检索结果-内蒙古大学图书馆

Spectral analysis of bone-conducted speech using modified linear prediction

International Journal of Speech Technology 2024年第4期27卷 1039-1053页

作者： Ohidujjaman Hasan, Mahmudul Zhang, Shiming Huda, Mohammad Nurul Uddin, Mohammad Shorif Computer Science and Engineering Daffodil International University Dhaka1216 Bangladesh Computer Science and Engineering Comilla University Comilla3506 Bangladesh School of Electrical and Information Northeast Agricultural University Harbin150030 China Computer Science and Engineering United International University Dhaka1212 Bangladesh Computer Science and Engineering Green University of Bangladesh Kanchon1460 Bangladesh Computer Science and Engineering Jahangirnagar University Savar1342 Bangladesh

This paper improves the performance of linear prediction (LP) in precise spectral estimation of bone-conducted (BC) speech. Inherently, BC speech contains a wide spectral dynamic range that causes ill conditioning in the autocorrelation (ACR) method and its variants, where the Levinson–Durbin (L–D) algorithm is commonly implemented. Instead of the conventional LP-based spectral estimation methods, we utilize the covariance-based method, specifically the modified covariance (MC) method, where the orthogonal decomposition algorithm is deployed. In this paper, we derive the MC method from the least squares (LS) technique for BC speech analysis. The MC method reduces the eigenvalue expansion that compresses the spectral dynamic range of the BC speech signal. The effect of spectral dynamic range compression declines the ill-conditioned properties of LP. Through the proposed method using synthetic BC speech, the resulting power spectrum provides more accurate peaks than the conventional methods. The validity of the proposed method is also analyzed by inspecting real BC speech. This study reveals the utmost use of BC speech in speech processing systems. The experimental results demonstrate that the proposed method provides more accurate spectral estimation for synthetic and real BC speeches compared with conventional spectral estimation methods. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Autocorrelation

来源：评论

学校读者我要写书评

暂无评论

HandRT: Simultaneous Hand Shape and Appearance Reconstruction with Pose Tracking from Monocular RGB-D Video

引用

IEEE Transactions on Pattern Analysis and Machine Intelligence 2025年第07期47卷 5625-5635页

作者： Kalshetti, Pratik Chaudhuri, Parag Indian Institute of Technology Bombay Department of Computer Science and Engineering India

We propose a method to reconstruct a personalized hand avatar, representing the user's hand shape and appearance, from a monocular RGB-D video of a hand performing unknown hand poses under unknown illumination. Our method, HandRT, jointly optimizes hand pose, shape, appearance, and lighting parameters using a physically-based shading model in a differentiable rendering framework incorporating Monte Carlo path tracing. HandRT extends our previous work, Intrinsic Hand Avatar, by relaxing the assumption of a known coarse hand pose and utilizing depth data in the optimization. Specifically, we introduce an articulated registration energy based on iterative closest point over the depth point cloud that enables reconstruction from unknown hand poses via tracking across frames. Thus, we can reconstruct the avatar from arbitrary poses with high accuracy without relying on an off-the-shelf 2D joint detector at each frame. Further, HandRT is capable of precisely tracking the reconstructed avatar from either RGB or RGB-D input. Our evaluation demonstrates that our method outperforms existing hand avatar reconstruction methods on all commonly used metrics while producing significantly accurate mesh compared to state-of-the-art hand mesh recovery methods by a large margin on public and our captured datasets. To facilitate future research, we release our code at https://***/pmkalshetti/handrt © 1979-2012 IEEE.

关键词： Mesh generation

来源：评论

学校读者我要写书评

暂无评论

Multi-feature Fusion Deep Network for Skin Disease Diagnosis

引用

Multimedia Tools and Applications 2025年第1期84卷 419-444页

作者： Gairola, Ajay Krishan Kumar, Vidit Sahoo, Ashok Kumar Diwakar, Manoj Singh, Prabhishek Garg, Deepak Department of Computer Science and Engineering Graphic Era Hill University Dehradun India Department of Computer Science and Engineering Graphic Era deemed to be university Dehradun India School of Computer Science Engineering and Technology Bennett University Greater Noida India School of Computer Science and Artificial Intelligence SR University Warangal Telangana Warangal506371 India

The skin acts as an important barrier between the body and the external environment, playing a vital role as an organ. The application of deep learning in the medical field to solve various health problems has generated increasing interest. The objective of this research is to create a new model using a deep network with multi-feature fusion that enables accurate identification of various skin diseases. This paper proposes a fully fused network (FFN), which includes an improved single block (ISB) and an improved fusion block (IFB) to achieve optimal performance. To implement this, a convolutional neural network-based model for multi-class recognition of skin images has been developed, in which ISB is used to segment diseases in skin images. The IFB module is designed to enhance the effectiveness of the fused network. The model is built and evaluated using our proprietary dataset (Skin_disease_v1) along with publicly available datasets, namely ISIC2016, ISIC2017, and HAM10000. The highest accuracy achieved in experiments was 86% for ISB, 90% for IFB, and 92% for FFN using HAM10000 (for ResNet101V2), HAM10000 (for Resnet50 + ResNet101V2), and HAM10000 (for Resnet50 + ResNet101V2). This shows an accuracy improvement of 9.2%, 13.2%, and 15.2% compared to the state-of-the-art ISB, IFB, and FFN approaches, respectively. Our network outperforms existing networks in terms of performance. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

SMILES2DTA: a CNN-based approach for identifying drug candidates and predicting drug-target binding affinity

引用

Neural Computing and Applications 2024年第4期37卷 2891-2910页

作者： Mukit, Hasanul Hossain, Sayeed Farabi, Mirza Milan Chowdhury, Mehrab Zaman Pritom, Ahmed Iqbal Rana, Humayan Kabir Department of Computer Science and Engineering Green University of Bangladesh Narayanganj-1461 Dhaka Bangladesh Department of Computer Science The University of Alabama at Birmingham BirminghamAL United States Department of Computer Science and Engineering Islamic University of Technology Gazipur Bangladesh

Computational approaches can speed up the drug discovery process by predicting drug-target affinity, otherwise it is time-consuming. In this study, we developed a convolutional neural network (CNN)-based model named SMILES2DTA (Simplified Molecular Input Line Entry System to Drug-Target Affinity) for predicting the binding affinity between a drug and a target protein. The model utilizes CNNs to process sequences of both drug SMILES and target proteins. SMILES2DTA generates multiple sequences from a single drug SMILES sequence, validates them based on Lipinski’s rule of five, and assesses their binding affinity against a target protein sequence. We evaluated our model using publicly available datasets and compared its performance to state-of-the-art methods. The results showed that SMILES2DTA outperformed the existing methods and demonstrated improved accuracy, mean squared error, and area under the precision-recall curve. SMILES2DTA has the potential to speed up the drug discovery process by reducing the time and cost complexity of identifying effective drugs. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Drug discovery

来源：评论

学校读者我要写书评

暂无评论

Drawing a parallel between the trend of confirmed COVID-19 deaths in the winters of 2022/2023 and 2023/2024 in Italy, with a prediction

引用

Mathematical Biosciences and engineering 2024年第3期21卷 3742-3754页

作者： Roccetti, Marco Department of Computer Science and Engineering University of Bologna Bologna Italy

We studied the weekly number and the growth/decline rates of COVID-19 deaths of the period from October 31, 2022, to February 9, 2023, in Italy. We found that the COVID-19 winter wave reached its peak during the three holiday weeks from December 16, 2022, to January 5, 2023, and it was definitely trending downward, returning to the same number of deaths as the end of October 2022, in the first week February 2023. During this period of 15 weeks, that wave caused a number of deaths as large as 8,526. Its average growth rate was +7.89% deaths per week (10 weeks), while the average weekly decline rate was -15.85% (5 weeks). At the time of writing of this paper, Italy has been experiencing a new COVID-19 wave, with the latest 7 weekly bulletins (October 26, 2023 – December 13, 2023) showing that deaths have climbed from 148 to 322. The weekly growth rate had risen by +14.08% deaths, on average. Hypothesizing that this 2023/2024 wave will have a total duration similar to that of 2022/2023, with comparable extensions of both the growth period and the decline period and similar growth/decline rates, we predict that the number of COVID-19 deaths of the period from the end of October 2023 to the beginning of February 2024 should be less than 4100. A preliminary assessment of this forecast, based on 11 of the 15 weeks of the period, has already confirmed the accuracy of this approach. ©2024 the Author(s), licensee AIMS Press.

关键词： COVID-19

来源：评论

学校读者我要写书评

暂无评论

3D brain image based tumor classification using ensemble of reinforcement transfer-based belief neural networks

引用

Multimedia Tools and Applications 2025年第14期84卷 14001-14028页

作者： Arora, Shraddha Lamba, Monika Department of Computer Science and Engineering The NorthCap University Haryana Gurugram India

The procedure of segmenting a brain tumour is an essential step in the field of medical image processing. In order to improve the efficacy of disease treatment choices and the likelihood of patient survival, the timely detection of a brain tumour is not only highly essential but a vital factor. For the purpose of cancer diagnosis, manually segmenting brain tumours from the large number of MRI images that are generated in clinical practice is a hard and time-consuming process. Automatic brain tumour image segmentation is a technique that absolutely has to exist. Using deep learning and machine learning architectures for segmentation and feature extraction, this research proposes a novel approach to the detection of brain tumours. In this case, the input consists of a collection of 3-D brain slice pictures that have been smoothed and with noise eliminated. After that, the picture that has been processed is segmented by using a convolutional radial function architecture that centres on contours. The segmented image features retrieval has been accomplished by the use of an ensemble of reinforcement transfer-based belief neural networks. The results of the experimental study have been completed with regard to precision, recall, accuracy, F-1 score, area under the curve, dice coefficient, and root mean square error. The strategy that was recommended was successful in achieving 97.57% accuracy, 97.15% precision, 89.37% recall, 83.56% F-1 score, 69.59% area under the curve, 63.83% dice coefficient, and 55.56% root mean square error. The sagittal, coronal, and axial planes of the pictures were shown in order to acquire the specific position of the tumour as well as its characteristics, which included its height, breadth, and depth. When it comes to the prediction of tumours, the findings of the assessment of semantic segmentation based on deep learning networks are highly positive. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Spri

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Enhanced prediction of agricultural CO2 emission using ensemble machine learning-based imputation approach

引用

Neural Computing and Applications 2024年第36期36卷 22867-22887页

作者： Thendral, M. Sathya Abinaya, S. Devi, M. K. Kavitha Department of Computer Science and Engineering Thiagarajar College of Engineering Madurai625015 India School of Computer Science and Engineering Vellore Institute of Technology Chennai600127 India

The agricultural sector contributes significantly to greenhouse gas emissions, which cause global warming and climate change. Numerous mathematical models have been developed to predict the greenhouse gas emissions from agriculture. However, the database utilized for prediction has thousands of missing values and imbalanced data records due to various factors, such as environmental disasters, sensor failure, maintenance issues, and budgetary constraints. Many researchers have either completely discarded records with missing values or have only partially addressed the issue, leading to less precise predictions. This study proposes a machine learning-based ensemble approach that uses multiple imputation by chained equations (MICE), K-nearest neighbors (KNN), and MissForest techniques to impute the missing values by reducing uncertainties and improves robustness in predictions. Furthermore, Synthetic Minority Oversampling Technique (SMOTE R) addresses data imbalance in CO2 emissions prediction by generating synthetic data to balance the target variable's distribution in the regression problem. Thereby the proposed SMOTE R and ensemble imputation approach aims to improve prediction accuracy and reliability by tackling data completeness and distribution issues simultaneously. The experiments are conducted on the GRACEnet database shows that the proposed approach outperforms the existing work in measure of R2, MAE, and RMSE metrics. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Greenhouse gas emissions

来源：评论

学校读者我要写书评

暂无评论

Privacy-preserving intrusion detection with federated learning: enhancing privacy and efficiency in IoT networks

引用

International Journal of Machine Learning and Cybernetics 2025年 1-19页

作者： Sharma, Hemant Yadav, Gyan Singh Department of Computer Science and Engineering IIIT Kota Rajasthan Kota India

Intrusion detection systems (IDS) are crucial in the identification of unauthorized activities on a digital network, enabling cybersecurity measures to initiate prevention protocols to protect the security of their networks and data. With Internet of Things (IoT) as a ubiquitous phenomenon, IDS has become increasingly important owing to the massive growth of connected devices. Machine learning algorithms can detect anomalies in large datasets, but centralized methods pose security risks by gathering all data in one place. Federated learning (FL) offers a safer alternative, allowing multiple clients to collaborate on a model without sharing data, reducing the risk of data leaks and enhancing privacy. In this paper, we propose a FL-based IDS. Herein, we have applied the FL approach to a number of clients that are grouped in sets where the number of clients in a set varies from 2 to 10. After that, we applied feature ranking via a random forest approach, wherein the features are ranked based on their importance to find the optimal and reduced dataset for reducing the inference time. Further, three models for each of the clients were trained using the refined dataset obtained. Next, the central server aggregates the data from the clients using three methods. We have updated the dataset with two types of scenarios, one with two classes (Benign and Attacks) and the second with a multi-class (Benign, Dodag, Flooding, Rank, Blackhole) for identifying the attack. The experimental results establish that the proposed methodology achieves superior performance as compared to similar existing methodologies. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

A systematic mapping to investigate the application of machine learning techniques in requirement engineering activities

引用

CAAI Transactions on Intelligence Technology 2024年第6期9卷 1412-1434页

作者： Shoaib Hassan Qianmu Li Khursheed Aurangzeb Affan Yasin Javed Ali Khan Muhammad Shahid Anwar School of Computer Science and Engineering Nanjing University of Science and TechnologyNanjingJiangsuChina Department of Computer Engineering College of Computer and Information SciencesKing Saud UniversityRiyadhSaudi Arabia School of Software Northwestern Polytechnical UniversityXianShaanxiChina Department of Computer Science School of PhysicsEngineering&Computer ScienceUniversity of HertfordshireHatfieldUK Department of AI and Software Gachon University Seongnam-siSeongnamSouth Korea

Over the past few years,the application and usage of Machine Learning(ML)techniques have increased exponentially due to continuously increasing the size of data and computing *** the popularity of ML techniques,only a few research studies have focused on the application of ML especially supervised learning techniques in Requirement engineering(RE)activities to solve the problems that occur in RE *** authors focus on the systematic mapping of past work to investigate those studies that focused on the application of supervised learning techniques in RE activities between the period of 2002–*** authors aim to investigate the research trends,main RE activities,ML algorithms,and data sources that were studied during this ***-five research studies were selected based on our exclusion and inclusion *** results show that the scientific community used 57 *** those algorithms,researchers mostly used the five following ML algorithms in RE activities:Decision Tree,Support Vector Machine,Naïve Bayes,K-nearest neighbour Classifier,and Random *** results show that researchers used these algorithms in eight major RE *** activities are requirements analysis,failure prediction,effort estimation,quality,traceability,business rules identification,content classification,and detection of problems in requirements written in natural *** selected research studies used 32 private and 41 public data *** most popular data sources that were detected in selected studies are the Metric Data Programme from NASA,Predictor Models in Software engineering,and iTrust Electronic Health Care System.

关键词： data sources machine learning requirement engineering supervised learning algorithms

来源：评论

学校读者我要写书评

暂无评论

M22: A Communication-Efficient Algorithm for Federated Learning Inspired by Rate-Distortion

引用

IEEE Transactions on Communications 2024年第2期72卷 845-860页

作者： Liu, Yangyi Rini, Stefano Salehkalaibar, Sadaf Chen, Jun McMaster University Department of Electrical and Computer Engineering HamiltonONL8S 4L8 Canada Department of Electrical and Computer Engineering Hsinchu30010 Taiwan University of Toronto Department of Electrical and Computer Engineering TorontoONM5A 4T4 Canada

In federated learning (FL), the communication constraint between the remote clients and the Parameter Server (PS) is a crucial bottleneck. For this reason, model updates must be compressed so as to minimize the loss in accuracy resulting from the communication constraint. This paper proposes 'M-magnitude weighted L2 distortion + 2 degrees of freedom' (M22) algorithm, a rate-distortion inspired approach to gradient compression for federated training of deep neural networks (DNNs). In particular, we propose a family of distortion measures between the original gradient and the reconstruction we referred to as ' M-magnitude weighted L{2} ' distortion, and we assume that gradient updates follow an i.i.d. distribution-generalized normal or Weibull, which have two degrees of freedom. In both the distortion measure and the gradient distribution, there is one free parameter for each that can be fitted as a function of the iteration number. Given a choice of gradient distribution and distortion measure, we design the quantizer to minimize the expected distortion in gradient reconstruction. To measure the gradient compression performance under a communication constraint, we define the per-bit accuracy as the optimal improvement in accuracy that one bit of communication brings to the centralized model over the training period. Using this performance measure, we systematically benchmark the choice of gradient distribution and distortion measure. We provide substantial insights on the role of these choices and argue that significant performance improvements can be attained using such a rate-distortion inspired compressor. © 1972-2012 IEEE.

关键词： Deep neural networks

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：