To address issues related to reduced neural training accuracy, increased latency, and high energy consumption due to resource limitations of edge servers in the Internet of Vehicles (IoV), this article proposes a joint optimization scheme for edge collaboration and resource allocation in the digital twin-assisted IoV (DT-IoV) with integrated sensing and communication (ISAC). First, a digital twin-assisted environment-aware collaborative mechanism (DT-EACM) is proposed. The DT system makes edge collaboration decisions based on task requirements and available resources, selecting either the ISAC mode or the coordinated environment-aware instruction transmission (InsT) mode. Second, an analysis of deviations in communication, sensing, and computing resources within the collaboration mechanism has been conducted, and a joint optimization problem for edge collaboration and resource allocation is formulated with the goal of minimizing cumulative latency and energy consumption. Finally, since the above problem is a mixed-integer programming problem, it is transformed into a Markov decision process (MDP). A resource scheduling algorithm based on twin-delayed deep deterministic policy gradient (TD3) is proposed, combined with the prioritized experience replay (PER) mechanism to improve algorithm convergence speed and stability, thus minimizing cumulative latency and energy consumption in edge task processing. The simulation results demonstrate that the proposed algorithm effectively reduces cumulative training latency and energy consumption in edge task collaboration compared to other algorithms.

This manuscript introduces a novel RF co-design methodology that allows to co-integrate the functionality of a single-/multi-band bandpass filter (BPF), an RF switch, a variable phase shifter (VP), a variable attenuator (VA), and an RF isolator within a single multi-functional RF component. It is based on a power reconfigurable rat-race coupler (RFRC), spatiotemporally modulated (STM) resonators and a switchable phase coupling element that controls three modes of operation. These include: 1) a co-designed single/multi-band reflectionless BPF, an RF isolator (enabled/disabled by turning ON/OFF STM) and a 360$^{\circ}$-variable phase shifter (BPF/BPFI-VP), 2) an RF switching mode and 3) a single/multi-band BPF, an RF isolator (enabled/disabled by turning ON/OFF STM) and a variable attenuator (BPF/BPFI-VA). A detailed theoretical framework is provided alongside a variety of single-band and multi-band design examples. The methodology is validated through the implementation of a single-band prototype (Prototype 1) and a dual-band prototype (Prototype 2) at 700 MHz. Specifically, Prototype 1 features a single-band BPF/BPFI-VP mode with continuously tunable phase shift of 360$^{\circ}$while having constant transfer function (TF), a wide reflectionless bandwidth (RBW) from 375 to 1093 MHz as well as uni-directional transmission with over 20 dB directivity ($D)$. When reconfigured in its BPF/BPFI-VA mode, a wide attenuation tuning range of 15.8 dB can be obtained with a well-preserved TF, low phase imbalance ($<$5$^{\circ})$and D$>$20 dB. Furthermore, it can be intrinsically switched-off with an isolation$>$30 dB. Prototype 2 supports a third order dual-band TF with 360$^{\circ}$tunable phase shift capability for both of its bands. Furthermore, it exhibits a RBW between 370-1200 MHz, switching-off capability with$>$30 dB of IS and with D$>$10 dB.

Human motion modeling is essential for video-based 3D human pose and shape estimation. Most existing methods model human motion by learning a deterministic mapping from the input videos to the human body parameters, while the uncertainties such as occlusions and depth ambiguities are ignored. To address this problem, we propose a probabilistic model based on conditional normalizing flows called FlowPose to learn the distribution of feasible 3D human motion. This model allows access to the most likely 3D human poses given a video input, which means that more accurate and temporally coherent human poses can be obtained. Additionally, a contrastive training strategy is utilized to maximize the mutual information between video features and their 3D human poses, resulting in an improvement on feature extraction of the conditional flow model. Experimental results on two benchmarks 3DPW and Human3.6M demonstrate that our method outperforms the state-of-the-art video-based methods.

This paper introduces a detailed methodology for the realization of multi-functional transversal resonator-based non-reciprocal bandpass filters (NR-BPFs) that combine the functionality of a single-/multi-band BPF and an RF isolator. Directionality is achieved through spatiotemporal modulation (STM) whereas transfer function reconfigurability is obtained by only tuning the resonant frequency of its constituent resonators. A detailed design methodology to synthesize the response of transversal resonator-based STM arrays is introduced in this work for the first time, facilitating the synthesis of advanced RF filtering transfer functions based of transversal resonator arrays with incorporated directionality. The operating principles and scalability of the design method are demonstrated through the analysis of four distinct transversal resonator-based STM filtering topologies that facilitate the realization of high-order and highly-modular single-band and multi-band transfer functions with multiple levels of RF tuning including frequency tuning, bandwidth tuning, band controllability, and intrinsic switch-off capabilities. The concept has been validated at UHF band through the manufacturing and testing of four lumped-element NR-BPFs.

For visual question answering on remote sensing (RSVQA), current methods scarcely consider geospatial objects typically with large-scale differences and positional sensitive properties. Besides, modeling and reasoning the relationships between entities have rarely been explored, which leads to one-sided and inaccurate answer predictions. In this article, a novel method called spatial hierarchical reasoning network (SHRNet) is proposed, which endows a remote sensing (RS) visual question answering (VQA) system with enhanced visual–spatial reasoning capability. Specifically, a hash-based spatial multiscale visual representation module is first designed to encode multiscale visual features embedded with spatial positional information. Then, spatial hierarchical reasoning is conducted to learn the high-order inner group object relations across multiple scales under the guidance of linguistic cues. Finally, a visual-question (VQ) interaction module is employed to learn an effective image–text joint embedding for the final answer predicting. Experimental results on three public RS VQA datasets confirm the effectiveness and superiority of our model SHRNet.

The multitemporal semantic change detection challenge track (Track MSD) in the 2021 Data Fusion Contest is to extract the land cover changes of the US state of Maryland from 2013 to 2017, but only low-resolution label data is provided. We present a multi-resolution training algorithm (MRTA) to alleviate the overfitting of the model on the coarse labels. First, using low-resolution coarse labels to train FCN, the average IoU can reach 0.5253. Generating pseudo-labels using this network, they are combined with coarse labels to form a multi-resolution label combination, and perform iterative fine-tuning and retrain. After that, by analyzing the loss and gain indicators of specific categories, it was found that the discrimination effect of the water area was poor, so a strong classifier was trained for the water area. We also implemented strategies such as model voting and weighted training to improve model performance. Finally, our method achieves 0.6445 mIoU on the test set, ranking 3rd in Track 2 of the IEEE Data Fusion Contest of 2021.