检索结果-内蒙古大学图书馆

Digital Analysis and Processing, Intelligent Computation (DAPIC), International Conference on

作者： Xinpeng Chen Yuzhang Li Ziyang Zhang School of Computer and Information Engineering Xinjiang Agricultural University Urumqi China School of Electrical Engineering and Artificial Intelligence Xiamen University Malaysia Sepang Malaysia School of Data Science and Computer Science Guangdong Peizheng College Shenzhen China

ISBN: (数字)9798331536169

ISBN: (纸本)9798331536176

As the fourth largest food crop in the world, potatoes are prone to a number of illnesses that drastically lower the quality and output during cultivation. Most traditional deep learning models for detecting potato diseases are trained on datasets with only three categories, which cannot accurately confirm the type of disease. Meanwhile, the background of datasets in traditional controlled environments is too homogeneous, which may not be able to adapt to the application scenarios of real agricultural production. The study suggests EffViT-PotatoNet, a deep learning model built on EfficientViT architecture, as a solution to this issue. This approach combines the multi-scale attention mechanism of EfficientViT with a multi-layer fully connected network and regularization techniques to improve the classification accuracy, especially the performance in complex backgrounds. A dataset of potato leaf illnesses from Uncontrolled Environment, which includes pictures of both healthy and diseased leaves, was used in this investigation. Considering the sample imbalance of the dataset, a weighted cross-entropy method was used for optimization and the images were augmented by RandAugment. EffViTPotatoNet was fine-tuned on top of the pre-trained EfficientViT, and the model was optimized by introducing Batch Normalization and Dropout layers, and finally the probability distributions of seven categories are output by Softmax. In the experiments, EffViT-PotatoNet achieved better performance, especially in the case of the diversity of the dataset and the high environmental interference, demonstrating strong classification ability and better generalization. This study provides an efficient and scalable solution for crop disease detection in agriculture with potential practical applications.

关键词： Smart agriculture Adaptation models Plant diseases Transfer learning Crops Production Predictive models Transformers Probability distribution Diseases

来源：评论

学校读者我要写书评

暂无评论

Generalized Phase Pressure Control Enhanced Reinforcement Learning for Traffic Signal Control

arXiv

引用

arXiv 2025年

作者： Liao, Xiao-Cheng Mei, Yi Zhang, Mengjie Chen, Xiang-Ling Centre for Data Science and Artificial Intelligence School of Engineering and Computer Science Victoria University of Wellington Wellington New Zealand

Appropriate traffic state representation is crucial for learning traffic signal control policies. However, most of the current traffic state representations are heuristically designed, with insufficient theoretical support. In this paper, we (1) develop a flexible, efficient, and theoretically grounded method, namely generalized phase pressure (G2P) control, which takes only simple lane features into consideration to decide which phase to be actuated;2) extend the pressure control theory to a general form for multi-homogeneous-lane road networks based on queueing theory;(3) design a new traffic state representation based on the generalized phase state features from G2P control;and 4) develop a reinforcement learning (RL)-based algorithm template named G2P-XLight, and two RL algorithms, G2P-MPLight and G2P-CoLight, by combining the generalized phase state representation with MPLight and CoLight, two well-performed RL methods for learning traffic signal control policies. Extensive experiments conducted on multiple real-world datasets demonstrate that G2P control outperforms the state-of-the-art (SOTA) heuristic method in the transportation field and other recent human-designed heuristic methods;and that the newly proposed G2PXLight significantly outperforms SOTA learning-based approaches. Our code is available online1 Copyright © 2025, The Authors. All rights reserved.

关键词： Heuristic methods

来源：评论

学校读者我要写书评

暂无评论

Closer through commonality: Enhancing hypergraph contrastive learning with shared groups

arXiv

引用

arXiv 2025年

作者： Roh, Daeyoung Han, Donghee Kim, Daehee Han, Keejun Yi, Mun Graduate School of Data Science KAIST Korea Republic of School of Computer Engineering Hansung University Korea Republic of Department of Industrial and Systems Engineering KAIST Korea Republic of

Hypergraphs provide a superior modeling framework for representing complex multidimensional relationships in the context of real-world interactions that often occur in groups, overcoming the limitations of traditional homogeneous graphs. However, there have been few studies on hypergraph-based contrastive learning, and existing graph-based contrastive learning methods have not been able to fully exploit the high-order correlation information in hypergraphs. Here, we propose a Hypergraph Fine-grained contrastive learning (HyFi) method designed to exploit the complex high-dimensional information inherent in hypergraphs. While avoiding traditional graph augmentation methods that corrupt the hypergraph topology, the proposed method provides a simple and efficient learning augmentation function by adding noise to node features. Furthermore, we expands beyond the traditional dichotomous relationship between positive and negative samples in contrastive learning by introducing a new relationship of weak positives. It demonstrates the importance of fine-graining positive samples in contrastive learning. Therefore, HyFi is able to produce high-quality embeddings, and outperforms both supervised and unsupervised baselines in average rank on node classification across 10 datasets. Our approach effectively exploits high-dimensional hypergraph information, shows significant improvement over existing graph-based contrastive learning methods, and is efficient in terms of training speed and GPU memory cost. The source code is available at https://***/Noverse0/***. © 2025, CC BY.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Differentially Private Spiking Variational Autoencoder 27th

Differentially Private Spiking Variational Autoencoder

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Yadav, Srishti Pundhir, Anshul Goyal, Tanish Raman, Balasubramanian Kumar, Sanjeev Mehta Family School of Data Science and Artificial Intelligence Indian Institute of Technology Roorkee Roorkee India Department of Computer Science and Engineering Indian Institute of Technology Roorkee Roorkee India Department of Mathematics Indian Institute of Technology Roorkee Uttarakhand Roorkee India

ISBN: (纸本)9783031783531

Spiking Neural Networks (SNNs) are poised to lead the next generation of artificial intelligence, offering energy efficiency and performance on par with traditional neural networks. With these advantages, SNNs are finding widespread applications across various domains. One significant area of interest is image generation using deep learning models like Variational Autoencoders (VAE). However, like other deep learning models, SNNs demand substantial training data to achieve desired outcomes, raising concerns about data privacy. Our pioneering contribution is the introduction of a Differentially Private Spiking Variational Autoencoder (DP-SVAE) for image generation and reconstruction. DP-SVAE employs standard Differentially Private Stochastic Gradient Descent (DP-SGD) to ensure privacy preservation. Additionally, we have evaluated the models against various adversarial attacks to highlight the importance of differential privacy. We comprehensively analyze the proposed model through extensive experimentation across publicly available benchmark datasets. This pioneering study marks the first exploration of privacy considerations in SNN-based VAEs and will catalyze further research in this domain. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Differential privacy

来源：评论

学校读者我要写书评

暂无评论

A Trace-Selection Preprocessing Approach of Discovering Structurally Complete ICN-Process Models from Process Logs 3rd

A Trace-Selection Preprocessing Approach of Discovering Str...

引用

3rd International Conference on Advances in Information and Communication Technology, ICTA 2024

作者： Pham, Dinh-Lam Batzaya, Tsagaantsooj Kim, Kyoung-Sook Chang, Jeong-Hyun Kim, Kwanghoon Pio CCSRI Kyonggi University Gyeonggi Suwon16227 Korea Republic of Department of Public Safety Bigdata Graduate School of Kyonggi University Gyeonggi Suwon16227 Korea Republic of Data Process Engineering Research Laboratory Faculty of AI Computer Science and Engineering Kyonggi University Gyeonggi Suwon16227 Korea Republic of

ISBN: (纸本)9783031809422

In business process life-cycle management and reengineering through process mining, it is crucial for the process mining system to discover structurally safe and complete business process models from process logs. However, most process mining systems typically suffer from discovering inaccurate and unstructured business process models due to various types of anomalous traces hidden in process logs. In this paper, we therefore propose a process log preprocessing approach that purifies a corresponding process log dataset by selecting a group of reasonable process log traces and applying to discover an essential business process model satisfying the structural completeness requirement of matched-pairing and proper-nesting properties. The theoretical basis of the proposed approach is the mathematical graph model of information control net model that is the typical representation of business process models. Finally, we practically demonstrate the conceptual excellence and functional correctness of the proposed preprocessing approach through experimental verifications using real business process enactment log datasets available in the 4TU Center for Research data. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Reengineering

来源：评论

学校读者我要写书评

暂无评论

Causal fMRI-Mamba: Causal State Space Model for Neural Decoding and Brain Task States Recognition

Causal fMRI-Mamba: Causal State Space Model for Neural Decod...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Weihao Deng Fei Han Qinghua Ling Qing Liu Henry Han School of Computer Science and Communication Engineering JiangSu University China Jiangsu Engineering Research Center of Big Data Ubiquitous Perception and Intelligent Agriculture Applications China School of Computer Jiangsu University of Science and Technology China School of Electronic and Information Engineering West Anhui University China School of Engineering and Computer Science Baylor University USA

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Deep learning advances neural decoding in functional magnetic resonance imaging (fMRI) tasks with convolution and attention-based methods. However, these methods struggle with capturing global spatiotemporal information due to high dimensionality, noise and inter-individual difference of fMRI, which also increase computational complexity and prior bias. To this end, a novel causal state space model, Causal fMRI-Mamba, is proposed for neural decoding and task state mapping. It effectively captures global spatiotemporal information via eliminating local redundancies and capturing long-distance dependencies. Meanwhile, a causal representation framework is designed to extract invariant high-order features and disentangle related causal features, enhancing model performance. Furthermore, a dense connection module is expanded to prevent significant causal information loss in hidden states of inter layers. On the HCP brain task state classification task, Causal fMRI-Mamba achieves better performance and generalization than comparison methods.

关键词： Deep learning Perturbation methods Redundancy Noise Functional magnetic resonance imaging Brain modeling Feature extraction Spatiotemporal phenomena Decoding Speech processing

来源：评论

学校读者我要写书评

暂无评论

Evolutionary Paths to Systematically Unlocking the Frequency-Selective Potential of 3D Meta-Atoms

Evolutionary Paths to Systematically Unlocking the Frequency...

引用

Australian Microwave Symposium (AMS)

作者： Xiaojing Lv Zhen Luo J. Scott Tyo Yang Yang School of Electrical and Data Engineering University of Technology Sydney Australia School of Mechanical and Mechatronic Engineering University of Technology Sydney Australia Department of Electrical and Computer Systems Engineering Monash University Australia

ISBN: (数字)9798350379969

ISBN: (纸本)9798350379976

This paper highlights the benefits of advanced multi-material 3D printing technology in enhancing the electromagnetic performance of metasurfaces. By systematically evolving geometric structures, the proposed bandpass filtering meta-atom outperforms existing designs in key metrics, including bandwidth, out-of-band suppression, in-band insertion loss, and transition band steepness, underscoring the significant potential of 3D unit cells. Through the integration of qualitative approach and semi-analytical design algorithm, the geometric topology is fully optimized without imposing substantial computational demands.

关键词： Measurement Electric potential Filtering Insertion loss Bandwidth Three-dimensional printing Metasurfaces Topology Microwave filters Electromagnetics

来源：评论

学校读者我要写书评

暂无评论

Meta Heuristic Image Optimisation in LiDAR Sensor Based on Cloud IoT Network and Deep Learning Algorithms

引用

Remote Sensing in Earth Systems Sciences 2025年第2期8卷 566-575页

作者： Meenakshi, K. Manoj, Gunna Diwakar, Meenakshi Prabakaran, Manjula Sravan, Padala Babu, Gaddam Ravindra Department of Networking and Communications School of Computing SRM Institute of Science and Technology Kattankulathur Chennai India Cyber Security Marri Laxman Reddy Institute of Technology and Management Telangana Hyderabad India Department of IT Panimalar Engineering College Chennai India Dept of CSE (Data Science) Madanapalle Institute of Technology & Science Andhra Pradesh Angallu India Computer Science and Engineering NIIT with learning partner Hainan Normal University Haikou China Department of Computer Science and Engineering Koneru Lakshmaiah Education Foundation Telangana Hyderabad India

LiDAR technology has become a vital tool for researching these atmospheric components because of its capacity to measure backscattered signals and investigate the atmosphere at extremely high spatial and temporal resolution. However, machine learning has become a potent technique for sophisticated data processing and analysis because of the complexity and diversity of lidar technology. The aim of this research is to propose a novel technique in cloud IoT network-based LiDAR sensor imaging with image optimization using a metaheuristic deep learning model. Input has been collected LiDAR image using a cloud IoT module and processed for noise removal and normalization. Then this processed image is classified and optimized utilizing a quantile transfer graph convolutional model with binary firefly whale colony optimization. The experimental analysis is carried out in terms of detection accuracy, random precision, recall, RMSE, and AUC for various LiDAR sensor-based datasets. Comparing the results of the various approaches reveals that the suggested approach outperforms them all with a recall of 94%, a random precision of 95%, and a detection accuracy of 98%. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025.

关键词： Cloud IoT network Convolutional model Image optimization LiDAR sensor Metaheuristic deep learning

来源：评论

学校读者我要写书评

暂无评论

Psmf: Prototype Network Subgraph with Multi-Head Attention Framework for Group Emotion Recognition

SSRN

引用

SSRN 2025年

作者： Li, Chenhao Huang, Wenti Long, Yitian Chen, Yunfei Yang, Zhan Long, Jun Computer Science and Technology Xinjiang University Urumqi830000 China Big Data Institute School of Computer Science and Engineering Central South University Changsha410083 China School of Computer Science and Engineering Hunan University of Science and Technology Xiangtan411100 China Data Science Institute Vanderbilt University Nashville37235 United States

Group Emotion Recognition (GER) is an essential branch of affective computing aimed at revealing the dynamic changes in emotional states within a group and the overall characteristics of collective emotions. Integrating emotional features from multiple individuals within a group has become a significant challenge in current research. Existing mainstream methods typically extract various emotion-related features from images and use multi-cue fusion to correlate these features, revealing emotional expressions across different individuals in the group. However, these methods face two key issues: 1) Poor fusion of features across different modalities caused by uneven data distribution and 2) Weak recognition ability for rare emotional expressions. To address these challenges, we propose a novel framework—Prototype Network Subgraph Bridging Multi-Head Attention Framework (PSMF). This framework leverages graph-based reasoning to extend the prototype network's classification ability at each stage, effectively integrating features across different cues. Specifically, PSMF utilizes the support and query sets in the feature graph to store and bridge recognition features in subgraphs. With the multi-head attention mechanism, PSMF can thoroughly learn the distances between each feature and the prototype network, balancing competition between categories. This enhances the recognition of rare emotional expressions while improving the fusion of features across different modalities. Extensive experiments on datasets such as GroupEmoW and AffectNet demonstrate that our framework achieves state-of-the-art performance, significantly enhancing feature fusion across different cues. © 2025, The Authors. All rights reserved.

关键词： Emotion Recognition

来源：评论

学校读者我要写书评

暂无评论

Unveiling Hidden Water Resources: Deep Learning and Remote Sensing for Subsurface Hydrology for Environmental Health

引用

Remote Sensing in Earth Systems Sciences 2025年第2期8卷 352-364页

作者： Babu, Jampani Satish G, Dinesh Suganya, R. Radhakrishnan, P. Sutar, Anand S, Alex David Rao, L. Malleswara Department of Computer Science and Engineering Koneru Lakshmaiah Education Foundation Guntur AP Vaddeswaram India Department of Computational Intelligence School of Computing SRM Institute of Science and Technology Kattankulathur (KTR) Chennai India Department of Computer Science and Engineering (Data Science) New Horizon College of Engineering Karnataka Bengaluru India Department of ECE Tagore Engineering College Chennai India Department of Electronics and Telecommunication Bharati Vidyapeeth College of Engineering Navi Mumbai India Department of Computer Science and Engineering Vel Tech Rangarajan Dr Sagunthala R & d Institute of Science and Technology Chennai India Department of Physics Sri Y N College (A) West Godavari District Andhra Pradesh Narsapur India

Water scarcity in arid regions necessitates innovative solutions to uncover and manage subsurface water resources for environmental health. This study explores the potential of GeoHydroNet, a specialized deep learning framework, in identifying hidden water resources within arid regions. Utilizing remote sensing data such as multispectral and hyperspectral imagery, geological maps, and soil moisture datasets, GeoHydroNet combines graph neural Networks (GNNs), AutoEncoders, and generative adversarial networks (GANs) to address challenges in subsurface hydrology. The model achieved an accuracy of 93.5%, surpassing traditional models like CNNs (85.4%), LSTMs (88.1%), and RNNs (82.6%). GeoHydroNet demonstrates significant precision (0.88), recall (0.87), and sensitivity to data imbalance (0.91). Its physics-informed learning approach incorporates hydrological principles, ensuring realistic predictions. By analyzing subsurface water patterns, recharge zones, and geological influences, the model not only enhances prediction accuracy but also aids in sustainable groundwater management. GeoHydroNet’s performance in identifying groundwater levels, with minimal errors ranging from 0.62 to 3.07%, indicates its robustness in tackling water scarcity challenges in data-constrained environments. These findings highlight its transformative role in hydrological modeling and sustainable water resource management in arid regions. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025.

关键词： Aquifers Arid regions Deep learning Generative adversarial networks GeoHydroNet Physics-informed learning Remote sensing Subsurface hydrology Sustainable water management

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：