检索结果-内蒙古大学图书馆

Deep Fourier-Based Task-Aware Metric Network for Few-Shot SAR Target classification

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2025年 74卷

作者： Zheng, Jia Li, Ming Chen, Hongmeng Zhang, Peng Wu, Yan Xidian Univ Natl Key Lab Radar Signal Proc Xian 710071 Peoples R China Beijing Inst Radio Measurement Beijing 100854 Peoples R China Xidian Univ Sch Elect Engn Remote Sensing Image Proc & Fus Grp Xian 710071 Peoples R China

The downsampling operations in convolutional neural networks (CNNs) lead to the loss of high-frequency texture features in images, posing challenges for synthetic aperture radar (SAR) automatic target recognition (ATR). To this end, some algorithms introduce the Fourier transform to retain and restore high-frequency information. However, these algorithms are prone to overfitting with a small number of SAR samples. Additionally, the similarity of different SAR target backgrounds further complicates distinguishing between targets and backgrounds during metric learning. To address these issues, a novel few-shot learning (FSL) method for SAR target classification, namely, the deep Fourier-based task-aware metric network, is proposed. First, we propose a frequency-domain feature extraction module. This module uses a fast Fourier transform (FFT) to convert the input image into the frequency domain, and then performs convolution on the amplitude and phase information to extract and enhance features. Then, the inverse FFT and convolution operations are applied to generate the final frequency-domain feature representation. Furthermore, we propose a parameter-free contrast metric module that identifies and reduces background-related features by calculating the similarity between the query sample and both the target and nontarget classes. This module fully considers the context of the entire task support set, effectively improving the model's recognition accuracy. Finally, to address the issue of increased sensitivity to noise and outliers due to unimportant local descriptors during the metric process, we design a weighted k-nearest neighbor (k-NN) method to enhance the model's focus on important information, thereby improving classification accuracy. The experimental results on two public SAR datasets demonstrate that our proposed algorithm achieves higher classification accuracy than other FSL algorithms and exhibits discriminative feature representation and metric capabilit

关键词： Measurement Synthetic aperture radar Feature extraction classification algorithms Radar polarimetry Target recognition Frequency-domain analysis Accuracy Fast Fourier transforms Training Automatic target classification contrastive metric module few-shot learning (FSL) Fourier transform synthetic aperture radar (SAR)

来源：评论

学校读者我要写书评

暂无评论

Artifacts Extraction From Video Head Impulse Test Data Using Time Series classification Methods and VOR Gain Analysis

引用

IEEE ACCESS 2025年 13卷 56520-56530页

作者： Baydadaev, Shokhrukh Usmankhujaev, Saidrasul Sung Kim, Kyu Woo Kwon, Jang Inha Univ Dept Elect & Comp Engn Incheon 22212 South Korea Inha Univ Coll Med Dept Otorhinolaryngol Inchon 22332 South Korea

The video head impulse test (vHIT) has become an essential tool in the examination of patients with dizziness and other balance disorders, providing significant data on all six semicircular canals. The clinical interpretation of the vestibulo-ocular reflex (VOR) dynamic function of the human brain in vertigo and balance disorders using the vHIT method poses a considerable challenge. We utilize VOR gain measurements to ascertain the health of the patient's vestibular system. However, all methods have inherent limitations due to the presence of noise and artifacts in the data, which can significantly affect the gain values of normal and abnormal impulses, leading to inaccuracies. This paper presents a comprehensive study, where we have created a dataset using vHIT data from 5,782 clinical patients from the Department of Otorhinolaryngology, College of Medicine, Inha University. We apply time series classification (TSC) algorithms to identify and filter artifact-affected impulses, ensuring more reliable VOR gain calculations. The encoder model achieved a classification accuracy of 94%, surpassing previous approaches such as SSNHLV (92%) and AI-based stroke (88%) classification. Statistical analysis confirms the significance of our method, with p-values (<0.05) demonstrating a clear distinction between normal, abnormal, and artifact impulses. By improving impulse classification, our approach enhances the precision of VOR gain calculations, contributing to more accurate clinical diagnoses of vestibular disorders.

关键词： Deep learning Accuracy Magnetic heads Head Time series analysis Gain measurement Data mining classification algorithms Statistical analysis Otorhinolaryngology Video head impulse test (vHIT) vestibulo-ocular reflex (VOR) VOR gain video-oculography (VOG) device time-series classification (TSC)

来源：评论

学校读者我要写书评

暂无评论

MDSC-Net: Multi-Modal Discriminative Sparse Coding Driven RGB-D classification Network

引用

IEEE TRANSACTIONS ON MULTIMEDIA 2025年 27卷 442-454页

作者： Xu, Jingyi Deng, Xin Fu, Yibing Xu, Mai Li, Shengxi Beihang Univ Sch Elect & Informat Engn Beijing 100191 Peoples R China Beihang Univ Shen Yuan Honors Coll Beijing 100191 Peoples R China

In this paper, we propose a novel sparsity-driven deep neural network to solve the RGB-D image classification problem. Different from existing classification networks, our network architecture is designed by drawing inspirations from a new proposed multi-modal discriminative sparse coding (MDSC) model. The key feature of this model is that it can gradually separate the discriminative and non-discriminative features in RGB-D images in a coarse-to-fine manner. Only the discriminative features are integrated and refined for classification, while the non-discriminative features are discarded, to improve the classification accuracy and efficiency. Derived from the MDSC model, the proposed network is composed of three modules, i.e., the shared feature extraction (SFE) module, discriminative feature refinement (DFR) module, and classification module. The architecture of each module is derived from the optimization solution in the MDSC model. To the best of our knowledge, this is the first time a fully sparsity-driven network has been proposed for RGB-D image classification. Extensive results verify the effectiveness of our method on different RGB-D image datasets.

关键词： Feature extraction Image coding Image classification Dictionaries Optimization classification algorithms Deep learning Convolutional neural networks Accuracy Encoding Convolutional sparse coding RGB-D image classification Discriminative features

来源：评论

学校读者我要写书评

暂无评论

PDCNet: A Polarimetric Data-Enhanced Contrastive Learning Network for PolSAR Land Cover classification

引用

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 10010-10025页

作者： Ren, Bo Hua, Chaoyue Hou, Biao Lv, Jian Yang, Chen Jiao, Licheng Chanussot, Jocelyn Xidian Univ Minist Educ China Key Lab Intelligent Percept & Image Understanding Xian 710071 Peoples R China Univ Grenoble Alpes Inria CNRS Grenoble INPLJK F-38000 Grenoble France

Polarimetric synthetic aperture radar (PolSAR) has rich polarization information, offering an efficient and reliable means of collecting information. However, how to effectively leverage these complex data to extract polarization features remains a key challenge. Recently, contrastive learning has been successful in computer vision, with fewer labeled and a large amount of unlabeled data. Inspired by it, a polarimetric data-enhanced contrastive learning network (PDCNet) for PolSAR land cover classification is proposed. The design process for polarimetric contrastive learning involves the construction of positive samples, the establishment of a PolSAR-based network architecture for contrastive learning, and the formulation of the loss function. The proposed method first constructs positive sample pairs for contrastive learning by leveraging the linear relationship between the coherency matrix (T-matrix) and the covariance matrix (C-matrix). In addition, since both of them are 3 X 3 complex conjugate symmetric matrices, we design an adaptive contrastive learning network to extract real and complex information. Specifically, the encoder of PDCNet is designed as an extraction module for a real-convolutional composite complex convolutional network. Thirdly, a loss function is designed to effectively compute the real and complex high-level semantics of the T- and C-matrix's features separately. We conducted experiments with labels of different quantities in four data sets. The results indicate that our proposed model outperforms state-of-the-art contrastive learning models.

关键词： Feature extraction Contrastive learning Convolutional neural networks Vectors Scattering Land surface Synthetic aperture radar Data mining classification algorithms Training Complex convolutional networks contrastive learning land cover classification polarimetric synthetic aperture radar (PolSAR) positive pairs

来源：评论

学校读者我要写书评

暂无评论

Urdu Toxic Comment classification With PURUTT Corpus Development

引用

IEEE ACCESS 2025年 13卷 21635-21651页

作者： Saeed, Hafiz Hassaan Khalil, Tahir Kamiran, Faisal Informat Technol Univ Dept Comp Sci Lahore 54600 Pakistan

This study addresses the critical gap in toxic comment classification in Urdu, a widely spoken language devoid of high-quality standard datasets. To address this gap, we employed an existing labeled Roman Urdu (RU) corpus, which was developed originally for Roman Urdu toxic comment classification, and supplemented that corpus by adding its Urdu equivalent transliterations. The motivation behind such an extension is twofold: firstly, to provide a large comprehensive dataset for the classification of toxic comments in Urdu;secondly, to facilitate bidirectional transliteration between Urdu and RU, however, transliteration is currently outside the scope of this study and is envisioned as a future research direction. We introduce the extended corpus as PURUTT (Parallel Urdu and Roman Urdu Corpus for Toxic Comments and Transliteration), boasting 72,771 labeled comments as parallel comments in both Urdu and Roman Urdu scripts. Specific to Urdu toxic comment classification, our methodology begins by training those classification models that were trained on the original Roman Urdu corpus. We leverage pre-trained Word2Vec and FastText Urdu word embeddings to evaluate model performance through transfer learning. Furthermore, we fine-tune five multilingual large language models capitalizing on their inherent multilingual capabilities. To further enhance the classification performance, this study proposes an ensemble approach that aggregates the strengths of multiple base models. Our extensive empirical validation demonstrates the superiority of the ensemble model, achieving a state-of-the-art F1-score of 91.65% on PURUTT, setting a benchmark F1-score on PURUTT corpus for Urdu toxic comment classification.

关键词： Translation Internet Encoding Deep learning Cyberbullying Bidirectional control Online services Guidelines Encyclopedias classification algorithms Urdu Urdu parallel corpus Urdu toxic comments Urdu toxic comment classification toxic comment classification transfer learning ensemble

来源：评论

学校读者我要写书评

暂无评论

Dual Strategy Based Improved Swarm Intelligence and Stacked LSTM With Residual Connection for Land Use Land Cover classification

引用

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 4188-4198页

作者： Nagaraju, Vinaykumar Vajjanakurike Jayachandra, Ananda Babu Stateczny, Andrzej Yogesh, Swathi Holalu Lingaraju, Raviprakash Madenur Veeranna, Balaji Prabhu Baluvaneralu Malnad Coll Engn Dept Informat Sci & Engn Hassan 573202 India Visvesvaraya Technol Univ Belagavi 590018 India Malnad Coll Engn Dept Comp Sci & Engn Artificial Intelligence & Mac Hassan 573202 India Gdynia Maritime Univ Dept Nav PL-81225 Gdynia Poland Kalpataru Inst Technol Dept Artificial Intelligence & Machine Learning Tiptur 572201 India Visvesvaraya Technol Univ Belagavi 590018 India

Land use land cover (LULC) classification using satellite images is crucial for land-use inventories and environment modeling. The LULC classification is a difficult task because of the high dimensional feature space, which affects the classification accuracy. This article proposes a dual strategy-based bald eagle search (DSBES) algorithm and stacked long short-term memory (LSTM) with residual connection for LULC classification. The dual strategy includes adaptive inertia weight and phasor operator strategy to select relevant features from the feature subset. The stacked LSTM contains multiple layers stacked on top of each other to capture high-level temporal data. By integrating residual connection with stacked LSTM, gradient flow is enabled directly among long sequences, reducing the vanishing gradient issue and fastening the convergence rate. The DSBES and stacked LSTM with residual connection performance are examined in terms of metrics of accuracy, precision, sensitivity, specificity, f1-score, and computational time. The DSBES and stacked LSTM with residual connection achieve higher accuracy values of 99.71%, 98.66%, 97.59%, and 99.24% for UCM, AID, NWPU, and EuroSAT datasets, respectively, as compared to VGG19 and optimal guidance whale optimization algorithm-bidirectional long short-term memory.

关键词： Feature extraction classification algorithms Land surface Accuracy Data mining Remote sensing Residual neural networks Convergence Transfer learning Sports Dual strategy-based bald eagle search (DSBES) land use land cover (LULC) phasor operator regional variability residual connection stacked long short-term memory (LSTM)

来源：评论

学校读者我要写书评

暂无评论

Comparative Analysis of Deep Learning Methods for classification of Ablated Regions in Hyperspectral Images of Atrial Tissue

引用

IEEE ACCESS 2025年 13卷 35029-35047页

作者： Gabrielyan, Yeva Hunanyan, Arpi Bezirganyan, Sona Davtyan, Lusine Avetisyan, Ani Sarvazyan, Narine Butavyan, Aram Yeghiazaryan, Varduhi Amer Univ Armenia Akian Coll Sci & Engn Yerevan 0019 Armenia LA Orbeli Inst Physiol NAS RA Yerevan 0028 Armenia George Washington Univ George Washington Sch Med & Hlth Sci Dept Pharmacol & Physiol Washington DC 20052 USA

Radiofrequency ablation (RFA) is used to treat atrial fibrillation (AF). Viability gaps between ablated regions can lead to AF recurrence;thus, correctly detecting RFA lesions is important for successful treatment. Hyperspectral imaging (HSI) has been previously shown to aid in visualizing RFA-affected areas. However, automated classification of ablated/unabled tissue in HSI remains open. We comparatively analyze state-of-the-art deep learning models for HSI classification from the field of remote sensing. We group them based on model architecture and utilization of spectral and spatial contexts. We deploy a pre-clinical HSI dataset obtained from excised left atrial tissue of four porcine and five bovine samples. The 45 classification algorithms are benchmarked through two types of experiments: objective and subjective. For objective evaluation, per HSI cube, we limit the training and testing to rectangular areas of known classes (ablated/unablated), as reliable classification masks for whole HSI cubes are not available. For both porcine and bovine data, models incorporating graph neural networks demonstrate superior performance. The best eight algorithms are further compared in a subjective experiment. We ask seventeen participants to rate RGB visualizations of model outputs. In this experiment, the performance of an algorithm is assessed on the whole HSI cube, and is subject to participants' interpretation of the extent of RFA-affected tissue. Our comparative analysis allows us to observe the suitability of different deep learning architectures for HSI classification of affected tissue. These efforts contribute towards the potential future utilization of HSI for surgical guidance in RFA or similar procedures.

关键词： Transformers Three-dimensional displays Feature extraction Hyperspectral imaging Deep learning Data models classification algorithms Vectors Bovine Nearest neighbor methods Atrial tissue ablation comparative analysis convolutional neural networks deep learning graph neural networks hyperspectral image classification survey transformer networks

来源：评论

学校读者我要写书评

暂无评论

Class Imbalance in Network Traffic classification: An Adaptive Weight Ensemble-of-Ensemble Learning Method

引用

IEEE ACCESS 2025年 13卷 26171-26192页

作者： Abbasi, Mahmoud Florez, Sebastian Lopez Shahraki, Amin Taherkordi, Amir Prieto, Javier Corchado, Juan M. Univ Salamanca BISITE Res Grp Salamanca 37007 Spain Univ Oslo Dept Informat N-0373 Oslo Norway Osaka Inst Technol Fac Engn Dept Elect Informat & Commun Osaka 5358585 Japan

Network Traffic classification (NTC) serves as a crucial element in network management, and the rapid progress in machine learning has inspired the utilization of learning methods to discern network traffic. The inherent characteristics of network traffics result in uneven class distributions when datasets are shaped, creating a phenomenon known as class imbalance. This phenomenon has garnered growing attention across various research fields. Despite encountering performance setbacks attributed to class imbalance, this challenge remains inadequately examined in the realm of network traffic classification. This paper introduces the Adaptive Weight Ensemble-of-Ensemble Learning (AWEE) method as an innovative solution to this challenge. The AWEE integrates multiple ensemble layers with a dynamic weight adjustment mechanism, showcasing the collaborative intelligence of diverse modeling strategies. Using a sliding window-based validation approach, the model enhances adaptability and robustness to address concept drift in dynamic data streams. Experimental studies on benchmark datasets demonstrate the superior performance of AWEE (achieving an outstanding accuracy rate of over 98%), highlighting its effectiveness in handling class imbalance challenges across diverse network traffic scenarios. AWEE, outperforms competitive methods, including algorithmic-level, cost-sensitive, and data-level techniques, showcasing its robustness and superior performance in addressing class imbalance challenges across a wide range of network traffic scenarios.

关键词： Telecommunication traffic Ensemble learning Robustness Adaptation models Accuracy classification algorithms Training Streams Costs Boosting Applications of artificial intelligence class imbalance ensemble learning implemented artificial intelligence network traffic classification

来源：评论

学校读者我要写书评

暂无评论

Quantum Feature-Empowered Deep classification for Fast Mangrove Mapping

引用

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 2025年 63卷

作者： Lin, Chia-Hsiang Tang, Po-Wei Huete, Alfredo R. Natl Cheng Kung Univ Dept Elect Engn Tainan 70101 Taiwan Natl Cheng Kung Univ Miin Wu Sch Comp Tainan 70101 Taiwan Natl Cheng Kung Univ Inst Comp & Commun Engn Dept Elect Engn Tainan 70101 Taiwan Univ Technol Sydney Sch Life Sci Sydney NSW 2007 Australia

A mangrove mapping (MM) algorithm is an essential classification tool for environmental monitoring. The recent literature shows that compared with other index-based MM methods that treat pixels as spatially independent, convolutional neural networks (CNNs) are crucial for leveraging spatial continuity information, leading to improved classification performance. In this work, we go a step further to show that quantum features provide radically new information for CNN to further upgrade the classification results. Simply speaking, CNN computes affine-mapping features, while quantum neural network (QNN) offers unitary-computing features, thereby offering a fresh perspective in the final decision-making (classification). To address the challenging MM problem, we design an entangled spatial-spectral quantum feature extraction module. Notably, to ensure that the quantum features contribute genuinely novel information (unaffected by traditional CNN features), we design a separate network track consisting solely of quantum neurons with built-in interpretability. The extracted pure quantum information is then fused with traditional feature information to jointly make the final decision. The proposed quantum-empowered deep network (QEDNet) is very lightweight, so the improvement does come from the cooperation between CNN and QNN (rather than parameter augmentation). Extensive experiments will be conducted to demonstrate the superiority of QEDNet.

关键词： Feature extraction Convolutional neural networks Indexes classification algorithms Vegetation mapping Quantum computing Forests Deep learning Carbon Satellite images Image classification mangrove mapping (MM) quantum computing quantum deep learning (DE) Sentinel-2 image sustainable development goals (SDGs)

来源：评论

学校读者我要写书评

暂无评论

An Explainable Artificial Intelligence Model for the classification of Breast Cancer

引用

IEEE ACCESS 2025年 13卷 5618-5633页

作者： Khater, Tarek Hussain, Abir Bendardaf, Riyad Talaat, Iman M. Tawfik, Hissam Ansari, Sam Mahmoud, Soliman Univ Sharjah Elect Engn Dept Sharjah U Arab Emirates Liverpool John Moores Univ Fac Engn Sch Comp Sci & Math Liverpool L3 3AF England Univ Sharjah Clin Sci Dept Sharjah U Arab Emirates Univ Hosp Sharjah Dept Oncol Sharjah U Arab Emirates Alexandria Univ Fac Med Pathol Dept Alexandria 5424041 Egypt Univ Khorfakan Sharjah U Arab Emirates

Breast cancer is the most common cancer among women and globally affects both genders. The disease arises due to abnormal growth of tissue formed of malignant cells. Early detection of breast cancer is crucial for enhancing the survival rate. Therefore, artificial intelligence has revolutionized healthcare and can serve as a promising tool for early diagnosis. The present study aims to develop a machine-learning model to classify breast cancer and to provide explanations for the model results. This could improve the understanding of the diagnosis and treatment of breast cancer by identifying the most important features of breast cancer tumors and the way they affect the classification task. The best-performing machine-learning model has achieved an accuracy of 97.7% using k-nearest neighbors and a precision of 98.2% based on the Wisconsin breast cancer dataset and an accuracy of 98.6% using the artificial neural network with 94.4% precision based on the Wisconsin diagnostic breast cancer dataset. Hence, this asserts the importance and effectiveness of the proposed approach. The present research explains the model behavior using model-agnostic methods, demonstrating that the bare nuclei feature in the Wisconsin breast cancer dataset and the area's worst feature Wisconsin diagnostic breast cancer dataset are the most important factors in determining breast cancer malignancy. The work provides extensive insights into the particular characteristics of the diagnosis of breast cancer and suggests possible directions for expected investigation in the future into the fundamental biological mechanisms that underlie the disease's onset. The findings underline the potential of machine learning to enhance breast cancer diagnosis and therapy planning while emphasizing the importance of interpretability and transparency in artificial intelligence-based healthcare systems.

关键词： Breast cancer Medical services Artificial neural networks Predictive models Machine learning Feature extraction classification algorithms Artificial intelligence breast cancer explainable machine learning model-agnostic permutation importance partial dependence plot SHAP supervised 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：