Traditional vision-based material perception methods often experience substantial performance degradation under visually impaired conditions, thereby motivating the shift toward non-visual multimodal material perception. Despite this, existing approaches frequently perform naive fusion of multimodal inputs, overlooking key challenges such as modality-specific noise, missing modalities common in real-world scenarios, and the dynamically varying importance of each modality depending on the task. These limitations lead to suboptimal performance across several benchmark tasks. In this paper, we propose a robust multimodal fusion framework, TouchFormer. Specifically, we employ a Modality-Adaptive Gating (MAG) mechanism and intra- and inter-modality attention mechanisms to adaptively integrate cross-modal features, enhancing model robustness. Additionally, we introduce a Cross-Instance Embedding Regularization(CER) strategy, which significantly improves classification accuracy in fine-grained subcategory material recognition tasks. Experimental results demonstrate that, compared to existing non-visual methods, the proposed TouchFormer framework achieves classification accuracy improvements of 2.48% and 6.83% on SSMC and USMC tasks, respectively. Furthermore, real-world robotic experiments validate TouchFormer's effectiveness in enabling robots to better perceive and interpret their environment, paving the way for its deployment in safety-critical applications such as emergency response and industrial automation.

Textual open intent classification is crucial for real-world dialogue systems, enabling robust detection of unknown user intents without prior knowledge and contributing to the robustness of the system. While adaptive decision boundary methods have shown great potential by eliminating manual threshold tuning, existing approaches assume isotropic distributions of known classes, restricting boundaries to balls and overlooking distributional variance along different directions. To address this limitation, we propose EliDecide, a novel method that learns ellipsoid decision boundaries with varying scales along different feature directions. First, we employ supervised contrastive learning to obtain a discriminative feature space for known samples. Second, we apply learnable matrices to parameterize ellipsoids as the boundaries of each known class, offering greater flexibility than spherical boundaries defined solely by centers and radii. Third, we optimize the boundaries via a novelly designed dual loss function that balances empirical and open-space risks: expanding boundaries to cover known samples while contracting them against synthesized pseudo-open samples. Our method achieves state-of-the-art performance on multiple text intent benchmarks and further on a question classification dataset. The flexibility of the ellipsoids demonstrates superior open intent detection capability and strong potential for generalization to more text classification tasks in diverse complex open-world scenarios.

Electrocardiogram (ECG) classification plays a pivotal role in computer-aided cardiovascular disease diagnosis. However, current models often underperform in real-world applications due to the scarcity of annotated data and their limited capacity for capturing high-level semantic information. To address these challenges, we propose MTECG, a novel multimodal self-supervised learning framework that bridges raw ECG signals with corresponding clinical text reports. At its core is MACFormer, an ECG encoder based on masked autoencoding and contrastive learning, designed to capture physiologically meaningful features from ECG signals without relying on labels. To enable cross-modal alignment, MTECG integrates MACFormer with pretrained language models using three self-supervised objectives: ECG-to-text reconstruction, text-to-ECG generation, and cross-modal contrastive learning. These objectives encourage the model to learn semantically grounded and generalizable representations without relying on manual labels. Extensive experiments on actual open ECG datasets have shown that MTECG significantly improves classification accuracy. Compared with the existing ECG classification baselines, MTECG achieves state-of-the-art in the macro-averaged F1-score metric on open datasets, and also has good accuracy in the zeroshot scenario, which cannot be achieved by the existing models.

UAV-to-Satellite geolocalization aims to estimate the location of an aerial-view query image taken by an Unmanned Aerial Vehicle (UAV) by matching it to satellite images annotated with known locations. However, it is difficult for existing methods to distinguish neighboring satellite images that exhibit a high degree of visual similarity. To address this issue, we introduce a plug-and-play adaptive confusion-aware contrastive loss (ACCL) to explicitly enhance the model’s discriminative ability, which gives more tolerance to high confusion query samples by means of elaborating a confusion metric function. As a plug-and-play loss module, ACCL can be easily incorporated into various UAV-to-Satellite geolocalization methods without additional modifications. To demonstrate the effectiveness of our proposed method, we conduct extensive experiments on one publicly available geolocalization dataset (i.e. NewYorkFly) and to further prove the effectiveness of our method in different scenarios, we collect two new geolocalization datasets (LasVegasFly and HollywoodFly), which contain drone-captured aerial images and dense sampled satellite images in various geomorphic regions. Experimental results indicate that our method can achieve an obvious performance improvement over the state-of-the-art methods on all three datasets. Our code and collected datasets are available at https://github.com/NWPU-CPS/ACCL.

On-device fixed-point deep neural network (DNN) accelerators are widely used, but the multiply-accumulate (MAC) units that perform the atomic operations in DNNs have become inefficient due to the overestimation of guard bits. To address this issue, we propose an overflow-aware management mechanism. This mechanism promptly adds the partial sum that is about to overflow to the previous partial sum stored in the output buffer and then writes it back into the buffer, thereby reducing the local guard bit overhead in the processing elements (PEs). We implemented and evaluated a PE array based on this mechanism, which reduced area overhead for adders by 43% and for registers by 35%, along with a 14% decrease in energy consumption.

This study addresses the limitations of traditional truck platoon cooperative control methods in optimizing dynamic fuel efficiency. The fixed-order truck platoon has a key flaw: it is unable to dynamically respond to real-time vehicle state changes. To address this, we introduce three innovative methods based on deep reinforcement learning. Compared with fixed-cycle formation transformation strategies, the number of formation transformations is reduced to varying degrees for truck platoons of different sizes. For truck platoons with identical specifications, the impact of different cycle sizes on driving mileage is found to be minimal. This study proves that the dynamic decision mechanism based on deep reinforcement learning can effectively balance formation transformation costs and long-term fuel-saving benefits. The core value lies in establishing an intelligent control paradigm with environmental adaptability. The new algorithm significantly improves fuel economy indicators through real-time state perception and probabilistic decision-making while maintaining formation stability. This method provides a new technical route for energy-saving control in complex transportation scenarios. The core framework can be extended to multi-objective collaborative optimization fields.

Unequal aerodynamic drag across electric vehicles (EVs) in platoons causes imbalanced energy consumption. This reduces overall efficiency and accelerates battery degradation. To address this, we formulate a dynamic reordering task as an Optimal Re-Sequencing (ORS) problem, with the goal of minimizing the final state-of-charge (SOC) variance among vehicles. Departing from conventional fixed-order or heuristic-based strategies, we propose a novel hybrid deep reinforcement learning (DRL) approach that combines NoisyNet-enhanced exploration with the dueling network architecture for value decomposition. Five DRL models, including the proposed Noisy Dueling DQN, are evaluated in a simulation calibrated with real-world highway data. The results show that our method reduces the SOC variance by 34.1% over Dueling DQN, while maintaining low computational cost and fast inference, suitable for deployment in V2X enabled systems. These findings demonstrate the effectiveness and deployability of DRL-based dynamic reordering in enhancing energy-aware EV platoon coordination.

This study addresses the issue of uneven energy consumption in electric vehicle (EV) platoons, arising from the static sequencing of vehicles within the platoon. Such an imbalance can negatively impact the efficiency of individual vehicles and the driving performance of the entire platoon. Our approach proposes dynamically altering the formation of the platoon during transit to balance energy use. The core challenge is to identify the most efficient vehicle sequence at predetermined re-sequencing points during the journey. To address this, we introduce three innovative methods based on deep reinforcement learning, chosen for their ability to handle complex, dynamic optimization problems. Our experimental studies, conducted on actual transportation networks, demonstrate these methods significantly enhance energy management and distribution efficiency in EV platoons, highlighting their potential for practical applications in intelligent transportation systems.

Correlation is a highly computationally intensive and data-intensive signal processing application that is used heavily in radio astronomy for imaging and other measurements. For example, the next generation radio telescope, Square Kilometer Array Low (SKA-L), needs a correlator that calculates up to 22 million cross products, which is a real-time system with continuous input data rates of 6 terabits per second and equivalent computation of 2 Peta-operations per second. Therefore, a flexible and scalable solution with high performance per watt is very urgent and meaningful. In this work, a flexible FX correlation architecture based on FPGA cluster is proposed, which can be fractal in subsystem level, engine level and calculation module level, simplifying the complexity of data distribution network to increase the system's scalability. The interconnect network between processing engines is a new two-stage solution, using self-developed data redistribution hardware to decouple full bandwidth correlation into several independent sub-bands' computation. And the most intensive calculations, cross-multiplications among all the antennas, are modularly designed under MATLAB Simulink and AMD Xilinx System Generator, which are parametrized to scale to arbitrary antenna numbers with optional parallel granularity to minimize development effort on different FPGA or for different applications. What's more, a fully FPGA-based FX correlator for a large array with 202 antennas, consisting of 26 F Engines based on AMD Xilinx Kintex-7 325T FPGAs, 13 X Engines based on AMD Xilinx Kintex ultrascale KU115 FPGAs, has been deployed in 2022, which is the largest full FPGA-based astronomical correlator as we know.

Music-driven 3D dance generation has become an intensive research topic in recent years with great potential for real-world applications. Most existing methods lack the consideration of genre, which results in genre inconsistency in the generated dance movements. In addition, the correlation between the dance genre and the music has not been investigated. To address these issues, we propose a genre-consistent dance generation framework, GTN-Bailando. First, we propose the Genre Token Network (GTN), which infers the genre from music to enhance the genre consistency of long-term dance generation. Second, to improve the generalization capability of the model, the strategy of pre-training and fine-tuning is adopted. Experimental results on the AIST++ dataset show that the proposed dance generation framework outperforms state-of-the-art methods in terms of motion quality and genre consistency1.

The art of communication beyond speech there are gestures. The automatic co-speech gesture generation draws much attention in computer animation. It is a challenging task due to the diversity of gestures and the difficulty of matching the rhythm and semantics of the gesture to the corresponding speech. To address these problems, we present DiffuseStyleGesture, a diffusion model based speech-driven gesture generation approach. It generates high-quality, speech-matched, stylized, and diverse co-speech gestures based on given speeches of arbitrary length. Specifically, we introduce cross-local attention and self-attention to the gesture diffusion pipeline to generate better speech matched and realistic gestures. We then train our model with classifier-free guidance to control the gesture style by interpolation or extrapolation. Additionally, we improve the diversity of generated gestures with different initial gestures and noise. Extensive experiments show that our method outperforms recent approaches on speech-driven gesture generation. Our code, pre-trained models, and demos are available at https://github.com/YoungSeng/DiffuseStyleGesture.

Abstract In recent years, deep convolutional neural networks (CNNs) have achieved a lot of outstanding results in super‐resolution with superior ability. However, the majority of CNNs only use a series of convolution kernels with the same size to extract features. This will cause limited receptive fields. In this work, we propose a parallel convolution attention network (PCAN) to extract features in an effective way. Specifically, a pair of parallel convolutions (PCs) with different kernel sizes is used in one layer in our network, which can extract features within different receptive fields, thereby making full use of the multiscale information. Meanwhile, we apply a channel‐spatial attention (CSA) module in each parallel convolution block to calculate and fuse channel attention and spatial attention. The obtained attention maps emphasize useful features. Experimental results demonstrate the superiority of our PCAN in comparison with the state‐of‐the‐art methods.

We first present a new multi-modular shunt active power filter system suitable for large-capacity compensation. Each module in the system has the same circuit topology, system functionality, and controller design, to achieve coordination control among the modules. The module’s reference signals are obtained by multiplying the total reference signal by the respective distribution coefficient. Next, a novel fault-tolerant approach is proposed based on split-phase control in the a - b - c frame and real-time bus communication. When a phase fault occurs, instead of halting the whole module, the proposed strategy isolates only the faulted bridge arm, and then recalculates the distribution coefficients and transfers the compensation capacity to the same phases of the other normal modules, resulting in a continuous operation of the faulted module and optimization of the remaining usable power devices. Through steady-state analysis of the post-fault circuit, the system stability and control reliability are proven to be high enough to guarantee its engineering application value. Finally, a prototype is established and experimental results show the validity and feasibility of the proposed multi-modular system and its fault-tolerant control strategy.

Parasitic capacitances of silicon carbide (SiC) MOSFET exert an significant influence on the switching performance with direct determination of the switching speed, switching loss and EMI noises, among which the nonlinear gate drain capacitance (Miller capacitance) dominates due to the well-known Miller effect. A precise and comprehensive model of the miller capacitance is proposed according to the structure of SiC DMOSFET at a physical level. Comparing with the traditional “switch model” of SIEMENS, the proposed model is more compact with less parameters while keeps the merit of precision. The detailed parameter acquisition procedure is also given by nonlinear fitting. In addition, the principle of the widely used “switch model” of SIEMENS is explained clearly, which is absent in other literature. The proposed model is verified on a commercial SiC MOSFET and a perfect matching is obtained between the modeled and measured C-V curve.

In this work, we propose two decision techniques that interchange between MIMO and diversity schemes for improving shadowing and alignment problems. In a visible-light communication (VLC) system, the transmission nature can be of two types: (1) all LEDs transmit the same signal stream simultaneously; or (2) each LED transmits different parts of a signal stream independently. The MIMO-diversity technique detects and computes the transmitted signal power during reception and conditionally informs the transmitter to switch between Type (1) and Type (2) transmission. Two experimental models have been tested to show the feasibility of such a technique. In the first model, we constructed a full transceiver and uses a switch IC to switch between MIMO and diversity. The second model uses a microcontroller and software decision to switch between two COM ports, each of them dedicated to MIMO and diversity output respectively. Results suggest that shadowing and alignment problems commonly encountered in visible-light communication systems can be readily solved using these methods. A focusing equation has also been formulated to predict signal intensity more accurately. The focusing gap between the concentrator and the photodiode is taken into consideration during channel computation.