Electrocardiogram (ECG) signal classification is an important task in healthcare as it plays a vital role in early prevention and diagnosis of cardiovascular diseases. In this work, we propose an attention-based hybrid ResNet-Transformer network (ART-Net) for 12-lead ECG signal classification. It is comprised of a stacked multi-scale attention-based ResNet and self-attention-based Transformer. At first, ECG signals are divided into several signal segments with the same length. Then multi-scale features are extracted by attention-based Resnet through signal segments, and attention mechanisms are used to adjust the weight of different channel features based on their importance. Next, these multi-scale features from a same ECG signal are integrated in chronological order as input to the Transformer network. In this end, extracting and fusing contextual information based on self-attention mechanism, and extracting the correlation between beats at different positions. The experimental results on CPSC2018 indicate that our model outperforms three state-of-the-art methods, and achieve 85.27% of accuracy, 86.01% of sensitivity and 85.59% of specificity, respectively.

Freeway bottlenecks caused by traffic incidents contribute significantly to large-scale traffic congestion. Traditional strategies, including variable speed limit (VSL) and ramp metering, are commonly used for freeway traffic congestion management. Recently, vehicle platooning has become a promising way to alleviate traffic bottlenecks. This work proposes a novel framework that combines VSL and vehicle platooning for freeway bottleneck, referred to as VSL-VP, in mixed traffic of connected and autonomous vehicles (CAVs) and human-driven vehicles (HDVs). First, the upstream road of a bottleneck is divided into two segments, called the former and the latter. VSL limits vehicle speed at the former segment, thereby reducing inflow traffic to the latter one. Then, deep reinforcement learning is employed for CAV platooning at the latter segment, where low traffic flow density and large car-following distance create conditions for smooth lane change and platoon formulation of CAVs. Simulation results demonstrate that VSL-VP significantly enhances the bottleneck throughput and reduces traffic congestion at elevated levels of CAV penetration rates.

Trip recommendation aims to provide users with a sequence of points of interest (POIs) according to their interests and requirements when exploring unfamiliar cities. In contrast to prior research on trip recommendation, our research deals with such a problem: If a user is scheduled to attend an academic conference at 2:30 PM, how might he/she make a visit to the city's attractions while still managing to attend the conference? To address this problem, a trip recommendation method based on mixed graph representation learning is proposed. Firstly, a mixed graph is used to describe the spatial temporal, and transition knowledge in users' check-in data. Then, we employ the graph convolutional network to integrate knowledge matrices extracted from the mixed graph. Finally, a trip inference module, which incorporates a dual decoder, POI popularity knowledge, and positional encoding, is designed to generate a trip for a given query. Experiments are conducted on five real-world trip datasets. The results demonstrate that the proposed method outperforms several widely-used baselines when recommending a trip with an FP.

Point of interest (POI) recommendation is a hot research topic. Current researches mainly focus on the analysis of personal check-in trajectories to obtain user preferences. However, a user's check-in data is generally sparse, and it is difficult to make accurate recommendation by only using the user's local information. Additionally, the public's check-in behavior may exhibit common visiting patterns, and incorporating global check-in information is beneficial for enhancing the learning of individual user preferences. Therefore, we propose a GCN-based model with Fusion of Global and Local information (GFGL) for the next POI recommendation. The model obtains global information such as spatial distance, social relationships., and transition probabilities from all users' visit trajectories, and utilizes graph convolution network (GCN) for learning of multi-dimensional global information. Next, we fuse global information with user local information through the user context information embedding module. Besides., a long short-term memory (LSTM) model and transformer model are used to learn the relationship between the user's sequential preference and non-adjacent visits in trajectories. Extensive experiments on two real-world datasets demonstrate the superiority of GFGL against state-of-the-art methods in the next POI recommendation.

In location-based services (LBS), trip recommendation accuracy is challenged by diverse user preferences and complex transfer behaviors. Previous studies overlook the reverse effect of following POIs on previously visited ones. To address this, we propose a Graph-based Double-layer Bidirectional Trip Recommendation (GDB-TR) model. This model uses a heterogeneous graph to model user check-in trajectories with spatial and temporal information. Subgraphs are extracted from the heterogeneous graph, and an adjacency matrix is built for each subgraph. These matrices are fused through a neural network to obtain vector representations for POIs and POI categories. GDB-TR's core is a double-layer bidirectional neural network: one layer describes POIs, the other POI categories. Bidirectional computation captures the influence of preceding POIs on following ones and vice versa. Experiments on five real-world datasets demonstrate GDB-TR's superiority over baseline models, measured by${\boldsymbol{F}}_{\boldsymbol{{1}}}$and pairs-${\boldsymbol{F}}_{\boldsymbol{{1}}}$metrics.

With the fast progress in autonomous driving and communication technologies, robotaxis emerge as a novel mode of transport. Optimization of robotaxi dispatch with ride-sharing can enrich the travel choices of residents and improve the network capacity of transportation systems. This work proposes a multi-objective optimization model for robotaxi dispatch. Unlike previous approaches, this is the first attempt to adjust simultaneously unreasonable boarding time (BT) and pick-up/drop-off (UO) points for passengers during the dispatch. It encourages passengers to walk to the recommended UO points or to be picked up slightly earlier or later, which aims to maximize the profit per kilometer of robotaxis and to minimize the total travel expense of passengers. Consequently, a nondominated sorting genetic algorithm with mass center (NSGA-MC) is proposed to address the model. Experimental results show that the proposed algorithm outperforms its peers from multiple metrics, which highlight the advantage in advancing intelligent public transportation systems.

The development of Connected and Automated Trucks (CATs) provides a new opportunity for freight industry to enhance fuel efficiency, increase traffic flow, and improve safety through platooning. Particularly at highway on-ramps, how to effectively form CAT platoons is a key research topic. In the process of CAT platooning, the timing, location, and speed of CAT merging significantly impact safety and energy consumption. Thus, this study proposes a hierarchical merging strategy, aimed at achieving effective autonomous CAT platooning at highway on-ramps by considering the interference of human-driven vehicles (HDVs). Specifically, we employ a model-free deep reinforcement learning method that guides CAT merging process by exploring optimal driving behaviors. It ensures the safety and efficiency of the CAT merging process. In addition, we use the real vehicle dynamics model in simulation. The proposed strategy can handle the variation of the CATs' initial positions and speeds at on-ramps, as well as interference caused by HDVs at highway mainline. The effectiveness of the proposed strategy has been validated through simulations. The results show that the proposed strategy can effectively coordinate CAT platooning at highway on-ramps.

Two-Wheelers (TWs) such as bikes, e-bikes, and motorcycles often occupy lanes illegally and exceed speed limits, which leads to many traffic accidents. Therefore, we use deep reinforcement learning to design driving strategies for Autonomous Vehicles (AVs) to avoid collision with TWs and reduce injury of TW riders with irregular riding behaviors at un-signalized occluded intersections. First, the collision-avoidance behaviors of TWs are modeled, respectively. The state spaces integrate a safe avoidance range of AVs, a new position of AVs after taking deceleration and a steering angle, a predicted acceleration, and position, speed, and steering angle of AVs and other vehicles. At the same time, a reward function is designed based on the injury of TW riders and the driving safety and comfort of AVs. Secondly, a reinforcement learning model for autonomous driving strategy is constructed. Finally, Soft Actor-Critic is used to train the model, and the randomness policy is used to help AVs flexibly deal with the uncertain behaviors of TW riders and realize the balance between exploring unknown behaviors and using existing information. The simulation results show that compared with an autonomous emergency braking system, the injury of the riders using the driving strategy is reduced by 18.02% on average; compared with a risk-aware high-level decision strategy, the injury is reduced by 41.24% on average.

Abstract For semantic segmentation of cardiac magnetic resonance image (MRI) with low recognition and high background noise, a fusion‐attention Swin Transformer is proposed based on cognitive science and deep learning methods. It has a U‐shaped symmetric encoding–decoding structure with an attention‐based skip connection. The encoder realizes self‐attention for deep feature representation and the decoder up‐samples global features to the corresponding input resolution for pixel‐level segmentation. By introducing a skip connection between the encoder and decoder based on fusion attention, the remote interaction of global information is realized, and the attention to local features and specific channels is enhanced. A public ACDC cardiac MRI image dataset is used for experiments. The segmentation of the left ventricle, right ventricle, and myocardial layer is realized. The method performs well on a small sample dataset, for example, the pixel accuracy obtained by the proposed model is 93.68%, the Dice coefficient is 92.28%, and HD coefficient is 11.18. Compared with the state‐of‐the‐art models, the segmentation precision has been significantly improved, especially for the low recognition and heavily occluded targets.

Abstract In the medical image analysis domain, medical image segmentation has a significant impact on the quantitative analysis of organ or tissue function, as the first and critical component of diagnosis and treatment pipeline. In this paper, a dense R‐CNN segmentation model based on dual‐attention are proposed for medical images multi‐target instance segmentation. The model combines channel and spatial attention mechanism to extract image features and fuse multi‐scale feature information hierarchically. It combines up‐sampling strategies such as dilated convolution and bilinear interpolation to strengthen the distinguishability between multi‐target instances and pixel‐level features in other regions. The multi‐target detection mechanism of R‐CNN is combined with the multi‐scale feature extraction and fusion ability of dense convolution network. In the encoding stage, the multi‐scale hybrid bottleneck module and deformable convolution are introduced to extract more accurate structural feature information and increase the receptive‐field. In the decoding stage, the bilinear interpolation and the adaptive hierarchical fusion mechanism are used to strengthen the distinguishability between the target region and other regions, and improve the accuracy of instance segmentation. Taking cardiac MRI segmentation as an example, the left and right ventricles, and left ventricular myocardium are selected as segmentation targets. The pixel accuracy is 90.82%, the class pixel accuracy is 87.91%, the mean intersection‐over‐union is 81.52%, the Dice coefficient is 89.82%, and Hausdorff distance is 9.2, which is improved compared with other methods. It verifies the accuracy and applicability of the proposed method for multi‐target instance segmentation of medical images.