检索结果-内蒙古大学图书馆

FM-RTDETR: Small Object Detection Algorithm Based on Enhanced Feature Fusion With Mamba

学校读者我要写书评

暂无评论

IEEE signal processing LETTERS 2025年 32卷 1570-1574页

作者： Yang, Yuchuan Dai, Jiahui Wang, Yong Chen, Yafei Chongqing Univ Technol Sch Artificial Intelligence Chongqing 401135 Peoples R China

Traditional real-time object detection networks deployed in autonomous aerial vehicles (AAVs) struggle to extract features from small objects in complex backgrounds with occlusions and overlapping objects. To address this challenge, we propose FM-RTDETR, a real-time object detection algorithm optimized for small object detection. We redesign the encoder of RT-DETRv2 by integrating the Feature Aggregation and Diffusion Network (FADN), improving the algorithm's ability to capture contextual information. Subsequently, we introduce the Parallel Atrous Mamba Feature Fusion Module (PAMFFM), which combines shallow and deep semantic information to better capture small object features. Furthermore, we propose the Cross-stage Enhanced Feature Fusion Module (CEFFM), merging features for small objects to provide richer and more detailed information. Finally, we propose STIoU Loss, which incorporates a penalty term to adjust the scaling of the loss function, improving detection granularity for small objects. FM-RTDETR achieves AP(50) scores of 54.0% and 56.3% on the VisDrone2019-DET and AI-TOD datasets. Compared with other state-of-the-art methods, our method shows great potential in small object detection from drones.

关键词： Object detection signal processing algorithms Feature extraction Transformers Training Real-time systems Accuracy Decoding Convolution Artificial intelligence Feature fusion RT-DETRv2 small object detection VisDrone

Low Cost Variable Step-Size LMS With Maximum Similarity to the Affine Projection Algorithm

学校读者我要写书评

暂无评论

IEEE OPEN JOURNAL OF signal processing 2024年 5卷 82-91页

作者： Ferrer, Miguel De Diego, Maria Gonzalez, Alberto Univ Politecn Valencia UPV Inst Telecommun & Multimedia Applicat iTEAM Valencia 46022 Spain

The LMS algorithm is widely employed in adaptive systems due to its robustness, simplicity, and reasonable performance. However, it is well known that this algorithm suffers from a slow convergence speed when dealing with colored reference signals. Numerous variants and alternative algorithms have been proposed to address this issue, though all of them entail an increase in computational cost. Among the proposed alternatives, the affine projection algorithm stands out. This algorithm has the peculiarity of starting from $N$ data vectors of the reference signal. It transforms these vectors into as many data vectors suitably normalized in energy and mutually orthogonal. In this work, we propose a version of the LMS algorithm that, similar to the affine projection algorithm, starts from $N$ data vectors of the reference signal but corrects them by using only a scalar factor that functions as a convergence step. Our goal is to align the behavior of this algorithm with the behavior of the affine projection algorithm without significantly increasing the computational cost of the LMS.

关键词： signal processing algorithms Convergence Projection algorithms Computational efficiency Approximation algorithms Costs Robustness Adaptive filters affine projection algorithm variable step-size

Binaural Multichannel Blind Speaker Separation With a Causal Low-Latency and Low-Complexity Approach

学校读者我要写书评

暂无评论

IEEE OPEN JOURNAL OF signal processing 2024年 5卷 238-247页

作者： Westhausen, Nils L. Meyer, Bernd T. Carl Von Ossietzky Univ Oldenburg Commun Acoust D-26129 Oldenburg Germany Cluster Excellence Hearing4All D-26129 Oldenburg Germany

In this article, we introduce a causal low-latency low-complexity approach for binaural multichannel blind speaker separation in noisy reverberant conditions. The model, referred to as Group Communication Binaural Filter and Sum Network (GCBFSnet) predicts complex filters for filter-and-sum beamforming in the time-frequency domain. We apply Group Communication (GC), i.e., latent model variables are split into groups and processed with a shared sequence model with the aim of reducing the complexity of a simple model only containing one convolutional and one recurrent module. With GC we are able to reduce the size of the model by up to 83% and the complexity up to 73% compared to the model without GC, while mostly retaining performance. Even for the smallest model configuration, GCBFSnet matches the performance of a low-complexity TasNet baseline in most metrics despite the larger size and higher number of required operations of the baseline.

关键词： Complexity theory Low latency communication Speech enhancement Convolution signal processing algorithms Microphones MIMO communication Binaural low-latency multi-channel real-time speaker-separation

Deep Learning Model for CS-Based signal Recovery for IRS-Assisted Near-Field THz MIMO System

学校读者我要写书评

暂无评论

IEEE OPEN JOURNAL OF VEHICULAR TECHNOLOGY

IEEE OPEN JOURNAL OF VEHICULAR TECHNOLOGY 2024年 5卷 1326-1335页

作者： Sharma, Vaishali Keshari, Prakhar Sharma, Sanjeev Deka, Kuntal Krejcar, Ondrej Bhatia, Vimal Indian Inst Technol Indore Dept Elect Engn Indore 453552 India Indian Inst Technol BHU Varanasi Dept Elect Engn Varanasi 221005 India Indian Inst Technol Guwahati Dept Elect & Elect Engn Gauhati 781039 India Skoda Auto Univ Mlada Boleslav 29301 Czech Republic Univ Teknol Malaysia Malaysia Japan Int Inst Technol Skudai 81310 Malaysia Soochow Univ Sch Elect & Informat Engn Suzhou 215006 Jiangsu Peoples R China

Terahertz (THz) communication is a cutting-edge technology for the sixth-generation (6G) networks, offering vast bandwidths and data rates up to terabits per second, significantly advancing vehicular connectivity and services. However, THz signals are impacted by attenuation, path loss, and beam misalignment. Furthermore, the requisite high Nyquist sampling rates for THz systems increase the computational and system complexity of the receiver. A promising solution to navigate these obstacles involves the use of intelligent reflecting surfaces (IRS)-enhanced multiple-input multiple-output (MIMO) technology, which steers THz wave propagation. However, the substantial dimensions associated with IRS and MIMO extend the near-field, particularly at THz frequencies, as indicated by the Rayleigh distance and suffer from beam squint. To reduce system complexity and reduce sampling to sub-Nyquist rate, we propose a novel receiver design for an IRS-assisted near-field MIMO THz system that employs low-complexity compressed sensing. This method introduces an IRS signal-matched (IRSSM) measurement matrix with beam squint for capturing the transmitted signal at a sub-Nyquist rate, taking advantage of the sparsity in the signal and THz channels, and signal recovery using the deep learning (DL) model. Simulation results for symbol error rate (SER) and normalized mean square error (NMSE) performance indicate that the proposed DL-based receiver outperforms conventional recovery algorithms based on orthogonal matching pursuit (OMP) CS-recovery and dictionary-shrinkage estimation (DSE).

关键词： Terahertz communications Receivers Terahertz radiation Channel estimation Array signal processing Symbols signal processing algorithms THz band symbol detection compressed sensing MIMO DNN near-field beam squint

With signal processing Support, Prosthetics Are Becoming Safer, More Natural, and Increasingly Sensitive

学校读者我要写书评

暂无评论

IEEE signal processing MAGAZINE 2021年第4期38卷 8-11页

作者： Edwards, John Technol Gilbert Gilbert AZ 85234 USA

to accommodate the prosthesis![Special Reports]

关键词： Prosthetics Biomdical equipment Medical devices Market research Sensor systems Machine learning algorithms signal processing algorithms Tactile sensors Medical robotics Perturbation methods

Body Motion Segmentation via Multilayer Graph processing for Wearable Sensor signals

学校读者我要写书评

暂无评论

IEEE OPEN JOURNAL OF signal processing 2024年 5卷 934-947页

作者： Deng, Qinwen Zhang, Songyang Ding, Zhi Univ Calif Davis Dept Elect & Comp Engn Davis CA 95616 USA Univ Louisiana Lafayette Dept Elect & Comp Engn Lafayette LA 70504 USA

Human body motion segmentation plays a major role in many applications, ranging from computer vision to robotics. Among a variety of algorithms, graph-based approaches have demonstrated exciting potential in motion analysis owing to their power to capture the underlying correlations among joints. However, most existing works focus on simpler single-layer geometric structures, whereas multi-layer spatial-temporal graph structure can provide more informative results. To provide an interpretable analysis on multilayer spatial-temporal structures, we revisit the emerging field of multilayer graph signal processing (M-GSP), and propose novel approaches based on M-GSP to human motion segmentation. Specifically, we model the spatial-temporal relationships via multilayer graphs (MLG) and introduce M-GSP spectrum analysis for feature extraction. We present two different M-GSP based algorithms for unsupervised segmentation in the MLG spectrum and vertex domains, respectively. Our experimental results demonstrate the robustness and effectiveness of our proposed methods.

关键词： Motion segmentation Computer vision Clustering algorithms signal processing algorithms signal processing Nonhomogeneous media Image segmentation multilayer graph signal processing unsupervised learning

Event-Type Identification in Power Grids Using a Spectral Correlation Function-Aided Convolutional Neural Network

学校读者我要写书评

暂无评论

IEEE OPEN ACCESS JOURNAL OF POWER AND ENERGY

IEEE OPEN ACCESS JOURNAL OF POWER AND ENERGY 2024年 11卷 653-664页

作者： Alaca, Ozgur Ekti, Ali Riza Joo, Jhi-Young Stenvig, Nils Oak Ridge Natl Lab Electrificat & Energy Infrastruct Div Oak Ridge TN 37831 USA Lawrence Livermore Natl Lab Computat Engn Div Livermore CA 94550 USA

Rapid and accurate identification of events in power grids is critical to ensuring system reliability and security. This study introduces a novel event-type identification method, utilizing a Spectral Correlation Function (SCF)-aided Convolutional Neural Network (CNN). The proposed method employs a six-stage cascaded structure consisting of: (1) data collection, (2) clipping, (3) augmentation, (4) feature extraction (FE), (5) training, and (6) testing. Real-world power grid signals sourced from the Grid Event Signature Library are used for both training and testing. To improve robustness, additive white Gaussian noise (AWGN) is introduced at various signal-to-noise ratio (SNR) levels to augment the dataset. The SCF-based FE method captures distinctive event-type characteristics by exploiting the spectral correlation of signals, allowing the CNN architecture to effectively learn and generalize event patterns. The proposed method is benchmarked against seven conventional techniques, using real-world power grid signals representing four distinct event types: blown fuse, line switching, low amplitude arcing, and transformer energization. Key performance metrics-prediction accuracy, mean absolute error (MAE), precision, recall, F1-score, and confusion matrix-are employed to evaluate the performance. Results demonstrate that the SCF-CNN method outperforms traditional approaches across all metrics and SNR levels, achieving over 99% prediction accuracy and nearly zero error for SNR values above 6 dB. This signifies its efficacy in reliable event-type identification for power grid applications.

关键词： Power grids Iron Fault diagnosis signal to noise ratio Convolutional neural networks Correlation Training Switches Transformers signal processing algorithms Data augmentation event and fault classification grid event signature library (GESL) low amplitude arcing machine learning spectral correlation function (SCF)

Distributed Combined Channel Estimation and Optimal Uplink Receive Combining for User- Centric Cell-Free Massive MIMO Systems

学校读者我要写书评

暂无评论

IEEE OPEN JOURNAL OF signal processing 2024年 5卷 559-576页

作者： Van Rompaey, Robbe Moonen, Marc Katholieke Univ Leuven Dept Elect Engn B-3000 Leuven Belgium

Cell-free massive MIMO (CFmMIMO) is considered as one of the enablers to meet the demand for increasing data rates of next generation (6G) wireless communications. In user-centric CFmMIMO, each user equipment (UE) is served by a user-selected set of surrounding access points (APs), requiring efficient signal processing algorithms minimizing inter-AP communications, while still providing a good quality of service to all UEs. This paper provides algorithms for channel estimation (CE) and uplink (UL) receive combining (RC), designed for CFmMIMO channels using different assumptions on the structure of the channel covariances. Three different channel models are considered: line-of-sight (LoS) channels, non-LoS (NLoS) channels (the common Rayleigh fading model) and a combination of LoS and NLoS channels (the general Rician fading model). The LoS component introduces correlation between the channels at different APs that can be exploited to improve the CE and the RC. The channel estimates and receive combiners are obtained in each AP by processing the local antenna signals of the AP, together with compressed versions of all the other antenna signals of the APs serving the UE, during UL training. To make the proposed method scalable, the distributed user-centric channel estimation and receive combining (DUCERC) algorithm is presented that significantly reduces the necessary communications between the APs. The effectiveness of the proposed method and algorithm is demonstrated via numerical simulations.

关键词： Channel estimation Antennas signal processing algorithms Rician channels Rayleigh channels Interference Estimation Cell-free massive MIMO CSI-free channel estimation distributed user-centric processing minimum-mean-squared-error (MMSE) uplink receive combining

Smoothing ADMM for Sparse-Penalized Quantile Regression With Non-Convex Penalties

学校读者我要写书评

暂无评论

IEEE OPEN JOURNAL OF signal processing 2024年 5卷 213-228页

作者： Mirzaeifard, Reza Venkategowda, Naveen K. D. Gogineni, Vinay Chakravarthi Werner, Stefan Norwegian Univ Sci & Technol NTNU Dept Elect Syst N-7032 Trondheim Norway Linkoping Univ Dept Sci & Technol S-58183 Linkoping Sweden Univ Southern Denmark Maersk Mc Kinney Moller Inst SDU Appl & Data Sci DK-5230 Odense Denmark Aalto Univ Dept Informat & Commun Engn Aalto 00076 Finland

This paper investigates quantile regression in the presence of non-convex and non-smooth sparse penalties, such as the minimax concave penalty (MCP) and smoothly clipped absolute deviation (SCAD). The non-smooth and non-convex nature of these problems often leads to convergence difficulties for many algorithms. While iterative techniques such as coordinate descent and local linear approximation can facilitate convergence, the process is often slow. This sluggish pace is primarily due to the need to run these approximation techniques until full convergence at each step, a requirement we term as a secondary convergence iteration. To accelerate the convergence speed, we employ the alternating direction method of multipliers (ADMM) and introduce a novel single-loop smoothing ADMM algorithm with an increasing penalty parameter, named SIAD, specifically tailored for sparse-penalized quantile regression. We first delve into the convergence properties of the proposed SIAD algorithm and establish the necessary conditions for convergence. Theoretically, we confirm a convergence rate of o(k(-1/4)) for the sub-gradient bound of the augmented Lagrangian, where k denotes the number of iterations. Subsequently, we provide numerical results to showcase the effectiveness of the SIAD algorithm. Our findings highlight that the SIAD method outperforms existing approaches, providing a faster and more stable solution for sparse-penalized quantile regression.

关键词： Convergence Smoothing methods Optimization signal processing algorithms Convex functions signal processing Prediction algorithms Quantile regression non-smooth and non-convex penalties ADMM sparse learning