GCN-based methods have achieved remarkable performance in skeleton-based action recognition. However, existing not explicitly attempted to remove temporal and spatial redundancy that might introduce additional computational costs. Inspired by the fact humans always tend glimpse at overall motion then zoom into most important spatio-temporal regions, we propose a Spatio Temporal Focused Dynamic Network (STFD-Net) trained with reinforcement learning for Specifically, first global extractor...

In recent years, numerous neuroscientific studies have shown that human emotions are closely linked to specific brain regions, with these regions exhibiting variability across individuals and emotional states. To fully leverage neural patterns, we propose an Adaptive Progressive Attention Graph Neural Network (APAGNN), which dynamically captures the spatial relationships among during processing. The APAGNN employs three specialized experts progressively analyze topology. first expert global...

As a type of multi-dimensional sequential data, the spatial and temporal dependencies electroencephalogram (EEG) signals should be further investigated. Thus, in this paper, we propose novel spatial-temporal progressive attention model (STPAM) to improve EEG classification rapid serial visual presentation (RSVP) tasks. STPAM first adopts three distinct experts learn topological information brain regions progressively, which is used minimize interference irrelevant regions. Concretely, former...

The increasing illegal parking has become more and serious. Nowadays the methods of detecting illegally parked vehicles are based on background segmentation. However, this method is weakly robust sensitive to environment. Benefitting from deep learning, paper proposes a novel vehicle detection system. Illegal captured by camera firstly located classified famous Single Shot MultiBox Detector (SSD) algorithm. To improve performance, we propose optimize SSD adjusting aspect ratio default box...

Fast and accurate vehicle detection in unmanned aerial (UAV) imagery is a meaningful but challenging task, playing an important role wide range of applications. Due to its tiny size, few features, variable scales imbalance sample problems UAV imagery, current deep learning methods used this task cannot achieve satisfactory performance both accuracy speed, which obvious classical trade-off problem. In paper, we propose single-shot detector, focuses on real-time imagery. We make contributions...

Graph Convolutional Networks (GCNs), which model skeleton data as graphs, have obtained remarkable performance for skeleton-based action recognition. Particularly, the temporal dynamic of sequence conveys significant information in recognition task. For modeling, GCN-based methods only stack multi-layer 1D local convolutions to extract relations between adjacent time steps. With repeat a lot convolutions, key with non-adjacent distance may be ignored due dilution. Therefore, these still...

Video Compressive Sensing (VCS) works to recover the scene video from limited compressed measurements. VCS was intended sense and in spatial-temporal sensing manner. It is difficult be performed due complexity of design optimization. The most current approaches measure only spatial or temporal domain. However, this would lose correlation VCS. Focus on issue, paper proposes a framework, which uses learned manner hybrid-3D recovery network. In terms technical study, we develop residual block...

Skeleton-based action recognition has attracted great interest due to low cost of skeleton data acquisition and high robustness external conditions. A challenging problem skeleton-based is the large intra-class gap caused by various viewpoints data, which makes modeling difficult for network. To alleviate this problem, a feasible solution utilize label supervised methods learn view-normalization model. However, since in real scenes acquired from diverse viewpoints, it obtain corresponding...

Graph convolutional network (GCN)-based methods have obtained remarkable performance and gained widespread attention for skeleton-based human action recognition. These typically apply 1-D local convolutions to model temporal correlations simply utilize multilayer stacking capture long-range dynamics. However, the convolution focuses on relations between adjacent time steps. Also, with repeat of a lot convolutions, key relation nonadjacent distance may be ignored due information dilution....

Vertical Ground Reaction Force (VGRF) signal obtained from foot-worn sensors, also known as plantar data, provides a highly informative and detailed representation of an individual's gait features. Existing methods, such CNNs, LSTMs Transformers, have revealed the efficiency deep learning in Parkinson's Disease (PD) diagnosis using VGRF signal. However, intrinsic topologic graph pressure transmission characteristics data are overlooked those approaches, which essential features for analysis....

Skeleton-based action recognition has attracted great interest in computer vision. For this task, a challenging problem concerns the large intraclass variances of skeleton data, which are mainly caused by diverse viewpoints and subjects, greatly increase difficulty modeling actions through network. To address above problem, we propose variance reduction (VaRe) framework for skeleton-based recognition, consists view-normalization generative adversarial network (VN-GAN), subject-independent...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Skeleton-based Sign Language Recognition (SLR) is a challenging research area mainly due to the fast and complex hand movement. Currently, Graph Convolution Networks (GCNs) have been employed in skeleton-based SLR achieved remarkable performance. However, existing GCN-based methods suffer from lack of explicit attention topology which plays an important role sign language representation. To address this issue, we propose novel Hand-Aware Network (HA-GCN) focus on topological relationships...

Ultrasonic echo methods have been widely researched for the application of flaw detection, where locations are identified by arrival time each echo. The main difficulty is that receiving echoes from consecutive flaws overlap in when close. Over last decades, sparse approximation and neural-network-based used to address this issue. However, these cannot achieve satisfactory performance high-noise severe overlapping scenarios. In paper, we propose a high-resolution ultrasonic detection method...

Conventional compressive sensing (CS) reconstruction is very slow for its characteristic of solving an optimization problem. Convolu- tional neural network can realize fast processing while achieving compa- rable results. While CS image recovery with high quality not only de- pends on good algorithms, but also measurements. In this paper, we propose adaptive measurement in which obtained by learning. The new consists a fully-connected layer and ReconNet. has low-dimension output acts as...

Deep neural networks have been applied to video compressive sensing (VCS) task recently. The existing DNN-based VCS methods compress and reconstruct the scene only in space or time dimensions, which ignores spatial-temporal correlation of video. And they generally utilize pixel-wise loss as function, causes results be over-smoothed. In this paper, we propose a perceptual network. network, compresses recovers both can preserve Besides, refine by selecting specific feature-wise terms adding...