检索结果-内蒙古大学图书馆

Spatial-Spectral Enhancement and Fusion Network for Hyperspectral Image classification With Few Labeled Samples

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 2025年 63卷

作者： Liu, Shuang Fu, Chuan Duan, Yule Wang, Xiaopan Luo, Fulin Chongqing Univ Coll Comp Sci Chongqing 400044 Peoples R China Henan Univ Technol Coll Informat Sci & Engn Zhenzhou 450001 Peoples R China Chongqing Geomat & Remote Sensing Applicat Ctr Chongqing 401147 Peoples R China

Deep learning has shown great potential in hyperspectral image (HSI) classification. However, training these models usually requires a large amount of labeled data. Since the collection of pixel-level annotations for HSIs is laborious and time-consuming, developing algorithms that can yield good performance in a small sample size situation is of great significance. Therefore, many research works focus on building a deep learning model for HSI classification with few labeled samples. However, prevalent solutions are unsatisfactory in feature discrimination and model overfitting, which greatly limits their performance. To remedy these drawbacks, we propose a novel spatial-spectral enhancement and fusion network for hyperspectral image classification with few labeled samples, named SSEFN. Specifically, we design a spatial-spectral enhancement strategy (SSES) to boost the feature discrimination from spatial and spectral perspectives, which enables the model to learn more easily with fewer samples. In addition, we propose an adaptive decision fusion (ADF) module to fuse the decisions of all enhanced features. Since each decision prediction may have a different trend of overfitting, combining multiple predictions alleviates the overfitting. Extensive experiments are conducted on four diverse hyperspectral image datasets. The results show that our method significantly outperforms the state-of-the-art approaches on all datasets and demonstrates the effectiveness and superiority of the proposed model for hyperspectral image classification with few labeled samples. Codes are available at https://***/liushuang963/SSEFN.

关键词： Overfitting Hyperspectral imaging Feature extraction Image classification Training Data models Adaptation models classification algorithms Transfer learning Semisupervised learning Decision fusion few labeled samples hyperspectral image (HSI) classification spatial-spectral feature enhancement

来源：评论

学校读者我要写书评

暂无评论

Investigating the Effective Dynamic Information of Spectral Shapes for Audio classification

引用

IEEE TRANSACTIONS ON MULTIMEDIA 2025年 27卷 1114-1126页

作者： Chen, Liangwei Zhou, Xiren Chen, Qiuju Xiong, Fang Chen, Huanhuan Univ Sci & Technol China Sch Artificial Intelligence & Data Sci Hefei 230027 Peoples R China Univ Sci & Technol China Sch Comp Sci & Technol Hefei 230027 Peoples R China Univ Sci & Technol China Sch Informat Sci & Technol Hefei 230027 Peoples R China Cent South Univ Xiangya Hosp Natl Clin Res Ctr Geriatr Dis Dept Otolaryngol Head & Neck Surg Changsha 410078 Peoples R China

The spectral shape holds crucial information for Audio classification (AC), encompassing the spectrum's envelope, details, and dynamic changes over time. Conventional methods utilize cepstral coefficients for spectral shape description but overlook its variation details. Deep-learning approaches capture some dynamics but demand substantial training or fine-tuning resources. The Learning in the Model Space (LMS) framework precisely captures the dynamic information of temporal data by utilizing model fitting, even when computational resources and data are limited. However, applying LMS to audio faces challenges: 1) The high sampling rate of audio hinders efficient data fitting and capturing of dynamic information. 2) The Dynamic Information of Partial Spectral Shapes (DIPSS) may enhance classification, as only specific spectral shapes are relevant for AC. This paper extends an AC framework called Effective Dynamic Information Capture (EDIC) to tackle the above issues. EDIC constructs Mel-Frequency Cepstral Coefficients (MFCC) sequences within different dimensional intervals as the fitted data, which not only reduces the number of sequence sampling points but can also describe the change of the spectral shape in different parts over time. EDIC enables us to implement a topology-based selection algorithm in the model space, selecting effective DIPSS for the current AC task. The performance on three tasks confirms the effectiveness of EDIC.

关键词： Mel frequency cepstral coefficient Data models Spectral shape Computational modeling Feature extraction Fitting Music classification algorithms Training Multiple signal classification Learning in the model space dynamic information of the spectral shape audio classification mel-frequency cepstral coefficients echo state network

来源：评论

学校读者我要写书评

暂无评论

Vehicle classification Based on Multi-Frequency Resistance and Reactance Magnetic Profiles

引用

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 2025年第4期26卷 5322-5331页

作者： Marszalek, Zbigniew Konior, Tomasz Izydorczyk, Jacek Szulik, Mateusz Duda, Krzysztof AGH Univ Krakow Dept Measurement & Elect PL-30059 Krakow Poland Univ Bielsko Biala Fac Management & Transport Dept Transport PL-43309 Bielsko Biala Poland APM PRO PL-43300 Bielsko Biala Poland Silesian Tech Univ Fac Automat Control Elect & Comp Sci Dept Telecommun & Teleinformat PL-44100 Gliwice Poland

This paper presents the application of inductive loop (IL) sensor technology to classify vehicles in traffic lanes. Two wide and two slim IL sensors were installed in the traffic lane. The wide and slim IL sensors feature distinct structural designs and varying levels of sensitivity. An advanced multi-frequency impedance measurement (MFIM) system was used to operate the IL sensors. For a passing vehicle, the impedance of every IL sensor at three different operating frequencies is computed and finally recorded at a sampling frequency of 1 kHz. Each of the 12 recorded signals provides a complex-value vehicle magnetic profile (VMP). Based on the VMPs from two IL sensors positioned one after the other, an accurate measurement of vehicle speed is obtained. Furthermore, the system can capture images of vehicles. A reference database of VMPs was created for various vehicle categories. The software selects 10 statistical features from each real and imaginary VMP part. Eight machine learning algorithms were implemented using ready-made Python3 implementations. Cross-validation accuracy was tested for five feature configurations, including slim and wide IL sensors. The Random Forest (RF) algorithm, utilizing 20 features from the complex VMP, achieved an accuracy of 99.8 % for the wide IL sensor. No errors were made by the Voting Classifier and RF algorithm when they incorporated a fusion of features from complex VMPs with MFIM system, utilizing both slim and wide IL sensors.

关键词： Accuracy classification algorithms Impedance Magnetic resonance imaging Resistance Motorcycles classification tree analysis Vectors Software Roads Vehicle magnetic profile multi-frequency impedance inductive loop vehicle classification machine learning logistic regression naive Bayes k-nearest neighbors support vector classifier decision tree random forest voting classifier artificial neural network

来源：评论

学校读者我要写书评

暂无评论

New Class Detection in Network Traffic classification Using Confidence Information Embedded Cascade Structure

引用

IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING 2025年第3期12卷 1692-1706页

作者： Lu, Haotian Dong, Yuning Wu, Zhiyuan Wei, Hua-Liang Lu, Guanming Nanjing Univ Posts & Telecommun Sch Commun & Informat Engn Nanjing 210003 Peoples R China Univ Sheffield Dept Automat Control & Syst Engn Sheffield S1 3JD England

Network traffic classification plays an important role in network management. With continued emergence of new applications, classifiers need to deal with unknown classes in an open set environment. However, the available open set flow recognition methods cannot well balance the performance of new class detection and the fine-grained classification of known classes. Moreover, these methods could pursue high accuracy at the cost of the classification speed. To address these problems, this paper proposes an unknown network traffic detection method based on confidence (difference) and a cascade structure, by analyzing the confidence distributions of both the known and new classes. The proposed method works as follows. Firstly, it uses a cascade structure to detect new class samples (having high confidence) which are difficult to identify using existing methods;secondly, it employs the maximum confidence difference to classify the new and known classes. In order to better detect new classes with high confidence, an algorithm is designed to select the pseudo-negative samples from the unlabelled dataset with an adaptive threshold. The proposed method is evaluated on real-world datasets. The results show that compared with the state-of-the-art methods, the proposed method can significantly improve the overall accuracy and the classification latency is also greatly reduced.

关键词： Accuracy Telecommunication traffic Training Support vector machines classification algorithms Generative adversarial networks Deep learning Data mining Adversarial machine learning Vectors Open set flow recognition confidence difference new class detection unlabelled dataset network traffic classification

来源：评论

学校读者我要写书评

暂无评论

A Fuzzy-Probabilistic Representation Learning Method for Time Series classification

引用

IEEE TRANSACTIONS ON FUZZY SYSTEMS 2024年第5期32卷 2940-2952页

作者： Erazo-Costa, Fabricio Javier Silva, Petronio C. L. Guimaraes, Frederico Gadelha Univ Fed Minas Gerais UFMG Grad Program Elect Engn BR-31270901 Belo Horizonte Brazil Univ Fed Ouro Preto UFOP Dept Elect Engn BR-35931008 Joao Monlevade Brazil IFNMG Fed Inst Northern Minas Gerais BR-39480000 Januaria Brazil Univ Fed Minas Gerais UFMG Dept Comp Sci BR-31270901 Belo Horizonte Brazil

Time series classification (TSC) is a supervised task in which time series data are associated with predefined classes. Time ordering and correlations of the samples should be considered and observations can have different lengths. Therefore, learning compact and useful representations is an important approach for TSC. Fuzzy time series (FTS) methods have the ability to find temporal patterns and transition rules. Probabilistic weighted FTS (PWFTS) uses the empirical probabilities to find the model of a TS. Despite its good performance, this method has been used exclusively for forecasting problems. Therefore, in this study we propose a novel TSC method using the PWFTS representation learning weights as features in conjunction to a classifier algorithm. We show that PWFTS method is able to learn useful representations for time series in low computational time. Our algorithm was tested in 24 datasets available in the UCR TSC archive and compared with state-of-the-art methods: fuzzy cognitive maps, Rocket, and TS2Vec in combination to the classifiers random forest and support vector machine. The classification accuracy and execution time required to find the features were evaluated. The results show that our fuzzy-probabilistic representation learning method is similar in accuracy to the other methods, and the time consumption for PWFTS computation was two to five orders of magnitude faster than the other methods. The results suggest that our approach can be used in time sensitive applications and in systems with hardware limitations.

关键词： Time series analysis classification algorithms Computational modeling Rockets Forecasting Training Data models Fuzzy temporal patterns (FTPs) fuzzy time series (FTS) representation learning time series classification (TSC)

来源：评论

学校读者我要写书评

暂无评论

Special Vehicle classification Algorithm-Based System for Dedicated Parking Zone Violation Detection in South Korea

引用

IEEE ACCESS 2025年 13卷 7883-7901页

作者： Park, Hyunseong Kim, Kapyeol Jeong, Incheol Jung, Jungil Cho, Jinsoo Gachon Univ Coll IT Dept IT Convergence Engn Seongnam Si 13120 South Korea PCT Co Ltd Seongnam Si 13449 South Korea Gachon Univ Dept Comp Engn Seongnam Si 13120 South Korea

To address the problem of managing dedicated parking zones arising from the increasing number of electric vehicles and vehicles for the physically challenged, this paper proposes a license plate recognition (LPR)-based parking control system that combines the YOLO and MobileNet algorithms. These two algorithms are designed for real-time object detection and efficient preprocessing, respectively, and can operate in real time in resource-constrained edge-device environments. In tests using data from more than 51,000 vehicles, the system achieved an accuracy rate of 95.76% in classifying electric vehicles and 97.18% in classifying vehicles for the physically challenged. The average CPU and RAM utilizations of the system were 34.54% and 45.04%, respectively. In addition, the processing time per image was recorded as approximately 1.04 s, demonstrating its potential to run reliably on edge devices. These results are expected to facilitate the efficient resolution of parking management problems in smart cities and effective operation of parking zones reserved for electric vehicles and vehicles for the physically challenged.

关键词： Real-time systems YOLO Accuracy License plate recognition classification algorithms Artificial intelligence Feature extraction Computational modeling Computational efficiency Smart cities bounding box license plate recognition (LPR) MobileNet optical character recognition (OCR) video processing

来源：评论

学校读者我要写书评

暂无评论

A Co-Evolutionary Genetic Algorithm Approach to Optimizing Deep Learning for Brain Tumor classification

引用

IEEE ACCESS 2025年 13卷 21229-21248页

作者： Ali, Abdelmgeid A. Hammad, Mohamed T. Hassan, Hassan S. Minia Univ Fac Comp & Informat Al Minya 61519 Egypt Univ Sadat City Fac Comp & Artificial Intelligence Sadat City 32897 Egypt

Brain tumors are among the deadliest diseases, leading researchers to focus on improving the accuracy of tumor classification-a critical task for prompt diagnosis and effective treatment. Recent advancements in brain tumor diagnosis have significantly increased the use of deep learning techniques, particularly pre-trained models, for classification tasks. These models serve as feature extractors or can be fine-tuned for specific tasks, reducing both training time and data requirements. However, achieving high accuracy in multi-class brain tumor classification remains a major challenge, driving continued research in this area. Key obstacles include the need for expert interpretation of deep learning model outputs and the difficulty of developing highly accurate categorization systems. Optimizing the hyperparameters of Convolutional Neural Network (CNN) architectures, especially those based on pre-trained models, plays a crucial role in improving training efficiency. Manual hyperparameter adjustment is time-consuming and often results in suboptimal outcomes. To address these challenges, we propose an advanced approach that combines transfer learning with enhanced coevolutionary algorithms. Specifically, we utilize EfficientNetB3 and DenseNet121 pre-trained models in conjunction with the Co-Evolutionary Genetic Algorithm (CEGA) to classify brain tumors into four categories: gliomas, meningiomas, pituitary adenomas, and no tumors. CEGA optimizes the hyperparameters, improving both convergence speed and accuracy. Experiments conducted on a Kaggle dataset demonstrate that CEGA-EfficientNetB3 achieved the highest accuracy of 99.39%, while CEGA-DenseNet121 attained 99.01%, both without data augmentation. These results outperform cutting-edge methods, offering a rapid and reliable method for brain tumor classification. This approach has great potential to support radiologists and physicians in making timely and accurate diagnoses.

关键词： Brain modeling Feature extraction Brain tumors Accuracy Genetic algorithms Optimization Magnetic resonance imaging Training Deep learning classification algorithms convolutional neural network (CNN) deep learning (DL) EfficientNetB3 DenseNet121 transfer learning (TL) CEGA hyperparameter optimization MRI images pre-trained models

来源：评论

学校读者我要写书评

暂无评论

Rapid Colorectal Tissue classification Using Data-Driven Raman Techniques

引用

IEEE ACCESS 2025年 13卷 29601-29612页

作者： Tomes, Jakub Janstova, Daniela Mohsen, Shayestegan Sinica, Alla Kovacova, Zuzana Petrtyl, Jaromir Mares, Jan Univ Chem & Technol Dept Math Informat & Cybernet Fac Chem Engn Prague 16628 Czech Republic Univ Pardubice Fac Elect Engn & Informat Pardubice 53002 Czech Republic Univ Chem & Technol Fac Chem Engn Dept Analyt Chem Prague 16628 Czech Republic Charles Univ Prague Fac Med 1 Internal Clin Clin Gastroenterol & Hepatol 4 Prague 11000 Czech Republic Gen Univ Hosp Prague Prague 12800 Czech Republic

Colorectal cancer is among the most widespread cancers globally, and the risk of developing this disease increases with age. This has led to the recommendation that screening should begin in middle-aged patients. Consequently, the implementation of prevention programs has resulted in a greater number of samples being available for histological analysis. Raman spectroscopy offers a diagnostic solution to this challenge. In this study, we present a rapid procedure for Raman tissue analysis and classification using machine learning methods, including both supervised and hybrid supervised-unsupervised approaches, to distinguish between healthy and pathological cases based on Raman spectroscopy data. Raman spectra were acquired using a handheld portable spectrometer with an automatic BubbleFill preprocessing algorithm, covering at spectral range of 500 to 1800 cm(-1). Various machine-learning algorithms have been evaluated for classification, including Long Short-Term Memory (LSTM), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGBoost), and hybrid supervised-unsupervised methods involving dimensionality reduction, clustering, and classification. The experimental results indicated promising classification accuracy, achieving up to 82% accuracy using the MLP algorithm. This suggests the potential effectiveness of neural network methodologies for the classification of Raman spectroscopy data for the diagnosis of colorectal cancer.

关键词： Long short term memory classification algorithms Raman scattering Principal component analysis Autoencoders Computer architecture Microprocessors Boosting Training clinical diagnosis machine learning supervised learning spectroscopy spectroscopy

来源：评论

学校读者我要写书评

暂无评论

Optimization Strategy of classification Model Based on Weighted Implicit Optimal Extreme Learning Machine

引用

IEEE ACCESS 2024年 12卷 74169-74184页

作者： Zhang, Na Wang, Xiaofeng Henan Finance Univ Sch Comp & Artificial Intelligence Zhengzhou 450046 Peoples R China SEGi Univ Grad Sch Business GSB Kota Damansara 47810 Petaling Jaya Malaysia

classification algorithms are one of the important research topics in the artificial intelligence, widely applied in various scientific and engineering fields. Extreme learning machine is a single hidden layer feed-forward neural network algorithm. Compared with traditional neural network models, the training speed and the generalization ability are also better. In terms of methodology, this study first innovatively improves the traditional Grey Wolf Optimization (GWO) algorithm to enhance its convergence and search ability. Specific improvement measures include implementing the reverse learning strategy to reduce the initial dependence of the algorithm on population distribution, and adding exploration perception strategy to enhance its global search ability by calculating heuristic factors, so as to identify the global optimal solution more effectively. The results showed that the improved W-DH-ELM model had excellent performance on multiple standard data sets. In particular, the average accuracy was more than 90%, which was significantly higher than other benchmark classification models. In terms of operation efficiency, the running time of the new model on different data sets was significantly reduced, accounting for less than 25%, and the lowest running time was only 4.89%. These experimental results verify the effectiveness of the introduced intelligent optimization algorithm in improving the performance of traditional ELM model without changing the original model structure. The improved W-DH-ELM model not only maintains the fast training performance of ELM, but also has higher accuracy and stability, which shows its superiority in dealing with complex classification tasks. In summary, the weighted two-hidden layer extreme learning machine optimized by the improved GWO proposed in this study has significant advantages in classification problems, providing a new perspective for future machine learning applications and research.

关键词： Optimization classification algorithms Convergence Predictive models Heuristic algorithms Prediction algorithms Feature extraction Learning systems Reverse logistics Two-hidden layer extreme learning machine Grey wolf optimization algorithm classification model reverse learning perception strategy

来源：评论

学校读者我要写书评

暂无评论

Deep Learning-Based Algorithm for classification of News Text

引用

IEEE ACCESS 2024年 12卷 159086-159098页

作者： Yu Li, Xiao Han, Ling Bo Feng Jiang, Zheng Guangdong Ocean Univ Coll Elect & Informat Engn Zhanjiang 524088 Peoples R China Guangxi Minzu Normal Univ Coll Math & Comp Sci Chongzuo 532200 Peoples R China

As online news grows exponentially, hotspot classification is becoming increasingly important. Although traditional machine learning-based text classification methods, such as plain Bayes, support vector machines (SVMs), and classification trees, provide a certain degree of interpretability, they are often incapable of handling complex semantic relations, which frequently results in poor classification accuracy. To address this issue, in this study we introduce a novel deep learning-based text classification (TC) method based on a convolutional neural network (CNN), long short-term memory (LSTM), and an attention mechanism. This method accurately predicts news popularity by combining the feature extraction ability of the CNN, the sequence modeling ability of LSTM, and the weighted summation ability of the attention mechanism. The experimental results demonstrated that, compared to other deep learning models, the proposed method achieved a higher accuracy, more effectively accounted for the context of the text data, and addressed the problem of poor classification accuracy. Along with model selection, feature engineering was also the key to improving the accuracy. Accordingly, we developed a plain Bayesian TC model based on feature extraction, using word embeddings to convert text into richer vector representations. Then we combined the model with the different plain Bayes distributions, proving that the polynomial plain Bayes was the most suitable model for TC. Consequently, we added the feature word classification expressiveness index to improve the term frequency-inverse document frequency (TF-IDF) feature extraction, which produced a classification accuracy of 96%. This demonstrates that the improved model is superior at understanding and classifying text.

关键词： Text categorization Feature extraction Bayes methods Convolutional neural networks Vectors classification algorithms Accuracy Long short term memory Deep learning Machine learning algorithms Convolutional neural network (CNN) naive Bayes news text classification TF-IDF

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：