检索结果-内蒙古大学图书馆

Re-Net: Multibranch Network With structural reparameterization for Landslide Detection in Optical Imagery

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2024年 17卷 2828-2837页

作者： Zhang, Rui-xuan Zhu, Wu Li, Zhen-hong Zhang, Bo-chen Chen, Bo Changan Univ Sch Geol Engn & Geomat Xian 710054 Peoples R China Changan Univ Big Data Ctr Geosci & Satellites Xian 710054 Peoples R China Changan Univ Sch Geol Engn & Geomat Xian 710054 Peoples R China Minist Nat Resources Key Lab Ecol Geol & Disaster Prevent Xian 710054 Peoples R China State Key Lab Loess Sci In Preparat Xian 710054 Peoples R China Shenzhen Univ Coll Civil & Transportat Engn Shenzhen 518060 Peoples R China Shenzhen Univ Key Lab Geoenvironm Monitoring Great Bay Area Minist Nat Resources Shenzhen 518061 Peoples R China

Deep learning techniques have been widely adopted in landslide detection by offering powerful feature extraction capabilities and automated processes. However, the pursuit of higher accuracy has led to increasingly complex network structures, which limits the efficiency of models in landslide detection. To tackle this challenge, we have developed a dynamic module, called five-branch feature extraction module (FFEM), based on the theory of structural reparameterization. This module is designed to reconstruct the encoder of the U-shaped network. Our novel network, Re-Net, effectively integrates information from multiple scales during training by utilizing its multibranch structure, which is facilitated by the FFEM. During inference, leveraging the structural reparameterization, the FFEM in Re-Net transforms as a convolutional layer, achieving an impressive 52.9% reduction in parameters, and 34.98% reduction in floating point operations. The efficiency improvement of Re-Net does not come at the expense of sacrificing landslide recognition accuracy. In the public dataset (Bijie dataset), Re-Net achieved improvements of 2.81% in intersection over union (IoU) and 1.93% in F1-Score. In postearthquake landslide detection tasks (Luding Dataset), Re-Net exhibited respective improvements of 2.29% and 1.52%. Moreover, in the task of Landslide Detection, Re-Net demonstrates superior segmentation accuracy compared to other convolutional neural networks, such as Unet++. When compared to other reparameterization modules, FFEM shows significant improvements in IoU and F1-Score in the Bijie dataset, with an average increase of 0.65% and 0.45%, respectively. Similarly, in the Luding dataset, FFEM demonstrates average improvements of 1.56% in IoU and 1.04% in F1-Score.

关键词： GF-2 imagery landslide detection Re-Net structural reparameterization

来源：评论

学校读者我要写书评

暂无评论

RepBoTNet-CESA:An Alzheimer’s Disease Computer Aided Diagnosis Method Using structural reparameterization BoTNet and Cubic Embedding Self Attention

引用

Computers, Materials & Continua 2024年第5期79卷 2879-2905页

作者： Xiabin Zhang Zhongyi Hu Lei Xiao Hui Huang College of Computer Science and Artificial Intelligence Wenzhou UniversityWenzhou325035China Key Laboratory of Intelligence Image Processing and Analysis Wenzhou UniversityWenzhou325035China

Various deep learning models have been proposed for the accurate assisted diagnosis of early-stage Alzheimer’s disease(AD).Most studies predominantly employ Convolutional Neural Networks(CNNs),which focus solely on local features,thus encountering difficulties in handling global *** contrast to natural images,structural Magnetic Resonance Imaging(sMRI)images exhibit a higher number of channel ***,during the Position Embedding stage ofMulti Head Self Attention(MHSA),the coded information related to the channel dimension is *** tackle these issues,we propose theRepBoTNet-CESA network,an advanced AD-aided diagnostic model that is capable of learning local and global features *** combines the advantages of CNN networks in capturing local information and Transformer networks in integrating global information,reducing computational costs while achieving excellent classification ***,it uses the Cubic Embedding Self Attention(CESA)proposed in this paper to incorporate the channel code information,enhancing the classification performance within the Transformer ***,the RepBoTNet-CESA performs well in various AD-aided diagnosis tasks,with an accuracy of 96.58%,precision of 97.26%,and recall of 96.23%in the AD/NC task;an accuracy of 92.75%,precision of 92.84%,and recall of 93.18%in the EMCI/NC task;and an accuracy of 80.97%,precision of 83.86%,and recall of 80.91%in the AD/EMCI/LMCI/NC *** demonstrates that RepBoTNet-CESA delivers outstanding outcomes in various AD-aided diagnostic ***,our study has shown that MHSA exhibits superior performance compared to conventional attention mechanisms in enhancing ResNet ***,the Deeper RepBoTNet-CESA network fails to make further progress in AD-aided diagnostic tasks.

关键词： Alzheimer CNN structural reparameterization multi head self attention computer aided diagnosis

来源：评论

学校读者我要写书评

暂无评论

RepMobile: A MobileNet-Like Network With structural reparameterization for Sensor-Based Human Activity Recognition

引用

IEEE SENSORS JOURNAL 2024年第15期24卷 24224-24237页

作者： Yu, Jianglai Zhang, Lei Cheng, Dongzhou Bu, Can Wu, Hao Song, Aiguo Nanjing Normal Univ Sch Elect & Automat Engn Nanjing 210023 Peoples R China Yunnan Univ Sch Informat Sci & Engn Kunming 650500 Peoples R China Southeast Univ Sch Instrument Sci & Engn Nanjing 210096 Peoples R China

How to design and deploy efficient neural network backbones always plays a crucial role in sensor-based human activity recognition (HAR) on mobile devices, which have targeted at reducing floating-point operations (FLOPs) and parameter count, hence offering a faster activity inference. However, since the most commonly used efficiency indicators such as FLOPs and parameter count are indirect metrics for activity inference, they could not directly translate into practical inference latency in real-world. This article is the first attempt to mitigate such discrepancy between theoretic computational efficiency and practical inference latency. Our main goal is to identify and analyze key performance bottlenecks including activations and branching, which might potentially incur higher inference latency for on-device activity inference due to the degree of parallelism and memory access (MAC) cost. To this end, we provide an in-depth analysis about how the indirect metrics and different structural design choices correlate with practical inference latency, which has been rarely explored in ubiquitous HAR scenario. In particular, based on the above analyses, motivated by an idea of structural reparameterization, we present an efficient architecture called RepMobile built on MobileNet by decoupling training-time and inference-time architectures, where a linearly overparameterized model is used to improve accuracy at training time, and a reparameterization is performed for faster activity inference. We validate the effectiveness and efficiency of our proposed method on several public HAR benchmarks, including PAMAP2, UNIMIB-SHAR, WISDM, and USC-HAD, as well as a resource-restrained mobile device, indicating a better tradeoff between accuracy and on-device latency.

关键词： Computer architecture Sensors Accuracy Training Human activity recognition Low latency communication Feature extraction Deep learning human activity recognition mobile devices sensor structural reparameterization

来源：评论

学校读者我要写书评

暂无评论

A Bearing Fault Diagnosis Framework Based on Few-Shot Learning with Distribution Consistency and structural reparameterization 7th

A Bearing Fault Diagnosis Framework Based on Few-Shot Learni...

引用

7th International Conference on Green Technology and Sustainable Development

作者： Gia-Bao Truong Nhu-Linh Than Manh-Hung Vu Van-Truong Pham Thi-Hue Nguyen Thi-Thao Tran Hanoi Univ Sci & Technol Sch Elect & Elect Engn Dept Automat Engn Hanoi Vietnam

ISBN: (纸本)9783031761966;9783031761973

Nowadays, end-to-end deep learning networks inherit the development of convolutional neural networks to have good performance, but they can only do well when they are learned on a large number of samples. However, building a large dataset is difficult for specific data, such as bearing faults. In this paper, a novel end-to-end few-shot model is proposed for bearing fault diagnosis even in the case of limited data with artificial and real faults. The architecture is designed to optimally exploit the feature extraction process as well as the covariance metric based on the distribution consistency for the few-shot classification tasks. In particular, we propose a module for the feature extraction stage with structural re-parameterization combined with a self-attention mechanism. Specifically, the input data goes through the STEM block and is then put into three stages, which consist of two stages called RepMixer block using re-parameterizing skip connections and the other stage called Attention block. We also use a covariance metric to help the feature extractor learn more precise features and enhance its discriminative performance during training. Experimental results based on the public imbalanced dataset CWRU, with different degrees of training samples, show that our model can significantly improve the accuracy of identifying faults-bearing to guarantee safety in industrial systems. Our code is released at https://***/giabao804/few-shot-structural-rep.

关键词： Bearing faults diagnosis Few-shot learning Self attention Mechanism structural reparameterization Distribution consistency

来源：评论

学校读者我要写书评

暂无评论

A Lightweight Edge Computing Neural Network for Online Photovoltaic Defect Inspection

引用

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2025年 74卷

作者： Ye, Jinlin Liu, Yuhan Yang, Liang Chen, Haiyong Wang, Changjun Zhou, Yidong Zhang, Wei Hebei Univ Technol Sch Artificial Intelligence Tianjin 300130 Peoples R China Peking Union Med Coll Hosp Dept Breast Surg Beijing 100041 Peoples R China

The photovoltaic energy plays a crucial role in the global transition to sustainable energy. Regular inspection of photovoltaic power plants is essential for the timely detection of potential faults and the maintenance of optimal power generation efficiency. Autonomous aerial vehicles (AAVs) are increasingly being applied as efficient and flexible tools for inspecting photovoltaic power plants. However, AAV inspection missions often rely on cloud computing and photovoltaic power plants are usually located in remote areas with limited access to cloud resources. Directly deploying cloud-based detection algorithms on AAVs can result in the AAVs being unable to run these algorithms in real-time due to hardware limitations. Therefore, this article proposes a structural reparameterization module (SRPM) and develops a lightweight neural network (GSDet), specifically designed for defect detection in photovoltaic plants under edge computing environments. In addition, the neuron dependence graph pruning method (NDGraph) is proposed to model neuronal dependencies, and a group-level sparse training method is employed to prune GSDet. To mitigate accuracy loss, the channel feature knowledge distillation (CFKD) is proposed. The performance of GSDet is evaluated on both a private dataset [autonomous aerial vehicle photovoltaic (AAVPV)] and a public dataset [photovoltaic electroluminescence (PVEL)]. Experimental results demonstrate that GSDet significantly reduces the number of parameters and computational complexity compared with baseline models, while substantially improving inference speed. After pruning, GSDet-Tiny further reduces the parameter count and computational load while maintaining accuracy and enhancing inference speed. In comparison with state-of-the-art general object detection models, GSDet and GSDet-Tiny achieve an excellent balance between computational efficiency, accuracy, and real-time performance.

关键词： Edge computing knowledge distillation model pruning model pruning neural network lightweighting neural network lightweighting photovoltaic plant inspection photovoltaic plant inspection structural reparameterization structural reparameterization structural reparameterization

来源：评论

学校读者我要写书评

暂无评论

Real-time surface defect detection algorithm on aluminum ingot alloy casting lines

引用

PHYSICA SCRIPTA 2025年第1期100卷

作者： Chen, Hong Yan, Feiya Yang, Jing Yan, Jianguo Qin, Tao Zhang, Jing Guizhou Univ Elect Engn Coll Guiyang 550025 Peoples R China Guiyang Aluminum Magnesium Design & Res Inst Co Lt Guiyang 550025 Peoples R China Guizhou Prov Key Lab InternetIntelligent Mfg Guiyang 550025 Peoples R China China Power Construct Grp Guizhou Engn Co Ltd Guiyang 550025 Peoples R China

Aluminum alloy ingot surface defect detection technology is crucial to ensure that product quality meets standards. Existing inspection model inspection precision and speed are difficult to meet requirements of fast and accurate detection on production lines, to address this problem, a new real-time surface defect detection algorithm Al-YOLO on aluminum alloy ingot casting line was proposed. Firstly, K-Means++ was embedded into YOLOv5 to optimize the detection anchor frame for the aluminum alloy ingot scenario. Secondly, the structural reparameterization technique was used to design the main trunk to speed up the model inference speed. Then combined with the dynamic head framework known as DyHead, the attention mechanism was unified with the detector's head, thereby enhancing the precision of small defect identification. Finally, the Focal function was introduced, and combined with EIoU to get the Focal-EIoU border regression loss function to address the uneven sample distribution. On the aluminum alloy ingot surface defects dataset, the mAP of Al-YOLO reaches 72.5%, and the FPS also reaches 57.1, which improves the mAP and FPS by 4.0% and 11.2, respectively, compared to the original model, and outperforms the current mainstream detection methods on the NEU-DET dataset. Experimental evidence Al-YOLO has better inspection precision and speed, which can better adapted to the requirements of fast and accurate detection in production lines.

关键词： aluminum alloy ingot surface defect detection YOLOv5 structural reparameterization dynamic head

来源：评论

学校读者我要写书评

暂无评论

Ship detection algorithm based on structural reparameterize dilated large-kernel convolution and spatial selective kernel mechanism

引用

JOURNAL OF REAL-TIME IMAGE PROCESSING 2025年第2期22卷 1-13页

作者： Shi, Yujing Wang, Haicheng Li, Shanqiang Fujian Univ Technol Sch Comp Sci & Math Fuzhou Peoples R China

In response to the issues of large target scale variations and complex background environments in ship detection, this paper presents a single-stage ship detection algorithm named DrbLSK, which is an improvement on YOLOv5. First, to enhance the model's ability to model the context of targets with different scales, a kernel selection module is introduced into the backbone network. By dynamically selecting the receptive field, environmental interference is reduced. Simultaneously, the combination of large-kernel dilated convolution and structural reparameterization is employed to reduce model parameters and enhance the feature expression capability of the backbone network. Next, a decoupled detection head is utilized to alleviate the conflict between classification and regression tasks in the target detection task. Moreover, CIoU is used in the detection head to replace the original loss function, thereby accelerating the convergence of the network. Experimental results show that, on the ABOships dataset, the improved model reduces the number of parameters while increasing accuracy by 2.5% compared to YOLOv5, with a mean average precision of 63.9.

关键词： Marine object detection Ship detection YOLOv5 structural reparameterization Convolution neural network Faster R-CNN

来源：评论

学校读者我要写书评

暂无评论

Lightweight cross-resolution coarse-to-fine network for efficient deformable medical image registration

引用

MEDICAL PHYSICS 2025年

作者： Liu, Jun Shen, Nuo Wang, Wenyi Li, Xiangyu Wang, Wei Yuan, Yongfeng Tian, Ye Luo, Gongning Wang, Kuanquan Harbin Inst Technol Sch Comp Sci & Technol Harbin Heilongjiang Peoples R China Harbin Inst Technol Shenzhen Sch Comp Sci & Technol Shenzhen Guangdong Peoples R China Harbin Med Univ Affiliated Hosp 1 Cardiovasc Inst Dept Cardiol Harbin Heilongjiang Peoples R China

BackgroundAccurate and efficient deformable medical image registration is crucial in medical image analysis. While recent deep learning-based registration methods have achieved state-of-the-art accuracy, they often suffer from extensive network parameters and slow inference times, leading to inefficiency. Efforts to reduce model size and input resolution can improve computational efficiency but frequently result in suboptimal *** address the trade-off between high accuracy and efficiency, we propose a Lightweight Cross-Resolution Coarse-to-Fine registration framework, termed *** method is built on an ultra-lightweight U-Net architecture with only 0.1 million parameters, offering remarkable efficiency. To mitigate accuracy degradation resulting from fewer parameters while preserving the lightweight nature of the networks, LightCRCF introduces three key innovations as follows: (1) selecting an efficient cross-resolution coarse-to-fine (C2F) registration strategy and integrating it into the lightweight network to progressively decompose the deformation fields into multiresolution subfields to capture fine-grained deformations;(2) a Texture-aware reparameterization (TaRep) module that integrates Sobel and Laplacian operators to extract rich textural information;(3) a Group-flow reparameterization (GfRep) module that captures diverse deformation modes by decomposing the deformation field into multiple groups. Furthermore, we introduce a structural reparameterization technique that enhances training accuracy through multibranch structures of the TaRep and GfRep modules, while maintaining efficient inference by equivalently transforming these multibranch structures into single-branch standard *** evaluate LightCRCF against various methods on the three public MRI datasets (LPBA, OASIS, and ACDC) and one CT dataset (abdomen CT). Following the previous data division methods, the LPBA dataset comprises 30 training image pairs

关键词： coarse-to-fine strategy deformation decomposition lightweight network medical image registration structural reparameterization

来源：评论

学校读者我要写书评

暂无评论

DBA-YOLO: A Lightweight Fire DetectionAlgorithm with structural Reparameterized Multiple Attention Mechanisms

引用

EARTH SCIENCE INFORMATICS 2025年第2期18卷 1-15页

作者： Guan, Shiyu Zhao, Xiao He, Lifeng Shaanxi Univ Sci & Technol Coll Elect Informat & Artificial Intelligence Xian 710021 Peoples R China Aichi Prefectural Univ Grad Sch Informat Sci & Technol Nagakute Aichi 4801198 Japan

In response to the challenges of detecting smoke-like objects in fire scenes and fire detection under various natural lighting conditions, we propose an improved fire detection algorithm based on YOLOv7-Tiny, named DBA-YOLO. This novel approach introduces an attention feature extraction module that combines structural re-parameterization, extending the feature extraction network architecture. It enhances feature propagation for flame target recognition and improves the performance of the network model. The integration of a focal modulation network into the head detection layer facilitates better feature exchange and enlarges the receptive field of the detection layer. By replacing the original model's detection head with a variable attention detection head, we significantly increase detection precision. It achieves an average detection precision of 85.4% and a mean Average Precision (mAP50) of 85.6%. Compared to the popular baseline algorithm YOLOv7, our proposed DBA-YOLO algorithm achieves a 3.5% increase in mAP and a 2.1% increase in precision. With an input image size of 640 x 640 pixels, the average detection time per frame is 6.5ms, which is 27% faster than the YOLOv7 model, surpassing most current target detection models in all fire detection metrics including precision, recall, and mean Average Precision (mAP). The proposed model effectively detects fire targets as well as fire-like and smoke-like objects, minimizing false positives with high precision and is applicable to target detection in other complex scenarios.

关键词： Fire Detection structural reparameterization Multiple Attention Smoke Detection Convolutional Neural Network

来源：评论

学校读者我要写书评

暂无评论

DR-Block: Convolutional Dense reparameterization for CNN Generalization Free Improvement

引用

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 2024年第11期34卷 10618-10631页

作者： Yan, Qingqing Li, Shu He, Zongtao Hu, Mengxian Liu, Chengju Chen, Qijun Tongji Univ Robot & Artificial Intelligence Lab RAIL Shanghai 201804 Peoples R China

As an emerging and popular technique for boosting CNNs, structural reparameterization (SR) decouples the training and inference structures to alter the training dynamics and achieve cost-free improvement of a given network. Existing SR methods often prioritize network expressiveness enhancement but have yet to investigate approaches to mitigate significant bias and non-robustness of model prediction due to over-reliance on training data distribution and image noise. To this end, inspired by the effective strength of implicit regularization on the problem, this paper introduces an extra balanced implicit regularization mechanism into SR techniques to enhance the generalization of a given network for the first time. Specifically, we propose a novel SR module named DR-Block, which is used to complicate each convolutional layer of a given CNN during training. It draws on the advantages of deep matrix factorization with the regularization effect and further improves singular value dynamics by introducing batch normalization and dense connections to alleviate network degradation. At inference time, DR-Block can be equivalently reparameterized back into a single convolution for deployment. Furthermore, we empirically demonstrate the role of each design in DR-Block and explicitly reveal its inherent mechanism, which lies in enhancing the movement of large singular values while countering the attenuation of small ones. This helps enhance the interpretability of SR techniques. Experiments illustrate that DR-Block is an impressive alternative for a regular convolution layer of any structure and outperforms the existing SR methods in improving mainstream network architectures on various visual tasks. The code is available at https://***/qyan0131/DRBlock.

关键词： structural reparameterization implicit regularization deep matrix factorization training dynamic network generalization.

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：