vertical GaN-on-Sapphire Schottky barrier diodes (SBD) with a reverse GaN p-n junction termination (RPN). The SBD has a current output of 1 kA/cm2 at $V_{F}=2.5$ V, a low $V_{on}$ of 0.66 V ± 0.017 V, a low $R_{on,sp}$ of 1.4 $\text{m}\Omega \cdot $ cm2, current ON/OFF ratio of over $10^{9}$ (−3 V~3 V). By introducing the RPN, the breakdown voltage can boost from 459 V to 1419 V, and power figure-of-merit (FOM) can reach 1438 MV/cm2. It is shown that the presence of the RPN with a suitable anode recess depth can generate an electric field (EF) opposite to the built-in EF at the center of the second top p-n junction, which can decrease the EF peak intensity and make the electric field more uniformly distributed inside the device. Finally, the leakage current of the SBD is inhibited and the breakdown voltage is increased.

Although emotion recognition techniques have been well developed, the understanding of the neural mechanism remains rudimentary. The traditional static network approach cannot reflect the entire brain activity at the time scale. Instead, a newly introduced temporal brain network is an optimal approach which can be used to investigate the dynamic functional connectivity (FC) of the human brain in different emotion states considering the time-varying brain regional interaction. In this study, we focused on emotion recognition and dynamic FC analysis with SEED dataset. First, multiband static networks were computed by the phase lag index (PLI). Then, subject-independent discriminative connection features of such static networks were selected to recognize the positive, neutral, and negative emotion types. In addition, we constructed the temporal brain network by sorting the static network according to time sequence. The experimental results show that the beta band is the most suitable for emotion recognition due to the best accuracy of 87.03%. And, the frontal and the temporal lobes are more sensitive to brain emotion-related activities. Moreover, we find the spatio-temporal topology of dynamic FC shows the small-world structure. Notably, the positive emotion is more distinguishable in the temporal global efficiency, especially between positive and neutral emotion states. Our findings provide new insight into the emotion-related brain regional coordination evolution and show the potential of dynamic FC for the investigation of the emotion-related brain mechanism.