Article
Computer Science, Hardware & Architecture
Tianyi Liu, Shuoyuan Wang, Yue Liu, Weiming Quan, Lei Zhang
Summary: This paper proposes a new method for implementing computationally lightweight convolution architectures on embedded platforms, which combines low-cost linear transformations with a convolution-based classifier to reduce computational and memory overhead while achieving satisfactory recognition performance.
JOURNAL OF SUPERCOMPUTING
(2022)
Article
Computer Science, Information Systems
Yong Li, Luping Wang, Fen Liu
Summary: In this paper, a multi-branch neural network based on attention-based convolution is proposed for sensor-based human activity recognition. Each branch consists of two layers of attention-based grouped convolution submodules. A dual attention mechanism is introduced to select important features and process sensor data from different locations independently. Experimental results show the superiority of the proposed method compared to existing methods.
Article
Engineering, Electrical & Electronic
Shige Xu, Lei Zhang, Wenbo Huang, Hao Wu, Aiguo Song
Summary: This article introduces a new deformable convolutional network for recognizing human activities from intricate sensory data. The proposed method achieves state-of-the-art recognition accuracies on several benchmark HAR datasets, indicating its advantage. Visual analysis shows that the deformable filter can be conditioned on different input activity samples, enhancing the interpretability of deep model behaviors. Evaluation on a Raspberry Pi shows that the deformable model maintains almost the same inference time, which is beneficial for activity recognition tasks.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Computer Science, Artificial Intelligence
Yunho Jeon, Junmo Kim
Summary: This study focuses on the convolution units within convolutional networks, proposing an active convolution unit (ACU) and grouped ACU, with a detailed analysis showing their efficiency. Experimental results demonstrate that these proposed units can replace existing convolution structures.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Information Systems
Dongzhou Cheng, Lei Zhang, Can Bu, Xing Wang, Hao Wu, Aiguo Song
Summary: Federated Learning (FL) has gained interest in sensor-based human activity recognition (HAR) tasks, but the Non-IID nature of sensor data in real-world environments poses challenges for FL methods. In this study, we propose ProtoHAR, a prototype-guided FL framework for HAR, which efficiently addresses the representation and classifier decoupling in heterogeneous FL settings. Experimental results demonstrate that ProtoHAR outperforms state-of-the-art FL algorithms in terms of performance and convergence speed on HAR datasets.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2023)
Article
Chemistry, Analytical
Sumya Akter, Rumman Ahmed Prodhan, Tanmoy Sarkar Pias, David Eisenberg, Jorge Fresneda Fernandez
Summary: Emotion recognition is an active research area with applications to improve people's lives. Most current image-based emotion recognition techniques are not accurate enough, so this paper proposes two convolutional neural network models with feature extraction methods to enhance recognition. The results show that these models outperform previous state-of-the-art models in terms of accuracy, especially the M2 model which achieves high accuracy with shorter time and data.
Article
Computer Science, Artificial Intelligence
Yan Huang, Jing Huang, Xiaoqiang Chen, Qicong Wang, Hongying Meng
Summary: This research proposes a novel similarity learning method in deep models that leverages global correlations to enhance discriminant feature learning. A global correlation inference model is used to induce the learning of a feature embedding model, and two loss functions are combined to provide feedback on global connection information for discriminant learning of the feature embedding model.
Article
Clinical Neurology
Yanhong Yu, Wentao Li, Yue Zhao, Jiayu Ye, Yunshao Zheng, Xinxin Liu, Qingxiang Wang
Summary: In this study, we used the Kinect V2 to collect skeletal data and proposed a novel spatial attention dilated TCN network for depression recognition. Our experiments and methods based on Kinect V2 not only identified and classified depression patients accurately but also observed the recovery level of depression patients during the recovery process.
FRONTIERS IN NEUROLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Ran He, Yi Li, Xiang Wu, Lingxiao Song, Zhenhua Chai, Xiaolin Wei
Summary: This study introduces a coupled adversarial learning (CAL) approach to address the challenging issue of VIS-NIR face matching, by conducting adversarial learning on both image and feature levels. Experimental results demonstrate that CAL not only synthesizes high-quality VIS or NIR images, but also achieves state-of-the-art recognition results.
PATTERN RECOGNITION
(2021)
Article
Chemistry, Multidisciplinary
Gokmen Ascioglu, Yavuz Senol
Summary: This study presents a wireless smart insole for activity monitoring, which uses force-sensitive resistors (FSRs) to monitor foot contact states and deep learning algorithms for activity recognition. The results show that a neural network using only foot contact states achieves high accuracy and performs better than other datasets for activity recognition.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Artificial Intelligence
Yun Xiong, Yizhou Zhang, Xiangnan Kong, Huidi Chen, Yangyong Zhu
Summary: This study investigates deep collective classification in Heterogeneous Information Networks (HINs) and proposes a method called GraphInception, a deep convolutional collective classification method that can learn deep relational features in HINs and generate a hierarchy of relational features with different complexities. Extensive experiments show that considering deep relational features in HINs can improve collective classification performance.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2021)
Article
Engineering, Electrical & Electronic
Rebeen Ali Hamad, Longzhi Yang, Wai Lok Woo, Bo Wei
Summary: Existing models for human activity recognition based on sensor data have achieved state-of-the-art performances. However, training separate models for each domain is time-consuming and computationally expensive. To address this issue, we propose a multi-domain learning network that transfers knowledge across related domains and mitigates isolated learning paradigms using a shared representation.
IEEE SENSORS JOURNAL
(2022)
Article
Agronomy
Jianhao Yin, Junfeng Wu, Chunqi Gao, Zhongai Jiang
Summary: With the continuous development of industrial aquaculture and artificial intelligence technology, the use of automation and intelligence in aquaculture is increasing. Individual fish recognition plays an important role in fish farming, but it faces challenges due to the complexity of underwater environment and high similarity among individual fish. This paper proposes a method based on lightweight convolutional neural network for individual fish recognition, which can accurately and efficiently extract visual features of underwater moving fish and provide unique identity recognition information for each fish.
Article
Engineering, Electrical & Electronic
Ruohong Huan, Chengxi Jiang, Luoqi Ge, Jia Shu, Ziwei Zhan, Peng Chen, Kaikai Chi, Ronghua Liang
Summary: In this paper, the optimal feature representation of human complex activities is studied, and a method for extracting multi-layer features from a hybrid CNN and BLSTM network is proposed. A new feature selection method is also introduced to generate mixed features by fusing different sources. Experimental results demonstrate that this method outperforms existing approaches in human complex activity recognition.
IEEE SENSORS JOURNAL
(2022)
Article
Computer Science, Information Systems
Mohamed Abdel-Basset, Hossam Hawash, Victor Chang, Ripon K. Chakrabortty, Michael Ryan
Summary: With the development of wireless sensing technology, the Internet of Things (IoT) has been widely applied in various fields. Human activity recognition (HAR) is an important component in intelligent systems for continuous monitoring of human behaviors. In this research, smartphone inertial sensors are used as a case study, and HAR is treated as an image classification problem. A lightweight model called multiscale image-encoded HHAR (MS-IE-HHAR) is proposed and achieves high accuracy in modeling HAR from heterogeneous data sources.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Automation & Control Systems
Yin Tang, Lei Zhang, Fuhong Min, Jun He
Summary: This article proposes a new CNN that uses a hierarchical-split idea to enhance multiscale feature representation ability in wearable human activity recognition. The experiments demonstrate that the proposed method outperforms baseline models and achieves higher recognition performance without increasing resource consumption. Ablation studies are conducted to evaluate the effect of receptive field variations on classification performance, and it is shown that multiscale receptive fields can help learn discriminative features.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2023)
Article
Green & Sustainable Science & Technology
Zhi Wang, Xianyong Peng, Shengxian Cao, Huaichun Zhou, Siyuan Fan, Kuangyu Li, Wenbo Huang
Summary: The study focuses on the selective catalytic reduction (SCR) denitrification efficiency of coal-fired boilers. Accurate predictions of NOx emissions at the SCR inlet can improve denitrification efficiency. Deep learning, random forest (RF) algorithm, and lightweight convolutional neural network (CNN) were used to develop a prediction approach. The experimental results showed that the proposed method reduced model complexity while ensuring prediction accuracy, making it suitable for online optimization of industrial pollutant control and cleaner production.
JOURNAL OF CLEANER PRODUCTION
(2023)
Article
Computer Science, Artificial Intelligence
Li Xu, Jun Zeng, Weile Peng, Hao Wu, Kun Yue, Haiyan Ding, Lei Zhang, Xin Wang
Summary: This work proposes a novel technological framework, MIA-SR, for sequential recommendation by modeling and predicting user preferences with multiple item attributes. The framework utilizes the item attribute information and a graph convolution network to enhance the representations of items and attributes. MIA-SR also employs a multi-tasking strategy to improve item recommendation. Experimental results show the effectiveness of MIA-SR.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Automation & Control Systems
Hailong Chen, Hao Wu, Jiahui Li, Xin Wang, Lei Zhang
Summary: Developers need to reuse web services and create mashups suitable for various scenarios. However, inexperienced developers may not be able to adequately express their requirements when using service recommendation systems, leading to inappropriate recommendations. To address this, a service-keyword correlation graph (SKCG) is defined to capture the relationship between services and keywords. A keyword-based deep reinforced Steiner tree search (K-DRSTS) approach is then proposed, which models the task of service discovery as a Steiner tree search problem and uses deep reinforcement learning to provide an efficient solution. Experimental results on real-world data sets demonstrate the effectiveness of K-DRSTS.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Engineering, Electrical & Electronic
Yiting Liu, Lianjie Sui, Peijuan Li, Lei Zhang, Qingzheng Wu, Junfeng Du, Yawen Liu, Hanqi Yu
Summary: This paper proposes a radar linear feature fitting algorithm that combines adaptive clustering and corner detection operators to effectively avoid the influence of noise points and a fixed segmentation threshold on corner point extraction. Experimental results show that the proposed algorithm accurately extracts corner features and performs linear positioning with higher calculation efficiency and position accuracy.
JOURNAL OF SENSORS
(2023)
Article
Computer Science, Information Systems
Yiting Liu, Lei Zhang, Peijuan Li, Tong Jia, Junfeng Du, Yawen Liu, Rui Li, Shutao Yang, Jinwu Tong, Hanqi Yu
Summary: In this paper, a laser radar data registration algorithm based on DBSCAN clustering is proposed, which avoids the search and establishment of the corresponding relationship. A kernel density estimation method is used to describe the registered point cloud, and K-L divergence is used to find the optimal value in the key clusters.
Article
Computer Science, Artificial Intelligence
Dongzhou Cheng, Lei Zhang, Can Bu, Hao Wu, Aiguo Song
Summary: In this paper, a novel approach called Contrastive Supervision is proposed to improve the classification accuracy of human activity recognition tasks. By leveraging contrastive learning and deeply-supervised learning, the approach learns time series augmentation invariances and effectively fuses multi-level features.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Information Systems
Dongzhou Cheng, Lei Zhang, Can Bu, Xing Wang, Hao Wu, Aiguo Song
Summary: Federated Learning (FL) has gained interest in sensor-based human activity recognition (HAR) tasks, but the Non-IID nature of sensor data in real-world environments poses challenges for FL methods. In this study, we propose ProtoHAR, a prototype-guided FL framework for HAR, which efficiently addresses the representation and classifier decoupling in heterogeneous FL settings. Experimental results demonstrate that ProtoHAR outperforms state-of-the-art FL algorithms in terms of performance and convergence speed on HAR datasets.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2023)
Article
Computer Science, Information Systems
Wenbo Huang, Lei Zhang, Hao Wu, Fuhong Min, Aiguo Song
Summary: Human activity recognition (HAR) using wearable sensors has gained significant attention recently, and this paper proposes a novel method called Channel Equalization to reactivate inhibited sensor channels and improve feature representation. Experimental results demonstrate that this method achieves higher recognition performance compared to baseline models, surpassing current state-of-the-art approaches.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Yueting Fang, Hao Wu, Yiji Zhao, Lei Zhang, Shaowei Qin, Xin Wang
Summary: Graph neural network (GNN) is a powerful model for processing non-Euclidean data, such as graphs, in recommendation tasks. However, existing GNN models lack attention to recommendation diversity. This work proposes a novel graph spreading network (GSN) model that addresses the accuracy-diversity dilemma in recommendation by modifying the propagation rule and developing a new sampling strategy. GSN effectively improves diversity while maintaining accuracy through a selective sampler.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Chaolei Han, Lei Zhang, Shige Xu, Xing Wang, Hao Wu, Aiguo Song
Summary: Deep convolutional networks have achieved great success in sensor-based human activity recognition. Most existing works focus on extracting multiscale activity features by increasing network depth or width, which is not suitable for resource-limited mobile devices. To address this issue, we propose a diverse-branch convolution (DBC) scheme, which strengthens the capacity of vanilla convolution by exploiting diverse branches of different scales and complexities. DBC can be transformed into a single convolution layer for activity recognition deployment while maintaining the model's performance and inference-time structure.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Qi Teng, Yin Tang, Guangwei Hu
Summary: This article proposes a resource-efficient dynamic network called RepHAR for low-cost hardware-constrained human activity recognition tasks. RepHAR achieves a trade-off between speed and accuracy by combining multibranch topologies and structured reparameterization techniques. Experimental results on four publicly available datasets demonstrate the effectiveness of the proposed method. RepHAR runs 72% faster than the multibranch CNN model on the embedded Raspberry Pi, showcasing its usefulness and practicality. Our model and codes will be released soon.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Automation & Control Systems
Can Bu, Lei Zhang, Hengtao Cui, Guangyu Yang, Hao Wu
Summary: Deep convolutional neural networks have shown impressive performance in human activity recognition using wearable sensors. However, their computational cost is typically high when using fixed-length sliding windows. This article proposes a novel approach to accelerate activity inference by reducing temporal redundancy in sensor data. The approach formulates the activity prediction problem as a dynamic inference process, where activity-discriminative intervals are selected from a window based on their discriminative importance. The proposed dynamic process adaptively decides when to halt computation for each sample, effectively reducing excessive computation and improving computational efficiency compared to previous static methods.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Review
Computer Science, Artificial Intelligence
Wei Gao, Shuangshuang Ge
Summary: This study provides a comprehensive review of slope stability research based on artificial intelligence methods, focusing on slope stability computation and evaluation. The review covers studies using quasi-physical intelligence methods, simulated evolutionary methods, swarm intelligence methods, hybrid intelligence methods, artificial neural network methods, vector machine methods, and other intelligence methods. The merits, demerits, and state-of-the-art research advancement of these studies are analyzed, and possible research directions for slope stability investigation based on artificial intelligence methods are suggested.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Khuong Le Nguyen, Hoa Thi Trinh, Saeed Banihashemi, Thong M. Pham
Summary: This study investigated the influence of input parameters on the shear strength of RC squat walls and found that ensemble learning models, particularly XGBoost, can effectively predict the shear strength. The axial load had a greater influence than reinforcement ratio, and longitudinal reinforcement had a more significant impact compared to horizontal and vertical reinforcement. The performance of XGBoost model outperforms traditional design models and reducing input features still yields reliable predictions.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Bo Hu, Huiyan Zhang, Xiaoyi Wang, Li Wang, Jiping Xu, Qian Sun, Zhiyao Zhao, Lei Zhang
Summary: A deep hierarchical echo state network (DHESN) is proposed to address the limitations of shallow coupled structures. By using transfer entropy, candidate variables with strong causal relationships are selected and a hierarchical reservoir structure is established to improve prediction accuracy. Simulation results demonstrate that DHESN performs well in predicting algal bloom.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Limin Wang, Lingling Li, Qilong Li, Kuo Li
Summary: This paper discusses the urgency of learning complex multivariate probability distributions due to the increase in data variability and quantity. It introduces a highly scalable classifier called TAN, which utilizes maximum weighted spanning tree (MWST) for graphical modeling. The paper theoretically proves the feasibility of extending one-dependence MWST to model high-dependence relationships and proposes a heuristic search strategy to improve the fitness of the extended topology to data. Experimental results demonstrate that this algorithm achieves a good bias-variance tradeoff and competitive classification performance compared to other high-dependence or ensemble learning algorithms.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Zhejing Hu, Gong Chen, Yan Liu, Xiao Ma, Nianhong Guan, Xiaoying Wang
Summary: Anxiety is a prevalent issue and music therapy has been found effective in reducing anxiety. To meet the diverse needs of individuals, a novel model called the spatio-temporal therapeutic music transfer model (StTMTM) is proposed.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Nur Ezlin Zamri, Mohd. Asyraf Mansor, Mohd Shareduwan Mohd Kasihmuddin, Siti Syatirah Sidik, Alyaa Alway, Nurul Atiqah Romli, Yueling Guo, Siti Zulaikha Mohd Jamaludin
Summary: In this study, a hybrid logic mining model was proposed by combining the logic mining approach with the Modified Niche Genetic Algorithm. This model improves the generalizability and storage capacity of the retrieved induced logic. Various modifications were made to address other issues. Experimental results demonstrate that the proposed model outperforms baseline methods in terms of accuracy, precision, specificity, and correlation coefficient.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
David Jacob Kedziora, Tien-Dung Nguyen, Katarzyna Musial, Bogdan Gabrys
Summary: The paper addresses the problem of efficiently optimizing machine learning solutions by reducing the configuration space of ML pipelines and leveraging historical performance. The experiments conducted show that opportunistic/systematic meta-knowledge can improve ML outcomes, and configuration-space culling is optimal when balanced. The utility and impact of meta-knowledge depend on various factors and are crucial for generating informative meta-knowledge bases.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
G. Sophia Jasmine, Rajasekaran Stanislaus, N. Manoj Kumar, Thangamuthu Logeswaran
Summary: In the context of a rapidly expanding electric vehicle market, this research investigates the ideal locations for EV charging stations and capacitors in power grids to enhance voltage stability and reduce power losses. A hybrid approach combining the Fire Hawk Optimizer and Spiking Neural Network is proposed, which shows promising results in improving system performance. The optimization approach has the potential to enhance the stability and efficiency of electric grids.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Zhijiang Wu, Guofeng Ma
Summary: This study proposes a natural language processing-based framework for requirement retrieval and document association, which can help to mine and retrieve documents related to project managers' requirements. The framework analyzes the ontology relevance and emotional preference of requirements. The results show that the framework performs well in terms of iterations and threshold, and there is a significant matching between the retrieved documents and the requirements, which has significant managerial implications for construction safety management.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Yung-Kuan Chan, Chuen-Horng Lin, Yuan-Rong Ben, Ching-Lin Wang, Shu-Chun Yang, Meng-Hsiun Tsai, Shyr-Shen Yu
Summary: This study proposes a novel method for dog identification using nose-print recognition, which can be applied to controlling stray dogs, locating lost pets, and pet insurance verification. The method achieves high recognition accuracy through two-stage segmentation and feature extraction using a genetic algorithm.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Shaohua Song, Elena Tappia, Guang Song, Xianliang Shi, T. C. E. Cheng
Summary: This study aims to optimize supplier selection and demand allocation decisions for omni-channel retailers in order to achieve supply chain resilience. It proposes a two-phase approach that takes into account various factors such as supplier evaluation and demand allocation.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Jinyan Hu, Yanping Jiang
Summary: This paper examines the allocation problem of shared parking spaces considering parking unpunctuality and no-shows. It proposes an effective approach using sample average approximation (SAA) combined with an accelerating Benders decomposition (ABD) algorithm to solve the problem. The numerical experiments demonstrate the significance of supply-demand balance for the operation and user satisfaction of the shared parking system.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Review
Computer Science, Artificial Intelligence
Soroor Motie, Bijan Raahemi
Summary: Financial fraud is a persistent problem in the finance industry, but Graph Neural Networks (GNNs) have emerged as a powerful tool for detecting fraudulent activities. This systematic review provides a comprehensive overview of the current state-of-the-art technologies in using GNNs for financial fraud detection, identifies gaps and limitations in existing research, and suggests potential directions for future research.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Review
Computer Science, Artificial Intelligence
Enhao Ning, Changshuo Wang, Huang Zhang, Xin Ning, Prayag Tiwari
Summary: This review provides a detailed overview of occluded person re-identification methods and conducts a systematic analysis and comparison of existing deep learning-based approaches. It offers important theoretical and practical references for future research in the field.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Jiajun Ma, Songyu Hu, Jianzhong Fu, Gui Chen
Summary: The article presents a novel visual hierarchical attention detector for multi-scale defect location and classification, utilizing texture, semantic, and instance features of defects through a hierarchical attention mechanism, achieving multi-scale defect detection in bearing images with complex backgrounds.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)