检索结果-内蒙古大学图书馆

An Endmember-Oriented Transformer Network for Bundle-Based Hyperspectral Unmixing

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 2025年 63卷

作者： Xiang, Shu Li, Xiaorun Chen, Shuhan Zhejiang Univ Coll Elect Engn Hangzhou 310027 Peoples R China

In recent years, researchers have focused on mitigating the impact of spectral variability (SV) on unmixing performance, leading to the development of various deep-learning-based unmixing networks. Currently, most unmixing networks that account for SV mainly rely on probabilistic generative models, which lacks specific constraints for SV and suffers from the instability of solutions generated by probabilistic models. This results in the generated endmembers or SV components lacking clear physical meaning. To avoid the problems above, we propose an endmember-oriented Transformer network (EOT-Net) that leverages the advantages of endmember bundles to introduce variability while providing stable endmember results with clear physical meaning. We design an endmember-oriented Transformer (EOT) to capture endmember-specific features through directional subspace projection and a low-redundancy attention (LRA) mechanism. Subsequently, the proposed network is divided into two branches: endmember generation and abundance estimation, to process endmember-specific features. In the endmember generation branch, endmember-specific features are transformed into intraclass weights that are used to combine signatures within the bundles, and a set of endmembers is generated for each pixel. In the abundance estimation branch, endmember-specific features are integrated using a heterogeneous information fusion (HIF) module that leverages the spatial distribution heterogeneity of the endmembers, ultimately producing the abundance results. We applied the proposed algorithm to both synthetic and real datasets, and the experimental results demonstrated the model's superiority.

关键词： Feature extraction Transformers Hyperspectral imaging Attention mechanisms Estimation classification algorithms Generators Data mining Change detection algorithms Accuracy Autoencoder (AE) network endmember bundle hyperspectral unmixing (HU) spectral variability (SV) Transformer

来源：评论

学校读者我要写书评

暂无评论

Fast Subspace and DOA Estimation Method for the Case of High-Dimensional and Small Samples

引用

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY 2025年第3期74卷 3958-3975页

作者： Zhang, Xuejun Feng, Dazheng Zheng, Weixing Xidian Univ Natl Lab Radar Signal Proc Xian 710071 Peoples R China Western Sydney Univ Sch Comp Data & Math Sci Sydney NSW 2751 Australia

It is well-known that classical direction of arrival (DOA) estimation methods work well in the case of large samples. However, these methods may be theoretically invalid in the case of small samples, which frequently occur in large array systems. Such a large array has two effects: i) The number of samples is relatively quite small, and ii) the dimension of samples is very large. To handle the above problems, a more appropriate method for solving DOA estimators in the case of high-dimensional and small samples is proposed in this paper. First, considering the special structure of received samples, an alternative well-estimated second-order statistic, known as the Gram matrix, is originally constructed to better utilize the spatial and statistical information of signals and noise contained by small samples. Second, two novel methods for estimating the number of targets are derived by combining the Gram matrix and information-theoretic criteria. Third, a novel object function and the corresponding algorithm based on the Gram matrix are designed to estimate the signal subspace more efficiently, and then the DOAs of targets are obtained by multiple signal classification methods. In particular, the theoretical analysis indicates that the improved signal subspace estimation algorithm only needs to decompose the low-dimensional Gram matrix instead of the high-dimensional sample covariance matrix. Finally, simulation results are provided to demonstrate the high accuracy and lower computational complexity of the proposed methods in the case of high-dimensional and small samples.

关键词： Direction-of-arrival estimation Estimation Prediction algorithms Noise Covariance matrices classification algorithms Signal processing algorithms Multiple signal classification Accuracy Vectors High-dimensional and small samples gram matrix direction of arrival number of targets estimation method subspace estimation algorithm

来源：评论

学校读者我要写书评

暂无评论

Vision Transformers Versus Convolutional Neural Networks: Comparing Robustness by Exploiting Varying Local Features

引用

IEEE ACCESS 2025年 13卷 65232-65245页

作者： Arslanoglu, Muhammed Cihad Albayrak, Abdulkadir Acar, Huseyin Dicle Univ Dept Elect & Elect Engn TR-21280 Diyarbakir Turkiye Dicle Univ Dept Comp Engn TR-21280 Diyarbakir Turkiye Mayo Clin Hosp Operat Dept Generat AI Program Operat Rochester MN 55905 USA

Robustness of deep learning models is a crucial point that should be evaluated in computer vision tasks. Deep learning models must be able to produce the same results under different conditions such as brightness, image rotation, etc. Deep learning models must have high level robustness to be reliable to address real world problems. Unreliable systems may be harmful for critical systems such as security, healthcare, and military. In this study, it is aimed to evaluate image rotation robustness of Convolutional Neural Networks (CNNs) and Transformer-based deep learning models: MobileNetV2, Residual Network 18 (ResNet18), Visual Geometry Group 16 (VGG16), EfficientNetB1, Vision Transformer (ViT), Data-Efficient Image Transformers (DeiT), and Pooling-based Vision Transformer (PiT). Canadian Institute for Advanced Research 100 classes (CIFAR100), Canadian Institute for Advanced Research 10 classes (CIFAR10) and California Institute of Technology 256 classes (Caltech256) datasets were used to evaluate the performance of the models. All deep learning models were trained with original train datasets. Image embeddings of train, test datasets, and their rotated forms were extracted. A Support Vector Classifier (SVC) was trained with embeddings of original dataset and concatenation of original and rotated form of train dataset. Performances on original test dataset and their rotated forms were evaluated. 10, 50, and 90 degrees rotated form of the datasets were used for rotated test dataset. F1-score metrics were used compare model performances. In conclusion, Transformer-based classification algorithms are more robust to image rotation than the CNN-based models.

关键词： Deep learning Robustness Transformers Feature extraction Computer vision Computational modeling Vectors Accuracy Training classification algorithms Convolutional neural networks vision transformers classification robustness feature extraction

来源：评论

学校读者我要写书评

暂无评论

Application of Big Data Technology to Assessments of Female Ovarian Reserve Dysfunction

引用

IEEE ACCESS 2025年 13卷 19408-19419页

作者： Xia, Ji'An Ma, Yunfei Wu, Yiyun Zhao, Youlin Ni, Haorang Liu, Xinyan Nanjing Vocat Univ Ind Technol Sch Comp & Software Nanjing 210023 Jiangsu Peoples R China Jiangsu Prov Hosp Chinese Med Dept Ultrasound Nanjing 210029 Jiangsu Peoples R China

This paper evaluated the potential application of big data technology to assessments of diminished ovarian reserve (DOR). The study enrolled 162 patients who underwent ovarian reserve function assessment for the first time in the Department of Ultrasound, Jiangsu Province Hospital of Chinese Medicine from January 2023 to December 2023. Patients were divided into normal ovarian reserve function (n = 68), early-stage DOR (n = 66), mid-stage DOR (n = 12), and late-stage DOR (n = 16) groups. Hadoop and Spark frameworks were used to build a big data platform, and the MLlib parallel machine learning library was used to implement three multivariate classification models-multilayer perceptron, one-vs-rest, and random forest classifiers-to classify and analyse the ovarian reserve function dataset and evaluate the platform's performance. In the big data platform, the random forest algorithm achieved the highest classification accuracy (89.47%), followed by the neural network (81.06%) and support vector machine (72.91%) methods. The random forest algorithm had the least time overhead for datasets smaller than 50 MB;for datasets exceeding 50 MB, the support vector machine algorithm had the least time overhead, followed by the random forest and neural network algorithms. The neural network algorithm's speedup ratio was lower than that of the other two algorithms for small datasets, but with increasing dataset size, its speedup ratio significantly exceeded those of the other two algorithms. The random forest algorithm showed substantial growth for large datasets.

关键词： Medical diagnostic imaging Big Data Machine learning algorithms Sparks classification algorithms Indexes Data models Random forests Biochemistry Support vector machines Diminished ovarian reserve three-dimensional power Doppler ultrasound medical big data machine learning classification algorithm performance evaluation

来源：评论

学校读者我要写书评

暂无评论

A Novel Expandable Borderline Smote Over-Sampling Method for Class Imbalance Problem

引用

IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING 2025年第5期37卷 2183-2199页

作者： Sun, Hao Li, Jianping Zhu, Xiaoqian Chinese Acad Sci Inst Sci & Dev Beijing 100190 Peoples R China Univ Chinese Acad Sci Sch Econ & Management Beijing 100190 Peoples R China

The class imbalance problem can cause classifiers to be biased toward the majority class and inclined to generate incorrect predictions. While existing studies have proposed numerous oversampling methods to alleviate class imbalance by generating extra minority class samples, these methods still have some inherent weaknesses and make the generated samples less informative. This study proposes a novel over-sampling method named the Expandable Borderline Smote (EB-Smote), which can address the weaknesses of existing over-sampling methods and generate more informative synthetic samples. In EB-Smote, not only minority class but also majority class is oversampled, and the synthetic samples are generated in the area between the selected minority and majority samples, which are close to the borderlines of their respective classes. EB-Smote can generate more informative samples by expanding the borderlines of minority and majority classes toward the actual decision boundary. Based on 27 imbalanced datasets and commonly used machine learning models, the experimental results demonstrate that EB-Smote significantly outperforms the other 8 existing oversampling methods. This study can provide theoretical guidance and practical recommendations to solve the crucial class imbalance problem in classification tasks.

关键词： Noise classification algorithms Accuracy Training Safety Noise measurement Sun Stochastic processes Overfitting Next generation networking Class imbalance oversampling machine learning classification decision boundary

来源：评论

学校读者我要写书评

暂无评论

Quick Unsupervised Hyperspectral Dimensionality Reduction for Earth Observation: A Comparison

IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING

引用

IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING 2025年 11卷 520-535页

作者： Lupu, Daniela Garrett, Joseph L. Johansen, Tor Arne Orlandic, Milica Necoara, Ion Natl Univ Sci & Technol Politehn Bucharest Automat Control & Syst Engn Dept Bucharest 060042 Romania POLITEHN Bucharest Precis Res Inst Bucharest 060042 Romania Norwegian Univ Sci & Technol Ctr Autonomous Marine Operat & Syst N-7491 Trondheim Norway Norwegian Univ Sci & Technol Dept Engn Cybernet N-7034 Trondheim Norway Norwegian Univ Sci & Technol Dept Elect Syst N-7491 Trondheim Norway Romanian Acad G Mihoc C Iacob Inst Math Stat & Appl Math Bucharest 050711 Romania

Dimensionality reduction can be applied to hyperspectral images so that the most useful data can be extracted and processed more quickly. This is critical in any situation in which data volume exceeds the capacity of the computational resources, particularly in the case of remote sensing platforms (e.g., drones, satellites), but also in the case of multi-year datasets. Moreover, the computational strategies of unsupervised dimensionality reduction often provide the basis for more complicated supervised techniques. In this work, eight unsupervised dimensionality reduction algorithms are tested on hyperspectral data from the HYPSO-1 earth observation satellite. Each particular algorithm is chosen to be representative of a broader collection of methods. Our extensive experiments probe the computational complexity, reconstruction accuracy, signal clarity, sensitivity to artifacts, and effects on target detection and classification of the different algorithms. No algorithm consistently outperformed the others across all tests, but some general trends regarding the characteristics of the algorithms did emerge. With half a million pixels, computational time requirements of the methods varied by 5 orders of magnitude, and the reconstruction error varied by about 3 orders of magnitude. A relationship between mutual information and artifact susceptibility was suggested by the tests. The relative performance of the algorithms differed significantly between the target detection and classification tests. Overall, these experiments both show the power of dimensionality reduction and give guidance regarding how to evaluate a technique prior to incorporating it into a processing pipeline.

关键词： Principal component analysis Dimensionality reduction classification algorithms Hyperspectral imaging Object detection Covariance matrices Matrix decomposition Image reconstruction Accuracy Pipelines Earth observation satellite images hyperspectral images linear and non-linear transformation of data unsupervised dimensionality reduction methods

来源：评论

学校读者我要写书评

暂无评论

Subpixel Target Detection in Hyperspectral Imaging Using a Deep Neural Network With a Variable Stepsize Gradient Descent Method

引用

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 7707-7727页

作者： Lo, Edisanter Josset, Damien B. Susquehanna Univ Dept Math & Comp Sci Selinsgrove PA 17870 USA NASA Ocean Sci Div US Naval Res Lab Stennis Space Ctr John C Stennis Space Ctr MS 39529 USA

The main difficulty in using artificial neural networks, which are designed for classification, to detect a rare subpixel target in hyperspectral imaging is that there is typically only one actual target pixel available for training the neural networks, which require a large sample of actual target pixels to train the target class. The first current detection algorithm for subpixel target detection is based on a single-layer neural network and gradient descent method with variable stepsize to solve the optimization problem, and the second one is based on a multilayer neural network and gradient descent method with variable stepsize. The objective of this article is to extend the current algorithms by developing a detection algorithm for subpixel target detection using a deep neural network with two hidden layers and the gradient descent method with variable stepsize instead of fixed stepsize. Implementing the gradient descent method with variable stepsize can reduce computational cost by improving convergence and speed of convergence. The decision boundary is also analyzed and is linear for the single-layer neural network and nonlinear for the multilayer neural network and deep neural neural network with two hidden layers. Experimental results from two hyperspectral images, one with simulated subpixel target pixels for training and validating the algorithm and the other with simulated subpixel target pixels for training and actual subpixel target pixels for validation, have shown that the proposed algorithm can perform better than conventional algorithms, which are based on generalized likelihood ratio test.

关键词： Neural networks Object detection Vectors Hyperspectral imaging Linear programming Training classification algorithms Artificial neural networks Convergence Earth neural network stochastic gradient descent (SGD) target detection

来源：评论

学校读者我要写书评

暂无评论

A Dynamic Selection Hybrid Model for Advancing Thyroid Care With BOO-ST Balancing Method

引用

IEEE ACCESS 2024年 12卷 78641-78656页

作者： Khalid, Majdi Shamrat, F. M. Javed Mehedi Alshanbari, Hanan Farrash, Majed Qadah, Thamir M. Umm Al Qura Univ Coll Comp Dept Comp Sci & Artificial Intelligence Makkah City 21955 Saudi Arabia Univ Malaya Dept Comp Syst & Technol Fed Terr Kuala Lumpur Kuala Lumpur 50603 Malaysia Umm Al Qura Univ Coll Comp Dept Comp & Network Engn Makkah City 21955 Saudi Arabia

Recently, thyroid disease has been a leading cause of mortality, underscoring the importance of early diagnosis to mitigate its impact. Researchers have randomly employed static selection ensemble methods aiming to forecast the disease in its initial stages. However, the use of such ensemble methods in healthcare diagnosis poses challenges related to performance consistency and potential mismatches with new data characteristics. Hence, this paper proposes a novel approach by introducing the Dynamic Selection Hybrid Model (DSHM) that leverages the most effective conventional classifiers using an appropriate ensemble method. Instead of going the conventional way, we evaluate various baseline classifiers to demonstrate their impact on the characteristics selected by two robust feature selection techniques. This evaluation employs an explainable AI (XAI) method, Permutation Feature Importance (PFI), and selects the most effective classifiers based on their characteristics impact. Then the selected classifiers are integrated by an appropriate ensemble method, based on a comparative evaluation between four efficient ensemble methods. Allowing the proposed DSHM to dynamically adjust its composition based on selecting conditions can potentially achieve robust performance by better adaptability on unseen data. Before the training DSHM, the methodology begins by addressing the dataset's imbalance issue using an effective data balancing method BOO-ST. To demonstrate the superiority of DSHM, various performance evaluation matrices, and a statistical test are employed. The experimental results reveal the effectiveness of our proposed DSHM, outperformed with an impressive 99.33% accuracy. Finally, to enhance transparency, trust, and patient outcomes, we applied the Local Interpretable Model-agnostic Explanations (LIME) to explain DSHM-provided outcomes. With a robust classification performance, our proposed DSHM aims to explain its outcomes, contributing to improved clinical deci

关键词： Thyroid Ensemble learning Diseases Adaptation models Support vector machines classification tree analysis classification algorithms Dynamic selection hybrid classifier feature selection explainable AI thyroid disease

来源：评论

学校读者我要写书评

暂无评论

Abnormal Electricity Detection of Users Based on Improved Canopy-Kmeans and Isolation Forest algorithms

引用

IEEE ACCESS 2024年 12卷 99110-99121页

作者： Wang, Jianyuan Li, Xiaoyao Northeast Elect Power Univ Key Lab Modern Power Syst Simulat & Control & Rene Minist Educ Jilin 132012 Peoples R China

Aiming at the existing user abnormal electricity consumption detection methods that have the problem of difficult classification of user similar electricity consumption patterns, this paper proposes an unsupervised isolation forest abnormal electricity consumption detection model based on the Canopy-Kmeans algorithm with weighted density improvement. To start, we propose a composite parameter analysis method for user electricity consumption patterns, volatility, trends, and correlations using Irish smart meter data. This method involves joint data cleaning, interpolation, and feature construction. Additionally, principal component analysis is introduced to fuse features across layers and reduce dimensionality in user electricity consumption. Subsequently, we introduce the weighted density improvement Canopy-Kmeans clustering algorithm. This algorithm determines the K value and clustering centers using the maximum weight product method, based on definitions of sample density, average intra-class sample distance, and inter-class distance in the multilayer fusion feature data. Finally, we propose a fusion mechanism of weighted density improvement Canopy-Kmeans and isolation forest algorithms to jointly construct a model for detecting abnormal power usage based on multilayer fusion feature data analysis. The results demonstrate that multilayer fusion feature parameters vary in size and discretization among different user types, enabling classification of users with diverse electricity consumption patterns. Moreover, the anomaly detection model based on multilayer fusion feature data analysis improves accuracy rates, recall rates, and F1 scores compared to other algorithms.

关键词： Electricity Clustering algorithms classification algorithms Anomaly detection Forestry Prediction algorithms Feature extraction Detection algorithms Energy consumption Unsupervised learning Abnormal detection of electricity consumption by users Canopy-Kmeans algorithm isolation forest algorithm unsupervised learning weighted density

来源：评论

学校读者我要写书评

暂无评论

SmartClus: Mobility and Energy-Aware Autonomous Aerial Vehicles' Clustering for B5G Cellular Networks

引用

IEEE ACCESS 2025年 13卷 59713-59724页

作者： Ahmed, Rabia Kumbhar, Farooque Hassan Mesbah, Wessam Elgabli, Anis Hasan Syed, Farrukh FAST Natl Univ Comp & Emerging Sci Sch Comp Karachi 44000 Pakistan King Fahd Univ Petr & Minerals Ctr Commun Syst & Sensing Dhahran 31261 Saudi Arabia King Fahd Univ Petr & Minerals Elect Engn Dept Dhahran 31261 Saudi Arabia King Fahd Univ Petr & Minerals Ind & Syst Engn Dept Dhahran 31261 Saudi Arabia

Autonomous aerial vehicles (AAVs) are key enablers for various beyond-5G (B5G) applications, including remote sensing, monitoring, emergency response, and delivery systems. However, communication with AAVs faces challenges such as energy constraints, dynamic environments, weather sensitivity, and high mobility. Ensuring connectivity for fast-moving AAVs while minimizing energy consumption to extend network lifetime remains critical. This paper presents SmartClus, a lightweight and dynamic clustering framework designed to address AAV group mobility and resource sharing. To enable efficient cluster formations, a multi-metric cluster head (CH) fitness and mobility similarity score process is formulated, that considers communication quality, spatial proximity, velocity, and direction. The extensive performance evaluation in the NS3 network simulator shows that SmartClus outperforms two state-of-the-art schemes. The results demonstrated that the proposed approach nearly doubled the average throughput, and reduced transmission delays by 2 to 4 seconds. Moreover, SmartClus achieved a packet delivery ratio (PDR) of 70% to 90% compared to previous schemes with 50% to 70% and 60% to 80%, and extended average CH lifetime (being CH longer and requiring no re-clustering).

关键词： Autonomous aerial vehicles Clustering algorithms Resource management Interference Heuristic algorithms Accuracy Location awareness classification algorithms Vehicle dynamics Spectral efficiency Clustering cooperative communications mobility-aware autonomous aerial vehicles beyond 5G cellular 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：