Search

Recently, deep learning-based methods have emerged as the preferred approach for ultrasound data analysis. However, these methods often require large-scale annotated datasets for training deep models, which are not readily available in practical scenarios. Additionally, the presence of speckle noise and other imaging artifacts can introduce numerous hard examples for ultrasound data classification. In this paper, drawing inspiration from self-supervised learning techniques, we present a pre-training method based on mask modeling specifically designed for ultrasound data. Our study investigates three different mask modeling strategies: random masking, vertical masking, and horizontal masking. By employing these strategies, our pre-training approach aims to predict the masked portion of the ultrasound images. Notably, our method does not rely on externally labeled data, allowing us to extract representative features without the need for human annotation. Consequently, we can leverage unlabeled datasets for pre-training. Furthermore, to address the challenges posed by hard samples in ultrasound data, we propose a novel hard sample mining strategy. To evaluate the effectiveness of our proposed method, we conduct experiments on two datasets. The experimental results demonstrate that our approach outperforms other state-of-the-art methods in ultrasound image classification. This indicates the superiority of our pre-training method and its ability to extract discriminative features from ultrasound data, even in the presence of hard examples.

Beyond 5G (B5G) networks rapidly growing to connect billions of Internet of Things (IoT) devices and the dense deployment of IoT devices leads the large-scale network conflict and obstacles the resource-efficient, which brings a great challenge for network resource management (NRM). To tackle this problem, hypergraph based resource-efficient collaborative reinforcement learning (CRL) was proposed for B5G massive IoT. Firstly, the hypergraph theory based network conflict model was formulated to quantify the conflict degree of the B5G massive IoT. Then, since the conflict-free resource management problem is a combinatorial optimization problem with NP-hard, the resource management based Markov decision process (MDP) model was built for NRM in B5G massive IoT. To reduce the computational load by distributing the training overhead throughout the entire B5G massive IoT and achieve distributed collaborative learning, the federated averaging advantage Actor-Critic (FedAvg-A2C) based resource management is proposed to handle the network conflict-free resource management problem and accelerate the training process. Simulation results show the proposed scheme has high network throughput and the resource-efficient in B5G massive IoT.

For subpixel mapping (SPM), ensuring the operational efficiency of the algorithm and mitigating the effect of abundance errors often cannot be achieved simultaneously. To solve the problem, we propose a new SPM method based on the spatial adaptive attraction model (SAAM) and conditional random fields (CRFs). First, the proposed SAAM obtains the spatial adaptive attraction value by adaptively adjusting the spatial attraction value obtained using the traditional spatial attraction model, thereby turning the display form of the abundance constraints in the SPM into an implicit form for expression, to perform the physical significance of the abundance constraints with the relative size of the attraction value of each subpixel. Second, the spatial adaptive attraction value of the implicitly represented abundance constraints and the local spatial smoothing prior are modeled in the CRFs, and the model makes full use of the spatial information in the label field while considering the abundance constraint. Third, Graph-cut is used to optimize the model, the proposed SPM can not only guarantee the operational efficiency, but also extinguish the influence of abundance error and decrease the noise artifact on the results of SPM. Experiments on three remote sensing images show that the proposed SPM accuracy is considerably better than the previously available SPM methods and is the least time-consuming. This study provides a new solution for the SPM of remote-sensing images.

To fully grasp the characteristics and provide a basis for establishing an evaporation duct communication system for the South China Sea (SCS), we constructed a regional statistical database based on remote sensing meteorological parameters and a numerical model. This database has the advantage of extensive spatiotemporal coverage and high resolution, which can provide accessibility and as a supplementary means for the maritime sixth Generation communication network. Specifically, work has been completed in three areas. First, the overall evaporation duct height (EDH) in the central SCS varies periodically and remains intense throughout the day, decreasing only just before sunrise during 04:00–08:00. Moreover, a 20.0 m high channel appears at the “Golden edge” in the northern coastal area, which can expand as far as 300 km from the coast during May–July. Second, the characteristics of specific regions were analyzed, and the highest average EDH was determined as 18.4 m in the Taiwan Strait. The height in the central SCS fluctuates little with monthly and annual variation, and it is maintained between 10.0 and 12.0 m. Third, based on the evaporation duct formation principle, we analyzed the spatiotemporal distributions of meteorological parameters. The Taiwan Strait maintains the highest EDH over the SCS, owing to the influence of the Kuroshio Current, and the average sea surface temperature (SST) reaches 28.41°C in June and 19.58°C in January. To the central SCS, high SST, low wind speed, and relative humidity of approximately 80% provide excellent conditions for maintaining an evaporation duct over the long term.

UAV route planning is the key issue for application of UAV in real-world scenarios. Compared with the traditional route planning methods, although the intelligent optimization algorithm has stronger applicability and optimization performance, it also has the problem of poor convergence accuracy and easy to fall into local optimization. Therefore, an intelligent route planning method for UAV based on chaotic random opposition-based learning and cauchy mutation improved Moth-flame optimization algorithm (OLTC-MFO) is proposed. First, the terrain environment is constructed by digital elevation map, and the threat model is established to realize the equivalent three-dimensional (3D) environment. Then, the opposite population is introduced to increase the diversity of solutions and improve the search speed of the algorithm. Then, the Logistic-Tent chaos map is introduced to realize random perturbation of flame position, which improves the global search capability of the algorithm. Finally, the probability operator and Cauchy mutation operator are introduced, which makes the algorithm not only accept the current solution with a certain probability, but also jump out of the current sub-optimal solution, thus enhancing the global search capability of the algorithm. The simulation results show that when the number of iterations is 1000, the length of route planning based on OLTC-MFO algorithm is 45.3716km shorter than MFO algorithm, and the convergence result of this method is stable and more accurate, which achieves the purpose of assisting UAV combat decision-making.

A database is presented that allows the investigation of machine learning (ML) tools and techniques in the signal integrity (SI), power integrity (PI), and electromagnetic compatibility (EMC) domains. The database contains different types of printed circuit board (PCB)-based interconnects and corresponding frequency domain data from a physics-based (PB) tool and represent multiple electromagnetic (EM) aspects to SI and PI optimization. The interconnects have been used in the past by the authors to investigate ML techniques in SI and PI. However, many more tools and techniques can be developed and applied to these structures. The setup of the database, its data sets, and examples on how to apply ML techniques to the data will be discussed in detail. Overall 78961 variations of interconnects are presented. By making this database available we invite other researchers to apply and customize their ML techniques using our results. This provides the possibility to accelerate ML research in EMC engineering without the need to generate expensive data.

Different from ship detection from synthetic aperture radar (SDSAR) and ship detection from spaceborne optical images (SDSOI), ship detection from visual image (SDVI) has better detection accuracy and real-time performance, which can be widely used in port management, cross-border ship detection, autonomous ship, safe navigation, and other real-time applications. In this paper, we proposed a new SDVI algorithm, named enhanced YOLO v3 tiny network for real-time ship detection. The algorithm can be used in video surveillance to realize the accurate classification and positioning of six types of ships (including ore carrier, bulk cargo carrier, general cargo ship, container ship, fishing boat, and passenger ship) in real-time. Based on the original YOLO v3 tiny network, we have made the following fine tunings. 1) The preset anchors trained on Seaship annotation data have the similar “dumpy” shape as the normal ships, helping the network to achieve faster and better training; 2) Convolution layer instead of max-pooling layer and expanding the channels of prediction network improve the small target detection ability of the algorithm. 3) Due to the problem that large-scale ships are easily disturbed by the onshore building, complex waves and light on the water surface, we introduced attention module named CBAM into the backbone network, which make the model more focused on the target. The detection accuracy of the proposed algorism is obviously better than that of the original YOLO v3 tiny work. Although it is slightly inferior to the Yolo v3 network, it has faster speed than Yolo v3. However, the proposed algorithm is a better trade-off between real-time performance and detection accuracy, and is more suitable for actual scenes. Compared with the SOAT algorithm in Z. Shao et al. (2020), our algorithm has a 9.6% improvement in mAP and a faster speed.

Differential Evolution (DE) is a simple and effective stochastic algorithm for optimization problems, and it became much more popular in recent year because of its easy-implementation and excellent performance. Nevertheless, the performance of DE algorithm is greatly affected by the trial vector generation strategy including mutation strategy and parameter control, both of which still exist some weaknesses, e.g. the premature convergence to some local optima of a mutation strategy and the misleading interaction among control parameters. Therefore in this paper, a novel Di-DE algorithm is proposed to tackle these weaknesses. A depth information based external archive was advanced in our novel mutation strategy, which can get a better perception of the landscape of objective in an optimization. Moreover, a novel grouping strategy was also employed in Di-DE and parameters were updated separately so as to avoid the misleading among parameters. Moreover, a cooperative strategy for information interchange was also advanced aiming at improving the efficiency of the exploration behavior. By absorbing these advancements, the novel Di-DE algorithm can secure better performance in comparison with other famous optimization algorithms. The algorithm validation was conducted on CEC2013 and CEC2017 test suites, and the results revealed the competitiveness of our Di-DE algorithm in comparison with those famous optimization algorithms including Particle Swarm Optimization (PSO) variants, QUasi-Affine TRansformation Evolution (QUATRE) variants and Differential Evolution (DE) variants.

Multiple SVR based on ensemble learning could be enhanced from the viewpoint of the performance, but the performance of modeling closely depends on the initial condition of the partitioning method and they are easily affected by noise and outliers. In this study, a multi-linear fuzzy support vector regression (MFSVR) robust to noise is proposed with the aid of the composite kernel function and $\varepsilon $ -fuzzy c-means (FCM) clustering based on insensitive data information. Here insensitive data information stands for the interval data information of “$\varepsilon $ ” which stands for insensitive loss parameter used in the $\varepsilon $ - insensitive loss function. The objective of this study is to reduce the effect of noise and to alleviate the overfitting problem through the synergistic effect of the following methods: First, $\varepsilon $ -FCM clustering based on insensitive data information is used for considering more impact on decision boundary and reducing the effect of noise. Second, the composite kernel based on multiple linear kernel expression is proposed for implementing multi-linear decision boundary to alleviate overfitting problem. In more detail, each training data point is assigned with corresponding membership degrees in the $\varepsilon $ -FCM clustering. Some data which are potentially to be noise or outlier are assigned with lower membership degrees and given small contribution (compensation) considered in composite kernel function. Then, the composite kernel function for multiple local SVRs is constructed according to the distribution characteristics of $\varepsilon $ -FCM clustering. The proposed MFSVR is tested with both synthetic and UCI data sets in order to verify the effectiveness as well as efficient performance improvement. Experimental results demonstrate that the proposed method shows the better performance when compared to other some methods studied so far.

The recommendation systems based on deep neural networks show high accuracy and precision in advertising prediction and rating prediction tasks, but the performance declines when they encounter sparsity problems. Furthermore, the low scalability framework makes them expand new users or items inefficiently. We propose a Rapid Scalable Embedding and Mapping model for Recommendation (RSEMR) to solve the problems. RSEMR uses the bi-directional mapping structure to build the transformation between the embedding vector and the unified vector space, which can rapidly expand new users or items into the existing domains and achieve a better initial performance of prediction. Two effective mapping structures, including vector elements-based single connection and autoencoder-based fully connection, can be used in RSEMR to capture linear or non-linear mapping functions, respectively. Besides, RSEMR does not require any additional information other than a very small amount of user behavior data, and it deals with the relationships between subjects based on the spatial distance in the unified vector space, which is more adaptable to the scarce attribute labels and sparse interaction behaviors. We select several open datasets for experiments and the results have demonstrated RSEMR's superiority over the mainstream neural network-based recommendation models in sparsity situation, in terms of accuracy, efficiency, and scalability.

The crosstalk suppression of multichannel coupled drivelines is studied for high-performance integration of micro-electro-mechanical system arrays. By using traditional drivelines, for example, microstrip lines, the distance between adjacent drivelines should be large enough to reduce the multichannel coupling effects. As a result, the size and the area of ultra-low crosstalk and closed-packaged driveline become an issue for design. To address it, we proposed a new scheme of multichannel spoof surface plasmon polaritons (SSPP)-based drivelines. The new drivelines consist of multiple parallel planar corrugated metallic strips. Thanks to the strong field confinement and localization characteristics of the SSPPs, the coupling effects between drivelines in different channels can be significantly suppressed with no need of any other coupling suppression circuits. For demonstration and investigation, both simulation and measurement are carried out up to 20 GHz. The results show that the crosstalk by using the proposed new drivelines can be reduced by about 10 dB on average. Therefore, the SSPP-based multichannel drivelines would be a good candidate for large-scale advanced integrated devices and systems with requirements of both miniaturization and coupling suppression.

Automatic modulation recognition (AMR) is considered one of most important techniques in the non-cooperative wireless communication systems. Traditional algorithms, e.g., support vector machine (SVM) based on high order cumulants (HOC), are hard to achieve the reliable performance. In this paper, we propose an effective AMR algorithm based on deep learning (DL) with capabilities of automatically extracting representative and effective features. Our proposed method resorts to in-phase and quadrature (IQ) samples which are IQ components of received baseband signal, respectively. We adopt convolutional neural networks (CNN) and recurrent neural networks (RNN) to classify six types of signal modulations over additive white Gaussian noise (AWGN) channel and Rayleigh fading channel, respectively. Simulation results show that DL-AMR is much better than traditional algorithms under two fading channels.

Stroke is a worldwide healthcare problem, which often causes long-term motor impairment, handicap, and disability. Optical motion analysis systems are commonly used for impairment assessment due to high accuracy. However, the requirement of equipment-heavy and large laboratory space together with operational expertise makes these systems impractical for local clinic and home use. We propose an alternative, cost-effective and portable, decision support system for optical motion analysis, using a single camera. The system relies on detecting and tracking markers attached to subject's joints, data analytics for calculating relevant rehabilitation parameters, visualization, and robust classification based on graph-based signal processing. Experimental results show that the proposed decision support system has the potential to offer stroke survivors and clinicians an alternative, affordable, accurate, and convenient impairment assessment option suitable for home healthcare and telerehabilitation.

Extensive research works demonstrate that the attention mechanism in convolutional neural networks (CNNs) effectively improves accuracy. Nevertheless, few works design attention mechanisms using large receptive fields. In this work, we propose a novel attention method named Rega-Net to increase CNN accuracy by enlarging the receptive field. To the best of our knowledge, increasing the receptive field of the CNN requires increasing the size of the convolution kernel, which also increases the number of parameters. For solving this problem, we design convolutional kernels to resemble the nonuniformly distributed structure inspired by the mechanism of the human retina. Then, we sample variable-resolution values in the Gabor function distribution and fill these values in retina-like kernels. This distribution allows essential features to be more visible in the center position of the receptive field. We further design an attention module including these retina-like kernels. Experiments demonstrate that our Rega-Net achieves 79.96% Top-1 accuracy on ImageNet-1k for classification and 43.1% mAP on COCO2017 for object detection. The mAP of the Rega-Net increased by up to 3.5% compared to baseline networks. [ABSTRACT FROM AUTHOR]