检索结果-内蒙古大学图书馆

A Dual-input Electromagnetic Inverse Scattering Algorithm Based on Improved U-net

Applied Computational Electromagnetics Society Journal 2024年第11期39卷 961-969页

作者： Yang, Chun Xia Meng, Jun Jie Wei, Shuang Tong, Mei Song Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal University Shanghai200234 China Department of Electronic Science and Technology Tongji University Shanghai201804 China

In this paper, we propose a dual-input inversion method based on deep learning to improve the accuracy of electromagnetic imaging using the back propagation algorithm (BP). An improved U-Net network is utilized to reconstruct the scatterers. Unlike other deep learning inversion methods, we input both the scatterer distribution data from BP imaging and the scattered field data received by the antennas into the neural network for training. This approach leads to a more accurate prediction of scatterer positions and characteristics. Compared to predicting the scatterers using only the scattered field as input, adding the BP imaging results at the input provides the neural network with more information, significantly reduces the learning difficulty, minimizes errors, and enhances the quality of imaging. To address potential gradient vanishing and spatial information loss during network training, we integrate attention mechanisms and residual modules into the basic U-Net network. The former helps the network extract important relevant information under different contrast conditions, while the latter focuses on solving the problems of gradient vanishing and explosion. Simulation experiments confirm that our dual-input inversion method significantly reduces the average error, outperforming traditional single-input reconstruction methods. © ACES.

关键词： Backpropagation

来源：评论

学校读者我要写书评

暂无评论

Improving low-complexity and real-time DeepFilterNet2 for personalized speech enhancement

引用

International Journal of Speech Technology 2024年第2期27卷 299-306页

作者： Wang, Shilin Guan, Haixin Wei, Shuang Long, Yanhua Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal University Shanghai200234 China Unisound AI Technology Co. Ltd Beijing China

DeepFilterNet2, a recently proposed real-time and low-complexity speech enhancement (SE) technique, has shown state-of-the-art SE performance in many deep noise suppression tasks. This paper proposes a new approach, termed pDeepFilterNet2, to generalize and improve the original DeepFilterNet2 for personalized speech enhancement (PSE) tasks under multi-talker noisy and reverberant conditions. Three improvements are investigated: a frame-wise speaker adaptation (FSA) is first proposed to achieve dynamic target speaker cues for generalizing the DeepFilterNet2 to pDeepFilterNet2;Then, we remove a redundant skip connection in original DeepFilterNet2 and add a simple layer of causal multi-head self-attention to enhance the model for aggregating global context information;Finally, a multi-domain loss function combing both time and frequency domain losses is introduced to further improve the PSE system performance. Moreover, different types of target speaker embedding are also investigated in this study to see their effectiveness. Our experiments are conducted on the DNS4 Challenge dataset and results show that the proposed pDeepFilterNet2 outperforms the original DeepFilterNet2 significantly across multiple evaluation metrics. Furthermore, it exhibits competitive performance when compared to our previously proposed sDPCCN, which was specifically designed for target speaker extraction. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.

关键词： Speech enhancement

来源：评论

学校读者我要写书评

暂无评论

Off-Grid Sparse Bayesian Inference with Biased Total Grids for Dense Time Delay Estimation

引用

Journal of shanghai Jiaotong university(Science) 2023年第6期28卷 763-771页

作者：魏爽李文瑶苏颖刘睿 College of Information Mechanical and Electrical EngineeringShanghai Engineering Research Center of Intelligent Education and BigdataShanghai Normal UniversityShanghai 200234China

For dense time delay estimation(TDE),when multiple time delays are located within a grid interval,it is dificult for the existing sparse Bayesian learning/inference(SBL/SBI)methods to obtain high estimation accuracy to meet the application *** solve this problem,this paper proposes a method named off-grid sparse Bayesian inference-biased total grid(OGSBI-BTG),where a mesh evolution process is conducted to move the total grids iteratively based on the position of the off-grid between two *** proposed method updates the off-grid dictionary matrix by further reconstructing an optimum mesh and offsetting the off-grid *** results demonstrate that the proposed approach performs better than other state-of-the-art SBI methods and multiple signal classification even when the grid interval is larger than the gap of true time *** this paper,the time domain model and frequency domain model of TDE are studied.

关键词： off-grid sparse Bayesian inference(SBI) time delay estimation(TDE) biased total grids(BTG)

来源：评论

学校读者我要写书评

暂无评论

A Machine Learning-Based Parameterized Tropical Cyclone Precipitation Model

引用

International Journal of Disaster Risk Science 2024年第6期15卷 972-985页

作者： Yi Lu Jie Yin Peiyan Chen Hui Yu Sirong Huang Shanghai Typhoon Institute China Meteorological AdministrationShanghai 200030China School of Geographic Sciences East China Normal UniversityShanghai 200241China Asia-Pacific Typhoon Collaborative Research Center Shanghai 200030China Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal UniversityShanghai 200030China

Current simulation models considerably underestimate local-scale,short-duration extreme precipitation induced by tropical cyclones(TCs).This problem needs to be addressed to establish active response policies for TC-induced *** shanghai,a coastal megacity,as a study area and based on the observations from 192 meteorological stations in the city during 2005–2018,this study optimized the parameterized Tropical Cyclone Precipitation Model(TCPM)initially designed for TCs at the national scale(China)to the local or regional scales by using machine learning(ML)methods,including the random forest(RF),extreme gradient boosting(XGBoost),and ensemble learning(EL)*** TCPM-ML was applied for multiple temporal scale hazard *** results show that:(1)The TCPM-ML not only improved TCPM performance for simulating hourly extreme precipitations,but also preserved the physical meaning of the results,contrary to ML methods;(2)Machine learning algorithms enhanced the TCPM ability to reproduce observations,although the hourly extreme precipitations remained slightly underestimated;(3)Best performance was obtained with the XGBoost or EL *** the strengths of both XGBoost and RF,the EL algorithm yielded the best overall *** study provides essential model support for TC disaster risk assessment and response at the local and regional scales in China.

关键词： Extreme precipitation Machine learning Parameterized model-shanghai Tropical cyclone

来源：评论

学校读者我要写书评

暂无评论

Teaching Feedback System Based on ViT Expression Recognition in Distance education 13

Teaching Feedback System Based on ViT Expression Recognition...

引用

13th International Conference on educational and Information Technology, ICEIT 2024

作者： Wang, Kexin Cheng, Meng The College of Information Mechanical and Electrical Engineering The Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal University Shanghai200234 China

ISBN: (纸本)9798350372663

This paper proposes an online emotional feedback system for distance education based on the Vision Transformer (ViT). The objective is to provide teachers with realtime information on students' emotional states, such that the teaching strategies can be properly adapted to improve the quality of education. In the proposed system, the OpenCV is used to call the camera to collect the video in class, and the facial information is then detected by the Dlib. After that, the acquired facial images are processed with the ViT model augmented with two attentive pooling ViT (APViT) modules, to calculate students' emotions during class. Then, the emotional states of the students will be analyzed and repeatedly fed back to the teachers. It has been found from the simulations that the APViT model can achieve a training accuracy of 94.68% with the OL-SFED dataset, exhibiting the advantage of applying the ViT technique to the field of distance education. © 2024 IEEE.

关键词： Students

来源：评论

学校读者我要写书评

暂无评论

Two-dimensional Microwave Through-wall Imaging Algorithm Based on Total Variational Compressive Sensing

Two-dimensional Microwave Through-wall Imaging Algorithm Bas...

引用

2024 Photonics and Electromagnetics research Symposium, PIERS 2024

作者： Zhou, C. Yang, C.X. Tong, M.S. Shanghai Normal University Shanghai Engineering Research Center of Intelligent Education and Bigdata 200234 China Tongji University Shanghai Institute of Intelligent Science and Technology 201203 China

ISBN: (纸本)9798350375909

Microwave imaging is a technique that uses information as a carrier. It enables non-contact acquisition of information from unknown targets, playing a pivotal role in various applications such as radar detection, medical imaging, and industrial inspection. However, practical applications often face challenges in meeting the requirements of traditional imaging algorithms due to limitations imposed by observation environments or economic constraints. Therefore, it is of great research value to explore how limited measurement data can be effectively utilized for reconstructing the target under test. Compressed sensing technology enables high-precision signal reconstruction with sampling frequencies lower than the Nyquist frequency, but its application scope is greatly restricted by the requirement for sparse target distribution measurements, particularly in non-uniform background imaging scenarios. In this study, we propose an inverse scattering method utilizing the total variation-based compressed sensing technique. And through-wall imaging scenes are chosen to validate the robustness of the algorithm. Firstly, by incorporating the gradient of the contrast function as a sparse prior, we reformulate the two-dimensional inverse scattering problem in a time-harmonic regime into an equivalent augmented Lagrange problem with the use of the first-order Born approximation. Subsequently, we construct a compressed sensing matrix. Then, we repeat the aforementioned steps and iteratively solve for the unknown total field and target to be reconstructed alternatively using the distorted Born iteration method framework until meeting certain convergence criteria. The proposed method effectively addresses the reconstruction problem of scattering targets with piecewise constant contrast, while remaining unaffected by non-uniform background. Numerical simulation results validate both the feasibility and efficiency of our proposed method. © 2024 IEEE.

关键词： Iterative methods

来源：评论

学校读者我要写书评

暂无评论

Leveraging Out-of-Domain Noise for Unsupervised Domain Adaptation in Speech Enhancement

Leveraging Out-of-Domain Noise for Unsupervised Domain Adapt...

引用

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

作者： Liao, Yu Guan, Haixin Wei, Shuang Long, Yanhua Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal University Shanghai China Unisound AI Technology Co. Ltd. Beijing China

ISBN: (纸本)9798350368741

When there's a mismatch between the training and test domains, supervised speech enhancement (SE) models trained on synthetic paired noisy-clean data often struggle in real-world scenarios, highlighting the industry's strong demand for unsupervised training and domain adaptation methods. In this study, we introduce PHA-ReMixIT, a novel approach for leveraging out-of-domain (OOD) noise signals to enhance unsupervised domain adaptation in SE. Our method builds upon the state-of-the-art ReMixIT by introducing a paired unsupervised remixing technique, which augments the diversity of target domain training data with OOD noise signals. We further propose a heterogeneous noise invariant training to align the OOD augmented noisy mixtures with their paired heterogeneous counterparts, encouraging the model to output cleaner speech. Additionally, an adaptive focal weighting mechanism is also introduced to dynamically emphasize the data importance of both in-domain and OOD noisy mixtures during model adaptation. Experiments on CHiME-7 unsupervised domain adaptation for conversational speech enhancement (UDASE) task demonstrate that PHA-ReMixIT significantly outperforms the ReMixIT baseline, boosting SE performance on both real and synthesized test sets. © 2025 IEEE.

关键词： adaptive focal weighting domain adaptation PHA-ReMixIT Speech enhancement

来源：评论

学校读者我要写书评

暂无评论

Leverage Learning Behaviour Data for Students' Learning Performance Prediction and Influence Factor Analysis

IEEE Transactions on Artificial Intelligence

引用

IEEE Transactions on Artificial Intelligence 2024年第5期5卷 2422-2433页

作者： Ni, Qin Zhu, Yanjin Zhang, Lingying Lu, Xiaochen Zhang, Lei Shanghai International Studies University Key Laboratory of Multilingual Education with AI Shanghai201620 China Donghua University College of Information Science and Technology Shanghai201620 China Donghua University Engineering Research Center of Digitized Textile and Apparel Technology Ministry of Education Shanghai201620 China Shanghai Normal University Shanghai Engineering Research Center of Intelligent Education and Bigdata and Lab for Educational Big Data and Policymaking Shanghai200234 China

Online education has become increasingly significant for university students and faculty, especially in the context of the modern remote education landscape. However, the inherent space-time separation in online education can create communication delays between teachers and students, making it challenging to monitor students' behaviors effectively. In an effort to understand the connection between students' online engagement and their learning performance, data annotation has been implemented to address the issue of accurately representing students' learning behaviors. Taking the online teaching data of shanghai Normal University platform as the research object, the primary online education problem is explored through data mining, which includes correlation analysis, Gini importance ranking, and principal component analysis (PCA). Then, constructing the learning performance prediction model using random forest (RF) based on PCA by comparing various machine learning algorithms. As a consequence, the most influential online learning behaviors are course duration time, document learning time, test average score, and video completion rate. The overall classification accuracy of the learning performance prediction model is 87.45%, and the highest prediction accuracy for a single category is 96.52%. © 2020 IEEE.

关键词： Students

来源：评论

学校读者我要写书评

暂无评论

Boosting Character-based Mandarin ASR via Chinese Pinyin Representation

引用

International Journal of Speech Technology 2023年第4期26卷 895-902页

作者： Li, Li Long, Yanhua Xu, Dongxing Li, Yijie Key Innovation Group of Digital Humanities Resource and Research and Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal University Shanghai200234 China Unisound AI Technology Co. Ltd. Beijing China

Current end-to-end automatic speech recognition (ASR) models have achieved good results in phonetic language such as English and French. However, Chinese character is a typical ideographic writing, and there is no direct correspondence between Chinese characters and phonetics, but Pinyin, as a mark of the pronunciation of Chinese characters, has an internal connection with these characters. Therefore, it’s crucial to introduce Pinyin into Mandarin ASR to assist the target units modeling of end-to-end speech recognition systems. In this work, we propose a method to boost character-based Conformer-Transducer Mandarin ASR system via Chinese Pinyin representation. Specifically, four new frameworks are investigated to enhance the target units modeling ability of end-to-end Mandarin ASR, including a PCM, PCM with PTLE or PALE and a TAE, they integrate the Pinyin and character information with different implementation in a CTC/transducer multi-task training framework. Experiments on both Aishell and accented ASR tasks show that, the proposed method significantly outperforms the conventional character-based model, it reduces the character error rate of Aishell test set and three accented ASR test sets by relative 6.4%, 5.6%, 4.2% and 4.5%, respectively. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.

关键词： Speech recognition

来源：评论

学校读者我要写书评

暂无评论

CI-Mix: cut instance mix for robust speaker verification

引用

International Journal of Speech Technology 2023年第4期26卷 851-857页

作者： Duan, Yibo Long, Yanhua Li, Yijie Key Innovation Group of Digital Humanities Resource and Research and Shanghai Engineering Research Center of Intelligent Education and Bigdata Shanghai Normal University Shanghai200234 China Unisound AI Technology Co. Ltd. Beijing China

Data augmentation is commonly used to help build a robust speaker verification system, especially when resources are limited. In this paper, we generalize the idea of CutMix to cut instance mix (CI-Mix) for augmenting the training data of speaker verification. The augmentation policy consists of performing the patches cut and paste in both instance-level time-domain and deep feature embedding within a minibatch, without increasing any training data size and computational resource. We apply CI-Mix on the widely used ECAPA-TDNN and ECAPA-CNN-TDNN end-to-end speaker verification systems. Our experiments are performed on both the VoxCeleb and VoxMovies tasks. Extensive results show that the proposed CI-Mix outperforms the state-of-the-art speech data augmentation methods, such as SpecAugment and Mixup, and it shows significant complementary information with these augmentation methods. Together with the simple adding noise augmentation, the CI-Mix achieves up to relative EER reduction of 19.7% on the VoxCeleb1-O, and 16.5%, 10.9%, 10.0%, 10.9% and 13.1% on VoxMovies E-1 to E-5 test sets, compared with the baseline systems. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.

关键词： Time domain analysis

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：