- Advanced Algorithms and Applications
- Advanced Measurement and Detection Methods
- Advanced Image and Video Retrieval Techniques
- Wireless Sensor Networks and IoT
- Advanced Graph Neural Networks
- Video Surveillance and Tracking Methods
- Advanced Computational Techniques and Applications
- Sentiment Analysis and Opinion Mining
- Advanced Neural Network Applications
- Image Retrieval and Classification Techniques
- Underwater Acoustics Research
- Topic Modeling
- Industrial Vision Systems and Defect Detection
- Machine Learning and Data Classification
- Simulation and Modeling Applications
- Imbalanced Data Classification Techniques
- Vehicle Routing Optimization Methods
- Artificial Immune Systems Applications
- Advanced Sensor and Control Systems
- Infrastructure Maintenance and Monitoring
- Domain Adaptation and Few-Shot Learning
- Image Enhancement Techniques
- Text and Document Classification Technologies
- Infrared Target Detection Methodologies
- Target Tracking and Data Fusion in Sensor Networks
Hebei University of Technology
2016-2025
Institute of Acoustics
2013-2022
Xinjiang Normal University
2010
Hebei Normal University
2009
Hebei University
2008
Northwestern Polytechnical University
2007
In copy-move forgery, the illumination and contrast of tampered genuine regions are highly consistent, which poses a greater challenge in forgery detection. this article, an end-to-end neural network is proposed based on adaptive attention residual refinement (AR-Net). Specifically, position channel features fused by mechanism to fully capture context information enrich representation features. Second, deep matching adopted compute self-correlation between feature maps, atrous spatial...
Gesture recognition as a natural, convenient and recognizable way has been received more attention on human-machine interaction (HMI) recently. However, visual-based gesture methods are often restricted by environments classical wearable device-based strategies suffered from relatively low accuracy or the complicated structures. In this study, we first design low-cost efficient data glove with simple hardware structure to capture finger movement bending simultaneously. Second, novel dynamic...
Nowadays, capsule network model is widely used in image processing, whose feature engineering not suitable for sentiment analysis based on texts obviously. In this paper, we propose a with BiLSTM named caps-BiLSTM to solve the problem, and introduce experimental results different datasets. At beginning of caps-BiLSTM, convolution layer transform instance hide vector. Then module constructs representation n-gram model. The state probability certain calculated by If given largest among all...
Intelligent video surveillance is a vital technique in smart city construction, where detection of objects generally achieved by subtracting estimated background from the raw video. Common wisdom estimation focuses on introducing meaningful structure or discriminative hypothesis to sparsity-based objectives. However, relaxation optimization, which always considered most effective solution, definitely leads information loss. So, this article, as preserve more information, new nonconvex...
Ship detection from synthetic aperture radar (SAR) images is a hot topic, but the difficulty in collecting labeled SAR may hinder development of deep-learning-based methods. Inspired by idea domain adaptation, this article, we propose hierarchical similarity alignment neural network (HSANet) for ship images, which adaptive (DA) approach with optical remote sensing as training samples. The kernel target HSANet to mine and align both global structure local instance information between where...
In this paper, classification of pavement surface distress and the statistics data are discussed. order to improve accuracy efficiency identify by image information, a new algorithm based on SVM is study, support vector (SVC), which novel effective algorithm, applied crack images classification. build an SVC classifier, parameters must be selected carefully. This study pioneered using genetic optimize SVC. The performances back-propagation neural network whose obtained trial-and-error...
Inspired by the biological nervous system, deep neural networks (DNNs) are able to achieve remarkable performance in various tasks. However, they struggle handle label noise, which can poison memorization effects of DNNs. Co-teaching-based methods popular learning with noisy labels. These cross-train two DNNs based on small-loss criterion and employ a strategy using either “disagreement” or “consistency” obtain divergence networks. these sample-inefficient for generalization scenarios. In...
Due to the exceptional learning capabilities of deep neural networks (DNNs), they continue struggle handle label noise. To address this challenge, pseudo-label approach has emerged as a preferred solution. Recent works have achieved significant improvements by exploring information involved in DNN predictions and designing straightforward method incorporate model into training process using convex combination original labels targets. However, these methods overlook feature-level contained...
Ant colony algorithm is one kind of new heuristic biological modelling method which has the ability parallel processing and global searching, but its convergence speed slow because poor pheromone on early path. In this paper, discuss a combines genetic algorithm. Genetic added to ant algorithm's every generation in proposed Making use advantage whole quick convergence, quickened. mutation mechanism improves avoid being trapped local optimal. The simulation shows that effective solving...
Deep Learning (DL) has been broadly applied to solve big data problems in biomedical fields, which is most successful image processing. Recently, many DL methods have analyze genomic studies. However, usually too small a sample size fit complex network. They do not common structural patterns like images utilize pre-trained networks or take advantage of convolution layers. The concern overusing motivates us evaluate methods' performance versus popular non-deep Machine (ML) for analyzing with...
Foreground-background separation of surveillance video, that models static background and extracts moving foreground simultaneously, attracts increasing attentions in building a smart city. Conventional techniques towards this always consider the as primary target tend to adopt low-rank constraint its estimator, which provides finite (equal value rank) alternatives when constructing background. However, practical missions, although general sketch is stable, some details change constantly....
As the current level of people-counting, on consideration cost acquisition equipment and accuracy, customer-counting system is constructed based RBF neural net using technology infrared photoelectric sensor. For better differential count, extant data segmentation method feature extraction improved passenger-counting continuous space-time sequence. Compared with traditional sensor, accuracy real-time in this situation customers who entry at same time can be identified lower error rate. It...
TSP is a typical combinatorial optimization problems and GA an adaptive searching algorithm for global with the natural parallelism. The running speed of GA's software implementation too slow, so new scheme hardware was put forward. pipelining structure designed facilitating implementation, other parallel mechanisms were also added to it, thus greatly enhanced speed. design genetic operators - selection, crossover mutation, individual population fitness memory given. overall idea, sub-module...
Particle swarm optimization algorithm is a kind of auto-adapted search based on community.But the standard particle has defects prematurely, stagnation when applied in optimizing problems and easily leading to local minimum. A modified proposed improve from initial solution precision. The results show computation precision increased, convergence improved, minimum phenomenon mainly avoided. experimental classic functions that PSO efficient feasible.
In the smart home environment, aiming at disordered and multiple destinations path planning, sequencing rule is proposed to determine order of destinations. Within each branching process, initial feasible set generated according law attractive destination. A sinusoidal adaptive genetic algorithm adopted. It can calculate crossover probability mutation adaptively changing with environment any time. According cultural-genetic algorithm, it introduces concept reducing turns by parallelogram...
Training noise-robust deep neural networks (DNNs) in label noise scenario is a crucial task. In this paper, we first demonstrates that the DNNs learning with exhibits over-fitting issue on noisy labels because of too confidence its capacity. More significantly, however, it also potentially suffers from under-learning samples clean labels. essentially should pay more attention rather than samples. Inspired by sample-weighting strategy, propose meta-probability weighting (MPW) algorithm which...
Routing optimization is a key technology in the intelligent warehouse logistics. In order to get an optimal route for vehicle, routing complex global dynamic environment studied. A new evolutionary ant colony algorithm based on RFID and knowledge-refinement proposed. The gets environmental information timely through updates map at same time. It adopts elite kept, fallback, pheromones limitation adjustment strategy. current population space optimized experiential knowledge. experimental...
With the development of communication technology, recently, a lot attempts related to Intelligent Transport System have come forth at home and abroad. Based on exploration wireless in intelligent public transport system, hybrid model GPRS Bluetooth is provided which fast, stable secure new transportation system established existing model. It introduces composition working principle, completes design implementation system. This has been used actual project achieved certain results.
Passive worms have posed serious security threats to unstructured P2P networks. A delayed SEIRS epidemic model with death, offline and online rate is constructed based on the actual situation of users. The basic eproduction number that governs whether a passive worm extinct or not obtained. In this model, time delay consists latent temporary immunity periods. impact different parameters studied simulation results, especially effect delay, which can provide an important guideline in control over