Search

Deep learning models face significant challenges in image classification due to the limited availability of training samples. To address this issue, few-shot learning, which enables model training with a small number of samples, has emerged. When applied to classification tasks, it is referred to as few-shot classification. Prototype networks are an effective approach for few-shot classification, however, the prototypes obtained by averaging sparse samples often lack sufficient representativeness, leading to reduced classification accuracy. To tackle these challenges, this paper proposes a novel prototype fusion completion network (ProfcNet). The proposed method enhances prototype representation and improves class representativeness using only the data of the prototype itself. Specifically, the network utilizes a fusion module to generate fusion feature maps with more class-specific characteristics, and a completion module to refine and complete the prototype using these fusion feature maps, yielding the final prototype. Additionally, an inference testing method tailored for the prototype fusion completion network is introduced. Extensive experiments on four publicly available few-shot classification datasets demonstrate that the proposed method significantly improves classification accuracy compared to prototype networks, achieving accuracy gains of nearly 30% and 20% in 1-shot and 5-shot classification tasks, respectively. Furthermore, ablation studies and visualization experiments further validate the effectiveness of the individual modules within the network.

Unmanned aerial vehicle (UAV) relay can effectively improve the coverage and performance of land communications, which makes it prospective to be deployed in various applications. However, there exist some practical challenges we have to face in UAV-enabled communication system designing, such as secrecy and wiretapping threats due to the air-to-ground line-of-sight communication link, and energy consumption issues because of the limitation in energy-constraint. To deal with these two key issues, we develop a novel UAV swarm assisted multi-hop mobile relay system which can be applied in special scenarios with severe blockage or long distance to enhance communication performance. An effective cooperative transmission scheme is proposed to address the security issues in a physical layer security perspective, where several UAVs serve as multi-hop relays forwarding information between ground users and the other UAVs are employed as friendly jammers confusing the ground eavesdroppers. The minimum secrecy energy efficiency (MSEE), namely the minimum achievable secrecy rate per energy consumption unit over legitimate links, is designed as the performance indicator to quantify the secure and energy-efficient transmission. Subject to information-causality, transmit power, and mobility constraints, the UAV swarm assisted MSEE maximization transmission scheme is formulated into a complicated non-convex problem. To solve the non-convex fractional optimization problem, we decouple the original problem into sub-problems and further propose an efficient algorithm by applying the block coordinate descent method, successive convex approximation (SCA) techniques, and Dinkelbach method. Numerical results demonstrate that the proposed scheme achieves considerable performance enhancement on MSEE compared with benchmark schemes.

Due to the air-to-ground line-of-sight communication links, it is challenging to deal with the security threats in the wireless system with unmanned aerial vehicles (UAVs) integrated. In a UAV-enabled mobile relay system, the paper proposes a novel cooperative communication scheme from a physical layer security perspective to address the security issue. Specifically, in a mobile relay system where a UAV relay is employed to forward confidential information between two ground users, an eavesdropper exists on the ground wiretapping the relay UAV. To enhance the security performance of the system, we introduce a friendly UAV jammer transmitting interference signals and confusing the eavesdropper. A secrecy rate maximization problem is then formulated, subject to mobility constraints, power constraints on both UAVs and the information-causality constraint on the relay UAV. To solve the non-convex optimization problem with closely coupled variables, we decouple the problem into more tractable subproblems. An iterative algorithm is thus proposed to optimize the transmit power and flight trajectory of both UAVs alternately via the update-rate-assisted block coordinate descent and successive convex approximation techniques. Simulation results demonstrate that proposed cooperative design can significantly improve the secrecy rate of the UAV-enabled mobile relay system compared to benchmark schemes.

Unmanned aerial vehicle (UAV) data collection is a promising research direction that can be applied to many practical scenarios. Due to the limited energy of sensors in wireless sensor networks (WSN), UAV, which is considered as a mobile fusion center, can effectively prolong the lifetime of sensor via supporting communication with the sensor directly. However, since the UAV's energy constrained, it is necessary that multiple UAVs provide data collection to sensors in large areas. In this paper, we consider a scenario where multiple UAVs collect data from a set of two-dimensional distributed sensors. The UAV can communicate with sensors while flying or hovering. The goal is to minimize the total time that all UAVs from data center and return to data center after completing all collection tasks, while giving each sensor a certain amount of data and energy. The problem of minimizing the task completion time of multiple UAVs is still a big challenge. We solve the multi-UAV problem by jointly optimizing UAV-sensor association mechanism and data collection method of UAVs. The numerical results show that the proposed multi-UAV data collection scheme shortens the task completion time.

Urbanization has been speeding up social and economic transformations in urban communities, the smallest social units in a city. However, urbanization brings challenges to urban management and security. Therefore, a system of risk prediction of crimes may be essential to crime prevention and control in urban communities and its system improvement. To tackle crime-related problems in urban communities, this paper proposes a model of daily crime prediction by combining Long Short-Term Memory Network (LSTM) and Spatial-Temporal Graph Convolutional Network (ST-GCN) to automatically and effectively detect the high-risk areas in a city. Topological maps of urban communities carry the dataset in the model, which mainly includes two modules — spatial-temporal features extraction module and temporal feature extraction module — to extract the factors of theft crimes collectively. We have performed the experimental evaluation of the existing crime data from Chicago, America. The results show that the integrated model demonstrates positive performance in predicting the number of crimes within the sliding time range.

Lamb wave modes with type I dispersion characteristics exhibit strong affinity toward multi transverse modes behavior above resonance frequency (fr) in the AlN Lamb wave resonators (LWRs), especially the high-transduction-efficiency modes: S0 and S1 mode. For conventional interdigital transducer (IDT) design, the IDT aperture and IDT gap are the two main factors impacting the transverse mode placements and strengths, according to the wave vector analysis and finite element method (FEM) simulation. Moreover, the convex slowness curve of the Lamb wave modes propagating in AlN platelets allows the waveguiding and weak lateral leakage into busbars by the high-velocity IDT gap region. Apodization, the standard technique to suppress the transverse modes for IDT-excited resonators, suffers from drawbacks, such as additional loss and reduction of the effective coupling coefficient (keff2). Type I Lamb wave modes in AlN show positive slope in the dispersion branch so that a border region of lower Eigen-resonance frequency is required to form piston mode structure for transverse spurious mode suppression and lateral leakage reduction. Based on dispersion calculations and 2.5D FEM simulations, we demonstrate that by designing the low-velocity border region, such as simply changing the IDT layout, the guiding can be improved and a piston mode can be obtained for the type I Lamb wave modes.

A unique type of refractive index (RI) sensor is proposed using a dual Mach-Zehnder interferometer (MZI) structure based on side-hole fibers (SHFs). The MZI structure contains two single-mode fibers (SMFs), two coreless fiber (CLF) sections and an SHF section, which are spliced together in the order of SMF-CLF-SHF-CLF-SMF. The SMFs and the CLFs enable light lead in/out and beam splitting/combining, respectively. As a special feature of the structure, one hole of SHF is exposed for liquid filling and to form two MZIs as well. Three types of sensors are fabricated, namely S1, S2 and S3. Numerical simulation and experimental studies have been conducted to characterize the sensing performance. The RI sensitivity of the S1 using the ~550 μm long SHF section reaches 14,000 nm/RIU. When a shorter SHF section is used in S2, the detectable RI range is broadened due to larger FSR. When the closed hole of SHF in S3 is filled with liquid to introduce the Vernier effect, the sensitivity can be further enhanced to over 44,000 nm/RIU (i.e., Refractive Index Unit), which corresponds to the detection limit at the level of 1.0 × 10-5 RIU. This sensor design is original and easy to package, which gives it potential for label-free biochemical analyses.

Point of interest (POI) recommendation is a significant task in location-based social networks (LBSNs) as it can help to suggest new locations and makes LBSNs more prevalent to users. Successive POI recommendation is a nature extension of the general POI recommendation which utilizes users' current state like the latest check-in records or timestamps to recommend subsequent POIs. Successive POI recommendation requires a well-constructed model for the transition patterns in POI sequences; however, existing works still have some limitations: 1) transitions are modeled on a relatively low level which cannot reflect users' real intentions hidden behind; 2) there lacks a balance between the transition patterns modeled globally and personally; and 3) most works only consider the correlations between adjacent check-ins, but longer dependencies should be captured as well. To resolve the above issues, we present a successive POI recommendation approach called TTR which is based on the personalized transition pattern analysis on the cluster level for the check-in data in LBSN. It first clusters the POIs based on their representation vectors learnt from Word2Vec model, then it models users' transition behavior on the cluster level through the additive Markov chain model, finally it recommends successive POIs based on a combination of personalized and global strategy. We conduct several experiments on the real datasets Gowalla and Brightkite to evaluate the performance of TTR, and the results show that the proposed method outperforms existing works in terms of precision and recall metrics, and the personalized strategy shows better performance while the global strategy can provide better diversity.