For the safety and effectiveness of autonomous driving, it is crucial to accurately detect lane lines and road structures. Modern methods based on the Transformer architecture, such as Maptr, have demonstrated powerful capability for online construction of high-definition maps on GPUs. However, for practical vehicular platforms, the deployment and optimization of related models have not caught sufficient attention in practice. To address such issue, this paper proposes an integrated deployment framework, which is tailored specifically for embedded automotive systems and can overcome the limitations of isolated optimization through co-optimizing the model algorithm and the target hardware platform. Furthermore, to validate the effectiveness of the proposed framework and demonstrate its practical performance, we implemented a complete inference visualization on the MDC610 platform. This visualization pipeline can directly project the structured map elements generated by the model, such as detected lane lines, onto the original sensor data.

Recently many learned image compression methods have surpassed traditional techniques in term of rate-distortion performance. However, these methods have yet to effectively eliminate local redundancy, and their complex structures and high computational costs also prevent their wide applications to power-sensitive, resource-limited devices, such as mobile phones. To resolve those issues, we propose a learned image compression approach that leverages a lightweight attention mechanism. Specifically, the proposed Local-Nonlocal Attention Module (LNAM) parallels the processing of local and non-local features, thereby eliminating redundancy more efficiently. Moreover, we further introduce a lightweight channel-wise entropy model to reduce the number of bits required for encoding. Experimental results demonstrate that our method achieves a favorable tradeoff among computational cost, latency, and rate-distortion performance across three public datasets, including Kodak, CLIC, Tecnick.

The deposition process of Fe atoms was simulated using molecular dynamics methods to study the growth modes and morphological changes of Fe thin films under different deposition conditions. The results show that an increase in temperature promotes the aggregation and growth of Fe atoms, resulting in larger and higher islands. Fe atoms exhibit both layer-by-layer and island growth modes on smooth surfaces. While on nano-rough surfaces, the presence of Ehrlich-Schwoebel barrier (ESB) drives Fe atoms to adopt an island growth mode directly. Due to the combined action of the ESB and the energy of the incident atom, the surface roughness of the Fe film on the nano-rough substrate increases first and then decreases as the atom energy increases. The incident angle has different effects on the growth morphology of atoms with varying energies. Specifically, the growth morphology of high energy atoms is significantly influenced by the incident angle, while that of low energy atoms is the opposite. This study provides a theoretical analysis for understanding the growth mechanisms of Fe thin films.

Ostwald ripening is an inevitable process during the preparation of catalysts, leading to an uneven particle size distribution of the catalyst particles. Preventing or controlling Ostwald ripening has always been a challenge in material preparation. This paper conducts molecular dynamics simulations of the deposition and aggregation process of Ni atoms on SiO2 substrates with different surface structures. The results show that by altering the substrate surface structure and increasing the Ehrlich-Schwoebel barrier (ESB), the Ostwald ripening process of Ni atoms can be delayed. At a 45 degrees incident angle, the growth of Ni atomic clusters in FS1 and RS2 shows a positive correlation with time. An optimal growth curve for Ni catalysts is predicted within the energy range of 7.65 x 10-degrees eV to 30 x 10-degrees eV in RS2, where the Ostwald ripening is effectively suppressed during Ni catalyst growth.

Recent advancements in learned image compression have demonstrated significant progress, particularly through the integration of attention modules. While these modules enhance model performance, they also introduce substantial computational complexity, limiting the applicability of these models on resource-constrained devices. To resolve this issue, we propose utilizing knowledge distillation to enable a lightweight model to mimic the behavior of a complex, attention-based large model. Our method reduces the computational complexity while maintaining high performance. Specifically, we train a lightweight student model under the supervision of a pre-trained teacher model with attention modules, ensuring that the student accurately mimics the intermediate features of the teacher. Experimental results demonstrate that the lightweight model trained via knowledge distillation achieves significant performance improvements in image compression tasks compared to the original student model. Our method provides a practical solution for deploying efficient image compression models in resource-constrained environments.

This paper investigates the mean square stability (MSS) of quantized Markov jump linear systems (MJLSs) based on networked control, where the control loop is formed in a digital feedback path. Taking channel capacity constraints into account, we mainly research on the effects of finite constant bits quantization on the MJLSs’ stability. The original system is stabilized by pre-processing coordination transformation of the system matrices and constructing a corresponding strategy of quantization and control. Sufficient and necessary bit rate conditions required to mean square stabilize the system are finally obtained, while these two conditions define the range that the minimum bit rate is located in. The gap between the conditions is narrowed by the method proposed by this paper compared to existed results. We also extend the bit rate conditions to the case that system state cannot be directly accessed through constructing a mode-dependent state observer. Numerical example is used to verify the achieved theoretical results.

This paper investigates the consensus problem of networked multi-agent systems employing event-triggering strategies. Agents establish communication among nodes via channels characterized by Markovian packet loss, where distributed packet losses occur independently on each communication channel. Due to communication bandwidth limitations, we introduce an event-triggering strategy based on consensus error and the state of the agent to design the state-feedback consensus protocol. We investigate the sufficiency conditions that ensure the system reaches mean-square consensus under Markovian packet loss. Furthermore, we do not consider the packet loss on each channel separately but model the overall communication topology of the multi-agent system using only one Markov process. Also, this paper takes into account the inevitable noise within the communication channel, recognizing its potential impact on system performance. Simulation experiments verify the effectiveness of the proposed control strategies.

This paper deals with the finite horizon optimal output feedback control problem for a class of discrete-time Markov jump linear system (MJLS) with multi-step input delays and additive noise. The main difficulties lie in solving a set of coupled difference Riccati equations (CDREs) with time delay terms and overcoming the influence of the invalidity of the separation principle in designing an output feedback controller. Firstly, the explicit expression of the optimal controller is derived from the maximum principle suitable for MJLS. It is shown that the optimal controller is a linear function of the current system state and the system’s historical input. Moreover, when the state is unavailable, a filter is designed by introducing an augmented system. Finally, based on the filter, an optimal output feedback controller is obtained.

We present a simple, robust, and cheap microfabrication method, based on thermally manipulating capillary action in poly(dimethylsiloxane) (PDMS) microholes, for preparing SU-8 curved microstructures. The microstructure morphology including convexity-concavity and curvature can be controlled via tuning the formation temperature. The convex SU-8 microspherical crowns with a height of 40 μm were formed at 10 °C, whereas the concave SU-8 microspherical crowns with a height of 90 μm were formed at 100 °C. The morphology of the microstructures is dictated by the thermally controlled combination of the pressure difference across the interface, contact angle, and surface tension. The fabricated microstructures with a spherical surface can be used as a microlens array or a mold for producing a microlens array. The clear and uniform images were observed using the generated microlens arrays. The equilibrium morphology of the microstructures can be predicted by numerical simulation, which can lessen the number of experiments and thus the design cost. The proposed method has the potential to find applications in industrial fields.

This paper investigates the stabilization of switched linear systems under denial-of-service (DoS) attacks with event-triggered strategies and finite bit rate quantization. Unlike previous research that only considers switching when DoS attacks are inactive, we investigate a general case where unknown switches are allowed during DoS activation, as well as asynchronous communication between controller modes and subsystem modes. We assume that switching instants follow the average dwell time and that DoS attacks with limited energy are described by restricted frequency and duration. Firstly, event-triggered strategies and finite bit rate quantized policy are designed to estimate the bounds of the state estimation error under unknown switches and DoS attacks, and achieve the quantized state feedback under finite bandwidth. Additionally, by using multiple Lyapunov functions, we establish a joint constraint of switching signals and DoS attacks energy, which implicates the influence of multiple switches in DoS attacks and quantization error on system stability. Based on the above discussions, the global asymptotic stability of the closed-loop system is established. Finally, simulations are conducted to confirm the obtained results.

The interfacial thermal resistance between adjacent parts is important in the thermal management of micro/nano-scale systems. In this paper, the temperature difference and heat flow methods are adopted to study the interfacial thermal resistance between the smooth surface of metal and the end of carbon nanotube (CNT) by molecular dynamics simulation. The effects of metal type, CNT diameter, and mean interfacial temperature on the interfacial thermal resistance are studied in detail, with the temperature and heat flow conditions applied to the two metal atom groups of a dumbbell-shaped metal-CNT-metal model. For a certain metal type, the diameter and temperature dependences of the interfacial thermal resistance obtained from both the temperature difference and heat flow methods are consistent, the interfacial thermal resistance decreases with increase of CNT diameter exponentially, and the thermal rectification occurs due to different interfacial temperatures. The thermal transfer mechanism at the metal-CNT interface is quantitatively analyzed by calculating the overlap area of the normalized vibrational density of states. The results of this paper will provide in-depth theoretical insights into the heat transfer enhancement at nano-scale interface.

Driver distraction and fatigue detection systems can effectively reduce car accidents and ensure the safety of traffic participants. Most of the existing vision-based approaches use facial landmarks as driver’s states. However, facial landmark detection is inaccurate under the large angle of head posture, which impacts the accuracy of further processing. This paper presents an effective method using convolution neural networks (CNNs). The method firstly deploys a modified MTCNN to detect the face region. Then, a lightweight multi-task CNN is proposed to detect eye regions, mouth landmarks and 3D head pose, and a simple CNN is used to detect eye closure independently. An angle-adapted loss function is applied to improve the landmark detection accuracy under the large posture. Finally, multiple abnormal behaviors are recognized to determine distraction and fatigue driving. Experiments show that our proposed method is superior to existing methods in both accuracy and running speed.

Myocardial infarction is one of the leading causes of mortality and disability all over the world. To find novel therapy for myocardial infarction, we have identified eudesmanolide from Salvia plebeia and evaluated its protective effects using H9c2 cardiomyocytes induced by hypoxia/reoxygenation. The results showed eudesmanolide improved the cell survival via increasing cell viability and blocking the release of LDH. Further explorations revealed eudesmanolide attenuated mitochondrial dysfunction, oxidative stress and apoptosis resulting from hypoxia/reoxygenation in H9c2 cardiomyocytes. In addition, it was observed eudesmanolide activated Nrf2 in H9c2 cardiomyocytes, which is closely associated with its protective effects. Taken together, these findings give evidences for the discovery of drugs targeting myocardial infarction and the application of eudesmanolide in clinical practice.

We report on curved film microstructure arrays fabricated through polydimethylsiloxane (PDMS) film buckling induced by mechanical stretching. In the process of the microstructure preparation, a PDMA film is glued on a bidirectionally prestretched PDMS sheet that has a square distributed hole array on its surface. After releasing the prestrain, the film microstructure array is created spontaneously. The fabricated microstructures possess a spherical surface and demonstrate very good uniformity. The film microstructure arrays can serve as microlens arrays with a focal length of 1010 μm. The microstructure formation mechanism is investigated via theoretical analysis and numerical simulation. The simulation results agree well with the experimental results. The prestrain applied by mechanical stretching during the fabrication has an important effect on the shape of the resulting film microstructures. The microstructure geometry can be easily tuned through controlling the applied prestrain.

Real-time detection of the degree of corn peeling is important to determine the operational status of the corn harvester. This paper presents a method to detect shucked corn. First, moving objects are detected from the background by using Gaussian Mixture Model (GMM) and morphological operations. Then, texture features, called Local Binary Pattern(LBP) features, are computed from multi-scale foreground images. Finally these texture features are sent to a trained support vector machine, which makes the decision whether a corn is shucked or not. Owing to the post-processing on the foreground image segmented after background modeling, our method can filter out redundant noise points. Due to the prominent difference of the LBP features of different objects, our method can make classification more robustly. Therefore, our method is accurate and efficient in the task of shucked corn detection, which is confirmed by experimental results.

Caching content in routers is the most significant feature of Named Data Networking (NDN) and therefore the cache performance is increasingly being concerned for NDN deployment. The default cache policy of NDN is Leave Copy Everywhere (LCE) that leaves the copies in each node the data packet passed, and most of the copies will not be requested again, which leads to the waste of cache resources. The Betweenness Strategy caches data in the node with the maximal betweenness value, which causes a high replacement rate. Considering the betweenness of nodes and the popularity of content, as well as the filter effect of cache, we propose Betweenness and Edge Popularity strategy (BEP) which caches the most popular content in the most important nodes. We also conduct comprehensive simulations based on ndnSIM. By evaluating BEP and other cache replication strategies on a virtual topology, it is indicated that BEP can achieve higher performance in terms of the cache hit ratio, server load and average delay.

Aiming at the redundancy and inefficiency of most caching strategy in Named Data Networking(NDN) architecture,a caching placement strategy BEP based on node median and edge content popularity is proposed.Combining the median centrality of node with the dynamic popularity of content,and considering the filtering effect of caches,the most popular content is placed on the most important node to make efficient use of scarce cache resources.Simulation results show that compared with the classical NDN caching strategies LCE and LCD,BEP can effectively improve the cache hit rate and reduce the server load.

Existing automated network flows protocol reverse method is difficult to infer message format accurately when dealing with protocols with a large number of binary message data.This paper proposes an improved automated network flows protocol reverse analysis method called PoKE.By adding the position attribute to the keywords,PoKE extracts the short-length keywords from the binary message data,uses keywords to mark the messages,and establishes a state transition model of the protocol according to the marked message sequences.At the same time,PoKE extracts more detailed keywords information through recursive looping mode based on message segmentation and keywords extraction.Experimental results show that PoKE method can extract more keywords information than Biprominer method,thereby establishing a more accurate binary protocol model.

In order to replacing cache efficiently in Named Data Networking(NDN),a cache replacement strategy combining dynamic popularity and cost cache replacement policy is proposed in this paper,which named DPCP.Each node calculates the Dynamic Popularity and Cost(DPC) value of every cache content in its content store.It replaces cache based on the DPC value of content store’s cache content,so as to keep content with high popularity and request cost.Furthermore,it divides contents into different types based on the DPC value and executes Differentiated Decision Policy(DDP) to choose cache node.Experimental results show that the proposed strategy can achieve higher cache hit ratio and reduce average request hop,compared with classical NDN cache algorithm.

This paper proposes a model-based event-triggered control strategy, which is utilized to achieve the input-to-state stability for a scalar continuous time nonlinear system. Due to the processing delay, the sampled feedback signal cannot be transmitted immediately, i.e., the transmission time instant of the feedback signal is not always equal to its sampling time instant. The gap between them measures how much information can be carried. A sufficient bit rate condition to stabilize a scalar continuous time nonlinear system is derived. The condition is only determined by the parameter of Lipschitz condition and the bound of processing delay. Compared with the periodic sampling control strategy, much less bits are used by the event-triggered control one for the particular case of stabilizing a scalar continuous time linear system. Simulations are done to verify the effectiveness of the stabilizing bit rate condition.