检索结果-内蒙古大学图书馆

31st International Conference on Computational Linguistics, COLING 2025

作者： Hou, Haowen Zeng, Peigen Ma, Fei Yu, Fei Richard Shenzhen China College of Computer Science and Software Engineering Shenzhen University Shenzhen China Shool of Information Technology Carleton University Canada

ISBN: (纸本)9798891761964

Visual Language Models (VLMs) have rapidly progressed with the recent success of large language models. However, there have been few attempts to incorporate efficient linear Recurrent Neural Networks (RNNs) architectures into VLMs. In this study, we introduce VisualRWKV, the first application of a linear RNN model to multimodal learning tasks, leveraging the pre-trained RWKV language model. We propose a data-dependent recurrence and sandwich prompts to enhance our modeling capabilities, along with a 2D image scanning mechanism to enrich the processing of visual sequences. Extensive experiments demonstrate that VisualRWKV achieves competitive performance compared to Transformer-based models like LLaVA-1.5 on various benchmarks. Compared to LLaVA-1.5, VisualRWKV has a speed advantage of 3.98 times and can save 54% of GPU memory when reaching an inference length of 24K tokens. To facilitate further research and analysis, we have made the checkpoints and the associated code publicly accessible at the following GitHub repository: https://***/howard-hou/VisualRWKV. © 2025 Association for Computational Linguistics.

关键词： Recurrent neural networks

来源：评论

学校读者我要写书评

暂无评论

TunSLR-25: A New Static Tunisian Sign Language Recognition System 41st

TunSLR-25: A New Static Tunisian Sign Language Recognition S...

引用

41st computer Graphics International Conference, CGI 2024

作者： Aloui, Sarra Wang, Yu Fan-fan, Zhou College of Computer Science and Software Engineering Hohai University Nanjing China First Affiliated Hospital of Kunming Medical University Kunming China

ISBN: (纸本)9783031818059

There are 466 M deaf across the world according to World Health Organisation. They are deprived of their basic human right which is communicating with others. Many solutions have been developed to help hard-of-hearing people overcome communication barriers. The existing sign language recognition research focuses on sign languages other than Tunisian Sign Language. Consequently, a Tunisian Sign Language dataset, TunSL-D, was developed for this research. The dataset contains static images that are more complex and can generalize the classification tasks. The images represent the most used words and alphabets in Tunisian Sign language. This paper proposes a novel model TunSLR-25 that recognizes and classifies static Tunisian sign language. The model is first trained on the words data, achieving an accuracy of 95.49%, then trained on both words and alphabets, achieving an accuracy of 88%, with a low parameter number of 1,098,901. The TunSL-D has been trained on SLRNet-8 CNN, a 4-layer CNN model, pre-trained models Resnet50, Resnet101, and InceptionV3. The paper compares the performance of these CNN models and the proposed CNN model TunSLR-25. The system will be integrated into a web application designed to be accessible and user-friendly for this community. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Classification SLR Tunisian Sign Language

来源：评论

学校读者我要写书评

暂无评论

Optimizing auditory brainstem response detection through NSGA-II guided feature selection

引用

Neural Computing and Applications 2025年第15期37卷 8925-8943页

作者： Majidpour, Jafar Hassanzadeh, Hiwa Khezri, Edris Arabi, Hossein Department of Software Engineering University of Raparin Kurdistan Region Rania Iraq Department of Computer Science University of Raparin Kurdistan Region Rania Iraq Department of Computer Engineering Islamic Azad University Boukan Branch Boukan Iran Division of Nuclear Medicine and Molecular Imaging Department of Medical Imaging Geneva University Hospital Geneva1211 Switzerland

Hearing-impaired people undergo auditory brainstem response (ABR) testing to assess their peripheral auditory nerve system. Audiologists apply diagnostic labels to ABR data using reference-based indicators such as peak latency, waveform shape, amplitude, and others. ABR test scoring errors may invalidate auditory nerve system integrity results. Machine learning, especially deep learning, may reduce human error in ABR analysis. This work offers a complete methodology to handle ABR analysis problems. First, extract ABR test images from PDF reports to create a dataset. The next step uses Elastic Distortion approach to create high-quality ABR images while conserving data. ResNet-50 deep learning extracts ABR image characteristics in the third phase. The fourth step finds the most relevant features using a trainable multi-objective non-dominated sorting genetic algorithm II optimizer. Finally, ABR images are classified using machine learning algorithms. The proposed model achieved significant performance metrics: an accuracy of 99.46%, precision of 99.33%, specificity of 99.57%, sensitivity of 99.57%, F1-score of 99.43%, and a mean squared error of 0.0054 when using a support vector machine classifier with both augmented and original images. These findings demonstrate the effectiveness of trainable feature selection and high-quality ABR images in improving ABR analysis accuracy. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.

关键词： Mean square error

来源：评论

学校读者我要写书评

暂无评论

MedVQA-APL: enhancing medical visual question answering with assistance prompt learning

MedVQA-APL: enhancing medical visual question answering with...

引用

2024 International Conference on Physics, Photonics, and Optical engineering, ICPPOE 2024

作者： Dai, Linling Lin, Zhi Computer Science and Technology South China Normal University Canton China Software Engineering South China Normal University Canton China

ISBN: (数字)9781510689121

ISBN: (纸本)9781510689114

In order to assist physicians in reducing repetitious chores and lightening their workload, Medical Visual Question Answering (Med-VQA) intends to streamline the process of extracting accurate responses from medical images and related queries. Existing medical visual question answering models are frequently pre-trained on extensive medical datasets before being fine-tuned on smaller, task-specific datasets. Despite the increased parameterization associated with these models, the enhancement in performance is often marginal. We present Medical Visual Question Answering with Assistance Prompt Learning (MedVQA-APL) in this study. It uses an encoder that is rather simple and still has strong performance. Our method starts with the creation of an assistance prompt generator module whose purpose is to generate an assistance prompt that is limited by the intended response. Next, we introduce an assistance prompt fusion module incorporated within a multi-modal fusion block. This module leverages the assistance prompt to derive clinically relevant features. In order to forecast the final answers, the integrated information is finally merged with image-language cross-modal information. Two publicly accessible Med-VQA datasets were used for experimental assessments, which show that MedVQA-APL outperforms the conventional model and produces performance improvements of 2.85% and 5.54% on VQA-RAD and Path-VQA, respectively. © 2025 SPIE.

关键词： Question answering

来源：评论

学校读者我要写书评

暂无评论

BiEfficient: Bidirectionally Prompting Vision-Language Models for Parameter-Efficient Video Recognition 17th

BiEfficient: Bidirectionally Prompting Vision-Language Model...

引用

17th Asian Conference on computer Vision, ACCV 2024

作者： He, Haichen Liu, Weibin Xing, Weiwei School of Computer Science and Technology Beijing Jiaotong University Beijing China School of Software Engineering Beijing Jiaotong University Beijing China

ISBN: (纸本)9789819609079

Vision-language models (VLMs) pre-trained on large-scale image-text pairs have shown great success in various image tasks. However, how to efficiently transfer such powerful VLMs into video domain is still an open problem. Given that full finetuning VLMs for video tasks could be computationally expensive, recent studies turn their focus on parameter-efficient finetuning (PEFT). The great potential of VLMs lies in leveraging the bidirectional semantic connections between the two modalities of vision and language. Nevertheless, most current PEFT methods use the vision-only framework and usually ignore the semantic connections between vision and language. In this paper, we propose a novel method called BiEfficient, which use bidirectional prompting schemes to efficiently transfer the VLM to video recognition task with a small number of tunable parameters: 1) Vision-to-Language: we propose two prompt mechanisms, Pre-Prompt and Post-Prompt, which act before and after the text encoder respectively to generate discriminative video-level text representation for each input video. 2) Language-to-Vision: we propose Word-Guided Visual-Prompt, which enhances the temporal modeling of videos using textual knowledge in an almost parameter-free manner. Experiments on Kinetics-400, UCF-101, HMDB-51 demonstrate that the proposed method can achieve comparable or even better performance to the full finetuning methods with much fewer tunable parameters across closed-set and zero-shot video recognition benchmarks. Code is available here: https://***/WbLiuBJTULab/BiEfficient. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Robust copy-move detection and localization of digital audio based CFCC feature

引用

Multimedia Tools and Applications 2025年第11期84卷 9573-9589页

作者： Wang, Dongyu Li, Xiaojie Shi, Canghong Niu, Xianhua Xiong, Ling Wu, Hanzhou Qian, Qing Qi, Chao School of Computer and Software Engineering Xihua University Chengdu610039 China The College of Computer Science Chengdu University of Information Technology Chengdu610225 China School of Information Shanghai University Shanghai200444 China School of Information Guizhou University of Finance and Economics Guiyang550000 China

Copy-move forgery is a common audio tampering technique in which users copy the contents of one speech and paste them into another region of the same speech signal, thus achieving the effect of tampering with the semantics. To verify the authenticity of the audio, this paper proposes a method to detect and localize audio copy-move forgery based on the cochlear filter of cochlear filter cepstral coefficients (CFCC) feature. The pitch tracking algorithm is used to distinguish the voiced and unvoiced segments in the audio, and then the CFCC coefficients are extracted for each voiced segment. The CFCC feature simulates the entire transmission process of signals in the cochlear basilar membrane using wavelet transformation. Finally, we use Pearson correlation coefficients (PCCs) and dynamic time warping (DTW) in combination to compare the similarity of voiced segments, accurately determining the tampered locations in the audio through threshold judgment. Through extensive experiments on relevant datasets, this algorithm achieves a precision rate of 98.39% and a recall rate of 98.00% in detecting audio without post-processing. Even when detecting audio that has undergone different post-processing, the precision and recall rates remain above 90%. Compared to existing methods, this approach not only achieves precise localization of replicated segments but also demonstrates superior experimental results. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Next-Generation Cybersecurity Solution: A Decentralized Ransomware Recovery Network (DRRN) with Secret Sharing

引用

SN computer Science 2025年第5期6卷 1-22页

作者： Ali, Sijjad Wang, Jia Leung, Victor C. M. Ali, Asad College of Computer Science and Software Engineering Shenzhen University Yuehai Campus Guangdong Shenzhen 518060 China College of Electronics and information Engineering Shenzhen University Yuehai Campus Guangdong Shenzhen 518060 China

The increasing frequency of ransomware attacks, with a reported growth rate of 59%, has exposed significant vulnerabilities in existing cybersecurity measures. Traditional defenses, such as routine data backups and endpoint protection systems, often fall short against sophisticated ransomware that encrypts data and demands exorbitant payments for decryption keys. These limitations are exacerbated by attackers leveraging centralized storage systems and static recovery methods, making data recovery costly, unreliable, and time-consuming. To address this pressing issue, this paper proposes a novel Decentralized Ransomware Recovery Network, which leverages decentralized storage and secret sharing theory to enhance data security and recovery. The proposed scheme introduces a decentralized encryption/decryption framework that distributes data and keys across the network, reducing the economic incentive for attackers by enabling secure and reliable data recovery without ransom payments. At its core, the system utilizes Shamir’s Secret Sharing Scheme, where encryption keys are divided into shares and distributed among nodes. In the event of an attack, these shares are reconstructed to restore access to the encrypted files. Additionally, the proposed scheme incorporates a dynamic key reconfiguration mechanism, where keys are updated after each recovery process to thwart repeated attacks. Through extensive simulations and evaluations under various scenarios, the proposed scheme demonstrates superior performance in terms of recovery efficiency, security resilience, and adaptability, achieving a significant reduction in attack success rates. By offering a robust and preventative defense mechanism, this research highlights the novelty and practicality of the proposed scheme as an effective cybersecurity strategy against ransomware. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.

关键词： Cybersecurity defense Decentralized Ransomware Recovery Network (DRRN) Ransomware Resilience Secret sharing schemes

来源：评论

学校读者我要写书评

暂无评论

Deformable Few-Shot Face Cartoonization via Local to Global Translation

引用

Computational Visual Media 2025年第2期11卷 269-287页

作者： Yang Zhou Shengshu Li Hui Huang College of Computer Science and Software Engineering Shenzhen University Shenzhen China

Cartoonizing portrait images is a stylish and eye-catching application in both computer vision and graphics. We aimed to train a face cartoonization model using very few (e.g., 5–10) style examples. The main difficulty in this challenging task lies in producing stylizations of high quality while preserving the identity of the input, particularly when the style examples contain strong exaggerations. To address this, we propose a novel cross-domain center loss for few-shot generative adversarial network (GAN) adaptation, which forces the distribution of the target domain to be similar to that of the source. We then employ it to solve this few-shot problem along with a two-stage strategy. Stage I generates an intermediate cartoonization for the input, where we first stylize the individual facial components locally and then deform them to mimic the desired exaggeration under the guidance of landmarks. Stage II focuses on global refinement of the intermediate image. First, we adapt a pretrained StyleGAN model using the proposed cross-domain center loss to the target domain defined by a few examples. Subsequently, the intermediate cartoonization from Stage I can be holistically refined through GAN inversion. The generative power of StyleGAN guarantees high image quality, while the local translation and landmark-guided deformation applied to facial components provide high identity fidelity. Experiments show that the proposed method outperforms state-of-the-art few-shot stylization approaches both qualitatively and quantitatively.

关键词： Translation Faces Generative adversarial networks Deformation Shape Feature extraction Visualization Neural style transfer Generators Adaptation models

来源：评论

学校读者我要写书评

暂无评论

A Combine CNN-RNN Based Approach for Augmenting the Performance of Speech Emotions Recognition

引用

Traitement du Signal 2025年第2期42卷 647-662页

作者： Sayed, Nasir Husnain, Ghassan Shoaib, Muhammad Ghadi, Yazeed Yasin Alajmi, Masoud Qahmash, Ayman Department of Computer Science CECOS University of IT and Emerging Science Peshawar25000 Pakistan Department of Computer Science Islamia College Peshawar Peshawar25100 Pakistan Department of Computer Science and Software Engineering Al Ain University Al Ain64141 United Arab Emirates Department of Computer Engineering College of Computers and Information Technology Taif University Taif21944 Saudi Arabia Department of Informatics and Computer Systems King Khalid University Abha61421 Saudi Arabia

Due to the advancement of neural networks and the increasing demand for accurate and real-time Speech Emotion Recognition (SER) in human-computer interactions, it is necessary to compare existing methods and databases in SER in order to arrive at feasible solutions and a complete understanding of this open-ended problem in SER. To detect and recognize the emotions expressed in speech, various techniques have been used in the literature, including well-established speech analysis and classification techniques. These techniques, including speech analysis and classification, have been used to extract emotions from signals. In this study, we propose a novel method for analyzing signals called Wavelet-Scaled Spectrogram which combines the frequency and scale spectrum of a signal using wavelet transform. This method is effective in analyzing signals at different scales and frequency content. In order to train models for speech emotion identification, a large number of handcrafted features and intermediary depictions i.e., frequency-time plot that have traditionally been utilized in data compilation, collection, and analysis. The development of end-to-end models which extract characteristics and learn directly from raw speech signals to improve speech recognition has recently been studied by researchers following the emergence of deep learning. After training and evaluation on the famous speech databases EmoDB, RAVDESS and IEMOCAP, the proposed model is evaluated on various speakers in both speaker-independent and speaker-dependent modes and on a variety of different voices. When advanced preprocessing techniques or data augmentation are omitted from the proposed architecture, the results demonstrate that it can produce products comparable to those produced by the current state of the art. Three concurrent CNN pipelines and a series of modified local features learning blocks (LFLBs) achieved the highest classification accuracy outperforming some advance state-of-the-approa

关键词： Recurrent neural networks

来源：评论

学校读者我要写书评

暂无评论

Enhancing Security in Distributed Drone-Based Litchi Fruit Recognition and Localization Systems

引用

computers, Materials & Continua 2025年第2期82卷 1985-1999页

作者： Liang Mao Yue Li Linlin Wang Jie Li Jiajun Tan Yang Meng Cheng Xiong Guangdong-Hong Kong-Macao Greater Bay Area Artificial Intelligence Application Technology Research Institute Shenzhen Polytechnic UniversityShenzhen518055China School of Computer Science and Software Engineering University of Science and Technology LiaoningAnshan114051China

This paper introduces an advanced and efficient method for distributed drone-based fruit recognition and localization, tailored to satisfy the precision and security requirements of autonomous agricultural operations. Our method incorporates depth information to ensure precise localization and utilizes a streamlined detection network centered on the RepVGG module. This module replaces the traditional C2f module, enhancing detection performance while maintaining speed. To bolster the detection of small, distant fruits in complex settings, we integrate Selective Kernel Attention (SKAttention) and a specialized small-target detection layer. This adaptation allows the system to manage difficult conditions, such as variable lighting and obstructive foliage. To reinforce security, the tasks of recognition and localization are distributed among multiple drones, enhancing resilience against tampering and data manipulation. This distribution also optimizes resource allocation through collaborative processing. The model remains lightweight and is optimized for rapid and accurate detection, which is essential for real-time applications. Our proposed system, validated with a D435 depth camera, achieves a mean Average Precision (mAP) of 0.943 and a frame rate of 169 FPS, which represents a significant improvement over the baseline by 0.039 percentage points and 25 FPS, respectively. Additionally, the average localization error is reduced to 0.82 cm, highlighting the model’s high precision. These enhancements render our system highly effective for secure, autonomous fruit-picking operations, effectively addressing significant performance and cybersecurity challenges in agriculture. This approach establishes a foundation for reliable, efficient, and secure distributed fruit-picking applications, facilitating the advancement of autonomous systems in contemporary agricultural practices.

关键词： Objective detection deep learning machine learning

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：