检索结果-内蒙古大学图书馆

31st International Conference on Multimedia Modeling, MMM 2025

作者： Yang, Enhui Zhang, Zhibin Engineering Research Center of Ecological Big Data Ministry of Education Key Laboratory of Wireless Networks and Mobile Computing Inner Mongolia University Hohhot010021 China

ISBN: (纸本)9789819620630

The Segment Anything Model (SAM), introduced in 2023, has made significant advancements in the field of computer vision. However, SAM faces two major challenges: limitations related to single-scale processing and high computational demands for large-scale data handling, which restrict its widespread adoption in practical applications. Therefore, we propose MS-SAM to achieve better segmentation performance at a lower cost. We introduce a multi-scale image encoder to handle images of varying scales and complexities. Additionally, we propose a dynamic weighted agent attention, which can flexibly and efficiently adjust attention distribution while reducing computational costs. We train MS-SAM on a 1/20 subset of the SA-1B dataset for instance segmentation using points or boxes as prompts. The results show that we maintain the accuracy with the SAM method while reducing the parameters by about 6 times. We give sufficient experimental results to demonstrate its effectiveness. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

CASSNet: Cross-Attention Enhanced Spectral–Spatial Interaction Network for Hyperspectral Image Super-Resolution

引用

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 2025年 18卷 11716-11730页

作者： Zhang, Zhanxu Yang, Linzi Zhang, Guanglian Deng, Jiangwei Bian, Lifeng Yang, Chen Guizhou University Power Systems Engineering Research Center Ministry of Education College of Big Data and Information Engineering Guiyang550025 China Fudan University Frontier Institute of Chip and System Shanghai200433 China Guizhou University China State Key Laboratory of Public Big Data Guiyang550025 China

Deep-learning-based super-resolution (SR) methods for a single hyperspectral image have made significant progress in recent years and become an important research direction in remote sensing. Existing methods perform well in extracting spatial features, but challenges remain in integrating spectral and spatial features when modeling global relationships. In order to take full advantage of the higher spectral resolution of hyperspectral images, this article proposes a novel hyperspectral image SR method (CASSNet), which integrates convolutional neural networks and cross-attention mechanisms into a unified framework. This approach achieves comprehensive integration of spectral and spatial information, with extensive exploration at both local and global levels. In the local feature extraction stage, parallel 3-D/2-D convolutions work in tandem to efficiently capture detail information from both spectral and spatial dimensions. In addition, a spectral–spatial dual-branch module employing the cross-attention mechanism is designed to capture the global dependencies within the features, where the reconstructed spectral–spatial module and the spectral–spatial interaction unit can effectively promote the interaction and complementarity of spectral–spatial features. The experiments on three publicly available datasets demonstrated that the proposed method obtained superior SR results, outperforming state-of-the-art SR algorithms. © 2008-2012 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Lightweight Dual Grouped Large-Kernel Convolutions for Salient Object Detection Network 31st

Lightweight Dual Grouped Large-Kernel Convolutions for Sali...

引用

31st International Conference on Multimedia Modeling, MMM 2025

作者： Liu, Jiajie Zhang, Zhibin Engineering Research Center of Ecological Big Data Ministry of Education Beijing China Key Laboratory of Wireless Networks and Mobile Computing Inner Mongolia University Hohhot010021 China

ISBN: (纸本)9789819620609

Most existing Salient Object Detection (SOD) methods focus on achieving better performance, often resulting in models with a large number of parameters. However, there is limited research on lightweight models in this field. To address this gap, our goal is to maintain performance while reducing the number of network parameters. Thanks to the development of large-kernel convolutions in recent years, we have improved the U2Net as the base network by adding our lightweight dual large-kernel fusion module. Our module better utilizes the depth information of U2Net and, due to the large receptive field of large-kernel convolutions, better captures the relationships between image elements. This allows our network to remain lightweight while maintaining performance. We designed a large-kernel (DLK) fusion module and a lightweight dual grouped large-kernel Unet network (DGLKUNET). Our network uses SRUS (an improved Residual U-blocks module, RUS) to construct DGLKUNET that predicts image contours and labels. Compared to the base RUS network (U2Net), our network reduces the number of parameters by up to 70% and the computational cost by 40% while maintaining performance. Evaluation results on five datasets demonstrate the performance of our lightweight network. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Object recognition

来源：评论

学校读者我要写书评

暂无评论

Listen With Seeing: Cross-Modal Contrastive Learning for Audio-Visual Event Localization

引用

IEEE Transactions on Multimedia 2025年 27卷 2650-2665页

作者： Sun, Chao Chen, Min Zhu, Chuanbo Zhang, Sheng Lu, Ping Chen, Jincai Huazhong University of Science and Technology Wuhan National Laboratory for Optoelectronics Key Laboratory of Information Storage System Engineering Research Center of Data Storage Systems and Technology Ministry of Education of China School of Computer Science and Technology Wuhan430074 China South China University of Technology School of Computer Science and Engineering Guangzhou510640 China Pazhou Laboratory Guangzhou510640 China

In real-world physiological and psychological scenarios, there often exists a robust complementary correlation between audio and visual signals. Audio-Visual Event Localization (AVEL) aims to identify segments with Audio-Visual Events (AVEs) that contain both audio and visual tracks in unconstrained videos. Prior studies have predominantly focused on audio-visual cross-modal fusion methods, overlooking the fine-grained exploration of the cross-modal information fusion mechanism. Moreover, due to the inherent heterogeneity of multi-modal data, inevitable new noise is introduced during the audio-visual fusion process. To address these challenges, we propose a novel Cross-modal Contrastive Learning Network (CCLN) for AVEL, comprising a backbone network and a branch network. In the backbone network, drawing inspiration from physiological theories of sensory integration, we elucidate the process of audio-visual information fusion, interaction, and integration from an information-flow perspective. Notably, the Self-constrained Bi-modal Interaction (SBI) module is a bi-modal attention structure integrated with audio-visual fusion information, and through gated processing of the audio-visual correlation matrix, it effectively captures inter-modal correlation. The Foreground Event Enhancement (FEE) module emphasizes the significance of event-level boundaries by elongating the distance between scene events during training through adaptive weights. Furthermore, we introduce weak video-level labels to constrain the cross-modal semantic alignment of audio-visual events and design a weakly supervised cross-modal contrastive learning loss (WCCL Loss) function, which enhances the quality of fusion representation in the dual-branch contrastive learning framework. Extensive experiments conducted on the AVE dataset for both fully supervised and weakly supervised event localization, as well as Cross-Modal Localization (CML) tasks, demonstrate the superior performance of our model compa

关键词： Information fusion

来源：评论

学校读者我要写书评

暂无评论

Revocable Hierarchical Identity-based Inner Product Function Encryption in Smart Healthcare

引用

IEEE Internet of Things Journal 2025年第11期12卷 15319-15332页

作者： Zhu, Yasi Zhou, Yanwei Wang, Jing Yang, Bo Zhang, Mingwu Shaanxi Normal University School of Computer Science Xi'an710119 China Sichuan University Key Laboratory of Data Protection and Intelligent Management Ministry of Education Chengdu610065 China Hubei University of Technology School of Computer Wuhan430068 China

With the development of cloud computing and the digital transformation of the medical industry, the application scenarios and effects of smart healthcare are constantly expanding and improving. Smart healthcare plays an important role in improving service quality and medical efficiency and reducing medical costs. However, in the process of data collection, storage, and transmission, the risk of patients' privacy being illegally accessed or leaked has become increasingly prominent. These medical data contain sensitive information about patients, and once leaked, it will violate the privacy of patients, cause medical accidents, and seriously damage the legitimate rights and interests of patients. An efficient and reliable privacy protection mechanism is the foundation and key to establishing a harmonious doctor-patient relationship and improving medical quality. Although traditional encryption methods can provide certain information security protection, they cannot balance data protection and data analysis and processing and are not suitable for intelligent medical environments that require more precise medical decisions. Therefore, we propose a hierarchical identity-based inner product functional encryption scheme with a malicious user revocation mechanism aimed at addressing the privacy and security issues of patients in smart healthcare. This solution allows specific access and analysis of data while ensuring that patients' sensitive information is protected from infringement, enabling doctors to diagnose and treat more accurately and optimize the allocation of medical resources to the greatest extent possible. Furthermore, we have formally demonstrated the security of the scheme and conducted a security analysis and performance comparison. The results show that our solution has better performance while ensuring security and is suitable for the efficient and secure data processing requirements of smart healthcare. © 2014 IEEE.

关键词： Differential privacy

来源：评论

学校读者我要写书评

暂无评论

KAN v.s. MLP for Offline Reinforcement Learning

KAN v.s. MLP for Offline Reinforcement Learning

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Guo, Haihong Li, Fengxin Li, Jiao Liu, Hongyan School of Information Renmin University of China China Institute of Medical Information Medical Library Chinese Academy of Medical Sciences Peking Union Medical College China Key Laboratory of Data Engineering and Knowledge Engineering Ministry of Education China School of Economics and Management Tsinghua University China

ISBN: (纸本)9798350368741

Kolmogorov-Arnold Networks (KAN) is an emerging neural network architecture in machine learning. It has greatly interested the research community about whether KAN can be a promising alternative to the commonly used Multi-Layer Perceptions (MLP). Experiments in various fields demonstrated that KAN-based machine learning can achieve comparable if not better performance than MLP-based methods, but with much smaller parameter scales and are more explainable. In this paper, we explore the incorporation of KAN into the actor and critic networks for offline reinforcement learning (RL). We evaluated the performance, parameter scales, and training efficiency of various KAN and MLP-based conservative Q-learning (CQL) on the classical D4RL benchmark for offline RL. Our study demonstrates that KAN can achieve performance close to the commonly used MLP with significantly fewer parameters. This allows us to choose the base networks according to the offline RL task requirements. © 2025 IEEE.

关键词： KAN Kolmogorov-Arnold networks MLP multilayer perceptrons offline reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

DSDIR: A Two-Stage Method for Addressing Noisy Long-Tailed Problems in Malicious Traffic Detection

DSDIR: A Two-Stage Method for Addressing Noisy Long-Tailed P...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Li, Guoliang Zhang, Ruiqi Sun, Zhe Xing, Lingkai Zhang, Yu College of Cyber Science Nankai University Tianjin China College of Computer Science Nankai University Tianjin China Tianjin Key Laboratory of Network and Data Security Technology Tianjin China Key Laboratory of Data and Intelligent System Security Ministry of Education Tianjin China

ISBN: (纸本)9798350368741

In recent years, deep learning based malicious traffic detection (MTD) systems have demonstrated remarkable success. However, their effectiveness tend to decrease because most malicious traffic datasets are suffered from noisy-labeled and long-tailed problems. While numerous approaches have been developed to address these two problems individually, they become inefficient when confronting the combined challenge of noisy long-tailed data, as they typically tackle only a single adverse factor at a time. This paper proposes a two-stage method called Distribution-aware sample Selection and Dynamic Instance-based Relabeling (DSDIR), which simultaneously addresses the impacts of noisy-labeled and long-tailed problems. In the first stage, a noise-independent clean sample selection method is designed to obtain a clean dataset, which converts negative effects of the long-tailed problem into positive ones. In the second stage, dynamic instance-based relabeling is designed to train a model and improve the dataset's quality simultaneously. Eventually, DSDIR not only produces a balanced and noise-tolerant model but also obtains a clean dataset. Experimental results demonstrate that in the high noise condition of 60% and 80%, the accuracy rate of DSDIR is 5% higher than the state-of-the-art methods. Our code is available at https://***/nku-ligl/DSDIR. © 2025 IEEE.

关键词： deep learning distribution estimation dynamic relabeling malicious traffic detection noisy long-tailed problem

来源：评论

学校读者我要写书评

暂无评论

Geomorphological evolution of small watershed on the Chinese Loess Plateau based on slope shape spectra

引用

Earth Science Informatics 2025年第2期18卷 1-23页

作者： Jiang, Hongtao Chen, Nan Wang, Chen Li, Sijia Ou, Mengyao Key Laboratory of Spatial Data Mining & Information Sharing Ministry of Education Fuzhou University Fuzhou 350116 China The Academy of Digital China (Fujian) Fuzhou University Fuzhou 350116 China

The Chinese Loess Plateau, the world's largest loess deposit, is characterized by distinctive landforms and severe soil erosion. Limited long-term observational data has hindered a comprehensive understanding of watershed evolution in this region. This study uses laboratory-based artificial rainfall simulations to investigate the long-term geomorphological evolution of watersheds on the Chinese Loess Plateau. Digital elevation models and terrain analysis techniques were employed to analyze geomorphological processes. Through correlation analysis and Sheffield entropy calculations, an optimal combination of terrain factors—including slope, aspect, plan curvature, profile curvature, and elevation—was identified to accurately describe slope characteristics. Multi-resolution image segmentation, Moran's I, and weighted variance were used to optimize segmentation parameters, achieving precise slope unit segmentation. Based on these parameters, we classified slope shapes, constructed slope shape spectra, and analyzed their changes over time. Results indicate that as the watershed evolves, the entropy of slope spectra increases while skewness and kurtosis decrease, indicating greater geomorphological complexity and diversity. The watershed's evolution progresses through three stages: slope shape development, slope shape growth, and slope shape maturity, each marked by distinct geomorphological features. By analyzing landscape indices at both class and landscape levels, we observed significant variation during the early development stage, followed by stabilization during the maturation phase. These findings demonstrate that slope shape spectra effectively capture the dynamic evolution of slopes in the Chinese Loess Plateau, offering new insights for understanding its geomorphological evolution and providing a novel method for quantitative geomorphology. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.

关键词： Evolution Loess landform Segmentation Simulated watershed Slope shape spectrum

来源：评论

学校读者我要写书评

暂无评论

Dual-Antenna GNSS/MEMS INS Tightly Coupled Algorithm for Agricultural Machinery Based on Adaptive Federated Filtering

引用

IEEE Sensors Journal 2025年第9期25卷 14780-14792页

作者： Feng, Yihang Huang, Guanwen Li, Xin Li, Zhenhong Zhang, Kai Li, Hang Jing, Ce Chang’an University College of Geological Engineering and Geomatics Xi’an710054 China Chang’an University College of Geological Engineering and Geomatics Key Laboratory of Ecological Geology and Disaster Prevention Ministry of Natural Resources Xi’an710054 China Universities of Shaanxi Province Engineering Research Center of Intelligent BDS Xi’an710054 China Chang’an University College of Geological Engineering and Geomatics Key Laboratory of Western China’s Mineral Resource and Geological Engineering Ministry of Education Big Data Center for Geosciences and Satellites Xi’an710054 China

Modern agricultural machinery relies on high-accuracy navigation systems;however, the common loosely coupled (LC) solution of dual-antenna global navigation satellite system (GNSS) and micro-electromechanical system inertial navigation system (MEMS INS) often fails to meet accuracy requirements in complex environments. Theoretically, the tightly coupled (TC) solution of the dual-antenna baseline constraint and MEMS INS offers better attitude accuracy. However, its state space is incomplete, comprising only attitude, gyro biases, and ambiguity. Moreover, previous studies have not conducted a state observability analysis on this model, which is essential for understanding its state estimation capabilities. Therefore, we derived the TC model of dual-antenna baseline constraint and MEMS INS within a complete state space and performed an observability analysis. Based on these results and considering computational efficiency, we integrated this model into the GNSS/MEMS INS TC model using federated filtering. To further improve the algorithm’s accuracy in complex agricultural environments, an adaptive robust positioning algorithm is proposed based on turning state detection. The proposed algorithm was validated through three sets of experiments, demonstrating position accuracy within 2 cm in both open and slightly occluded environments, with heading accuracy within 0.6°, and maintaining optimal accuracy even in severely occluded environments. © 2001-2012 IEEE.

关键词： Adaptive filtering

来源：评论

学校读者我要写书评

暂无评论

Auxiliary Tasks Benefit Skeleton-based Action Recognition

Auxiliary Tasks Benefit Skeleton-based Action Recognition

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Yang, Yuheng Chen, Haipeng College of Computer Science and Technology Key Laboratory of Symbolic Computation and Knowledge Engineering Ministry of Education Jilin University Changchun China

ISBN: (纸本)9798350368741

Skeleton-based action recognition has long been a fundamental and intriguing problem in machine intelligence. This task is challenging due to pose occlusion and rapid motion, which typically results in incomplete or noisy skeleton data. State-of-the-art methods tend to learn human motion directly from these corrupted skeletons as if they were reliable. Unfortunately, this might lead to unsatisfactory results when key regions of the skeleton are occluded or disturbed. To tackle the problem, we propose a novel framework that integrates auxiliary tasks into a motion modeling network. These auxiliary tasks corrupt partial human skeletons with masking or noise and then force the network to recover the corrupted data, explicitly facilitating robust feature representation learning. We further propose supervising the auxiliary tasks with mutual information losses, mathematically ensuring feature consistency and spatial alignment between the recovered and original skeleton data. Empirically, our approach sets the new state-of-the-art performance on three benchmark datasets. © 2025 IEEE.

关键词： auxiliary task mutual information skeleton-based action recognition

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：