检索结果-内蒙古大学图书馆

EXPERT SYSTEMS WITH APPLICATIONS 2025年 276卷

作者： Patnaik, Nitesh Raj, Rishi Misra, Indranil Kumar, Vinod Kalinga Inst Ind Technol KIIT Univ Sch Comp Engn Bhubaneswar 751024 Odisha India Indian Inst Management Visakhapatnam Informat Syst Visakhapatnam India Indian Space Res Org Space Applicat Ctr Ahmadabad India Indian Inst Technol BHU Dept Comp Sci & Engn Varanasi 221005 Uttar Pradesh India

Synthetic Aperture Radar (SAR) imagery is indispensable for earth observation, offering the ability to capture data under challenging conditions such as cloud cover and darkness. However, its grayscale format and speckle noise hinder interpretability and pose significant challenges for traditional processing methods. This study introduces an innovative framework for SAR image colorization, leveraging an Attention-Based WGAN-GP (Wasserstein GAN with Gradient Penalty). The model incorporates multi-head self-attention mechanisms to enhance feature extraction, capture long-range dependencies, and dynamically suppress noise through a novel variance-based attention adjustment mechanism. Extensive evaluations on Sentinel-1 and Sentinel-2 datasets across diverse terrains, including agriculture, urban areas, barren land, and grasslands, demonstrate the model's superiority over existing approaches. It achieves an LPIPS score of 0.27, SSIM of 0.76, and an average inference time of 0.22 s, showcasing its ability to preserve spatial coherence and perceptual quality even in complex, noisy environments. This capability enables real-time applications in disaster management, flood monitoring, and urban planning, providing actionable insights and advancing the state-of-the-art in SAR image processing.

关键词： Attention deep learning Generative adversarial networks image colorization image restoration Multihead attention Noise SAR Sentinel

来源：评论

学校读者我要写书评

暂无评论

Speed Matters, a robust infrared and visible image matching method at real-time speed

引用

JOURNAL OF real-time image processing 2024年第1期21卷 15-15页

作者： Chang, Rong Yang, Chuanxu Zhang, Hang Xie, Housheng Zhou, Chengjiang Pan, Anning Yang, Yang Yunnan Power Grid Corp Yuxi Power Supply Bur Yuxi 653100 Yunnan Peoples R China Yunnan Power Grid Corp Informat Ctr Kunming 650500 Yunnan Peoples R China Yunnan Normal Univ Lab Pattern Recognit & Artificial Intelligence Kunming 650500 Yunnan Peoples R China Baoshan Univ Sch Big Data Baoshan 678000 Peoples R China Yunnan Normal Univ Sch Phys & Elect Informat Kunming 650500 Peoples R China

image matching is a crucial step in executing many complex visual tasks, which involve identifying the same or similar visual patterns across various images. Matching images between infrared and visible becomes quite challenging due to the significant non-linear intensity differences. In this paper, we propose using a lightweight network for feature matching of infrared and visible images, combining global and local feature information, and reducing computational costs, enabling real-time inference on most desktop-level GPUs. To fully leverage the powerful matching capabilities of existing state-of-the-art models, we introduce knowledge distillation to obtain more robust features. Moreover, to address the issue of insufficient datasets for network training in existing methods, we propose using image style transfer techniques to synthesize paired datasets of infrared and visible. Experimental results show that our method achieves results comparable to the most advanced methods in infrared and visible image matching. Furthermore, our method has a significant advantage in inference speed, which is beneficial for tasks that require real-time completion.

关键词： Multimodal real time Feature matching deep learning

来源：评论

学校读者我要写书评

暂无评论

Agry: a comprehensive framework for plant diseases classification via pretrained EfficientNet and convolutional neural networks for precision agriculture

引用

MULtimeDIA TOOLS AND APPLICATIONS 2024年第24期83卷 64813-64851页

作者： Saleki, Sheida Tahmoresnezhad, Jafar Urmia Univ Technol Fac Informat Technol & Comp Engn Orumiyeh Iran

Diagnosing plant diseases is a vital issue in maintaining and developing of agricultural products. These diseases occur with changes in the tissue of different parts of plants. While the previous researches have only been conducted on certain species of plants and specific parts, we propose a comprehensive approach that has reached a high accuracy in diagnosing and classifying the disease of offending plant species by examining their different parts, including leaves, fruits, tree trunks, and seeds. We extract features from different layers of pre-trained AlexNet, ResNet50, VGG16, EfficientNetB0, EfficientNetB3 and EfficientNetB7 deep models with a spatial attention module to classify samples with an SVM classifier with RBF kernel. In order to automate the detection of plant diseases by manned or unmanned agricultural machines, our proposed Agry requires only 0.04089 seconds for image processing and decision making in real time. Also, due to the use of transfer learning, the cost of building the proposed model, including time and resources, is minimized. The results of the tests show a significant improvement compared to the previous works, and in most cases the classification is done without errors.

关键词： Precision agriculture Plant disease Machine vision deep learning Transfer learning deep convolutional neural networks Pre-trained models

来源：评论

学校读者我要写书评

暂无评论

A survey on few-shot class-incremental learning

引用

NEURAL NETWORKS 2024年 169卷 307-324页

作者： Tian, Songsong Li, Lusi Li, Weijun Ran, Hang Ning, Xin Tiwari, Prayag Chinese Acad Sci Inst Semicond Beijing 100083 Peoples R China Univ Chinese Acad Sci Sch Elect Elect & Commun Engn Beijing 100049 Peoples R China Univ Chinese Acad Sci Sch Integrated Circuits Beijing 100083 Peoples R China Beijing Key Lab Semicond Neural Network Intellige Beijing 100083 Peoples R China Old Dominion Univ Dept Comp Sci Norfolk VA 23529 USA Halmstad Univ Sch Informat Technol S-30118 Halmstad Sweden

Large deep learning models are impressive, but they struggle when real-time data is not available. Few-shot class-incremental learning (FSCIL) poses a significant challenge for deep neural networks to learn new tasks from just a few labeled samples without forgetting the previously learned ones. This setup can easily leads to catastrophic forgetting and overfitting problems, severely affecting model performance. Studying FSCIL helps overcome deep learning model limitations on data volume and acquisition time, while improving practicality and adaptability of machine learning models. This paper provides a comprehensive survey on FSCIL. Unlike previous surveys, we aim to synthesize few-shot learning and incremental learning, focusing on introducing FSCIL from two perspectives, while reviewing over 30 theoretical research studies and more than 20 applied research studies. From the theoretical perspective, we provide a novel categorization approach that divides the field into five subcategories, including traditional machine learning methods, meta learning-based methods, feature and feature space-based methods, replay-based methods, and dynamic network structure-based methods. We also evaluate the performance of recent theoretical research on benchmark datasets of FSCIL. From the application perspective, FSCIL has achieved impressive achievements in various fields of computer vision such as image classification, object detection, and image segmentation, as well as in natural language processing and graph. We summarize the important applications. Finally, we point out potential future research directions, including applications, problem setups, and theory development. Overall, this paper offers a comprehensive analysis of the latest advances in FSCIL from a methodological, performance, and application perspective.

关键词： Few-shot learning Class-incremental learning Catastrophic forgetting Overfitting Performance evaluation

来源：评论

学校读者我要写书评

暂无评论

Evaluation of Transfer learning for Visual Multiple Target Tracking 1

Evaluation of Transfer Learning for Visual Multiple Target T...

引用

1st International Conference on Advances in Computer Science, Electrical, Electronics, and Communication Technologies, CE2CT 2025

作者： Jadhav, Anisha Anil California State University Department of Computer Science Monterey Bay Seaside United States

ISBN: (纸本)9798331518578

deep learning has revolutionized high-level image processing tasks, notably image classification and segmentation, by effectively handling multi-dimensional features in image space. This report investigates the application of transfer learning, utilizing features extracted from a pre-trained VGG deep neural network to visual multiple-target tracking (MTT). Traditional feature detectors like the Lucas-Kanade Tracker and Good Features to Track (GFTT) algorithm are computationally efficient but limited in dynamic, cluttered environments, such as those encountered by unmanned aerial systems (UAS). Replacing GFTT with deep feature detectors from VGG is proposed to assess tracking accuracy and improve computational efficiency. Our experiments reveal that, though promising, deep feature extraction results in lower frame rates and less smooth trajectories than traditional methods. This study highlights the potential and current limitations of integrating deep learning features into real-time visual tracking systems. Plans include optimizing the deep learning model for real-time processing, automating filter selection for feature extraction, and exploring additional pre-trained networks to improve tracking performance and efficiency. © 2025 IEEE.

关键词： image segmentation

来源：评论

学校读者我要写书评

暂无评论

A deep learning based Scalable and Adaptive Feature Extraction Framework for Medical images

引用

INFORMATION SYSTEMS FRONTIERS 2024年第4期26卷 1279-1305页

作者： Loukil, Zainab Mirza, Qublai Khan Ali Sayers, Will Awan, Irfan Univ Gloucestershire Cyber & Appl Comp Pk Cheltenham GL50 2RH Gloucestershire England Univ Gloucestershire Games Technol Pk Cheltenham GL50 2RH Gloucestershire England Univ Bradford Fac Engn & Informat Comp Sci Richmond Rd Bradford BD7 1DP West Yorkshire England

Features extraction has a fundamental value in enhancing the scalability and adaptability n of medical image processing framework. The outcome of this stage has a tremendous effect on the reliability of the medical application being developed, particularly disease classification and prediction. The challenging side of features extraction frameworks, in relation to medical images, is influenced by the anatomical and morphological structure of the image which requires a powerful extraction system that highlights high- and low- level features. The complementary of both feature types reinforces the medical image content-based retrieval and allows to access visible structures as well as an in-depth understanding of related deep hidden components. Several existing techniques have been used towards extracting high- and low-level features separately, including deep learning based approaches. However, the fusion of these features remains a challenging task. Towards tackling the drawback caused by the lack of features combination and enhancing the reliability of features extraction methods, this paper proposes a new hybrid features extraction framework that focuses on the fusion and optimal selection of high- and low-level features. The scalability and reliability of the proposed method is achieved by the automated adjustment of the final optimal features based on real-time scenarios resulting an accurate and efficient medical images disease classification. The proposed framework has been tested on two different datasets to include BraTS and Retinal sets achieving an accuracy rate of 97% and 98.9%, respectively.

关键词： Features extraction deep learning High-level features Low-level features Optimal features fusion

来源：评论

学校读者我要写书评

暂无评论

Electric load forecasting under false data injection attacks via denoising deep learning and generative adversarial networks

引用

IET GENERATION TRANSMISSION & DISTRIBUTION 2024年第20期18卷 3247-3263页

作者： Mahmoudnezhad, Fayezeh Moradzadeh, Arash Mohammadi-Ivatloo, Behnam Zare, Kazem Ghorbani, Reza Univ Politecn Madrid Escuela Tecn Super Ingn Ind Madrid Spain Univ Tabriz Fac Elect & Comp Engn Tabriz *** Iran LUT Univ Sch Energy Syst Lappeenranta Finland Univ Hawaii Manoa Dept Mech Engn Renewable Energy Design Lab REDLab Honolulu HI USA

Accurate electric load forecasting at various time periods is considered a necessary challenge for electricity consumers and generators to maximize their economic efficiency in energy markets. Hence, the accuracy and effectiveness of existing electric load forecasting approaches depends on the data quality. Nowadays, with the implementation of modern power systems and Internet of Things technology, forecasting models are faced with a large volume of data, which puts the security and health of data at risk due to the use of numerous measuring devices and the threat of cyber-attackers. In this study, a cyber-resilient hybrid deep learning-based model is developed that accurately forecasts electric load in short-term and long-term time horizons. The architecture of the proposed model systematically integrates stacked multilayer denoising autoencoder (SMDAE) and generative adversarial network (GAN) and is called SMDAE-GAN. In the proposed framework, SMDAE layer is used to pre-process and remove real fs and intentional anomalies in data, and GAN layer is also utilized to forecast electric load values. The effectiveness of the SMDAE-GAN structure is studied using realistic electrical load data monitored in the distribution network of Tabriz, Iran, and meteorological data measured in weather station there. Compared with other conventional load forecasting approaches, the developed framework has the highest accuracy in both cases of using normal data with real-world noise and damaged data under false data injection attacks. As far as we know, this paper is considered to be the first study that, by presenting a deep learning-based hybrid forecasting model, was able to simultaneously consider all the prominent indicators related to electric load forecasting, such as the forecast time horizon, data pre-processing, time-series modelling, anomaly detection, data de-noising, and cyber-security. image

关键词： artificial intelligence data integrity demand forecasting load forecasting smart power grids

来源：评论

学校读者我要写书评

暂无评论

Speech enhancement using deep complex convolutional neural network (DCCNN) model

引用

SIGNAL image AND VIDEO processing 2024年第12期18卷 8675-8692页

作者： Iqbal, Yasir Zhang, Tao Fahad, Muhammad Rahman, Sadiq ur Iqbal, Anjum Geng, Yanzhang Zhao, Xin Tianjin Univ Sch Elect & Informat Engn 92 Weijin Rd Tianjin 300072 Peoples R China Dalian Univ Technol Sch Software Technol Dalian 116024 Peoples R China

In cases with highly non-stationary noise, single-channel speech enhancement is quite challenging, mainly when the noise includes interfering speech. In this situation, deep learning's success has contributed to speech enhancement to boost intelligibility and perceptual quality. Existing speech enhancement (SE) works in time-frequency domains only aim to improve the magnitude spectrum via neural network learnings;the latest research highlights the significance of phase in perceptual speech quality. Motivated by multi-task machines and deep learning this paper, proposes an effective and novel approach to the task of speech enhancement using an encoder-decoder architecture based on deep Complex Convolutional Neural Networks. The proposed model takes input from the spectrograms of the noisy speech signals, consisting of real and imaginary components for complex spectral mapping, and it simultaneously enhances the magnitude and phase responses of speech. Considering unseen non-stationary noise categories, which interfere with speech, the proposed model enhances speech quality by approximately, 0.44 MOS points compared to state-of-the-art single-stage techniques. Moreover, it outperforms all reference techniques constantly and improves intelligibility under low-SNR settings. In contrast, against the baselines, we find an incredible enhancement of over 3 dB in SNR, and 0.2 in STOI. In addition, our method outperforms baseline SE techniques in low-SNR conditions in terms of STOI.

关键词： Background noise Speech signals Speech enhancement Noise suppression deep learning deep complex convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Adaptive routing sign transformer framework

引用

SIGNAL image AND VIDEO processing 2023年第4期17卷 1715-1722页

作者： Chen, Yuming Mei, Xue Nanjing Tech Univ Coll Elect Engn & Control Sci Nanjing Peoples R China

Although deep learning-based continuous sign language translation (CSLT) models have made great progress in recent years, they are still faced with various difficulties and limitations when applied to practical scenarios. In order to better apply the technology of deep learning, we propose the adaptive route sign transformer framework for CSLT. The adaptive routing strategy is proposed to solve the problem that the accuracy of the deep learning model trained in the laboratory scene is greatly reduced when it is applied to the real scene, and the back-end part of the model, we present, adopts transformer-style decoder architecture to real-time translate sentences from the spatiotemporal context around the signer. By means of network layer visualization, we demonstrate that the attention mechanism of the model captures the hand and face regions of signers, which is often crucial for semantic analysis of video sign language. In this paper, we introduce the Chinese sign language corpus of the business scene which show sign language communication in a bank, a station, etc. It has certain impetuses for further research on video sign language translation. Experiments are carried out the PHOENIX-Weather 2014T (RWTH Aachen University, Germany);the proposed model outperforms the state-of-the-art in inference times and accuracy using only raw RGB as input.

关键词： Sign language recognition Domain adaption Transfer learning Transformer

来源：评论

学校读者我要写书评

暂无评论

Improvement of a Subpixel Convolutional Neural Network for a Super-Resolution image

引用

APPLIED SCIENCES-BASEL 2025年第5期15卷 2459-2459页

作者： Agalday, Muhammed Fatih Cinar, Ahmet Mardin Artuklu Univ Vocat Sch Dept Comp Programing TR-47100 Mardin Turkiye Firat Univ Fac Engn Dept Comp Engn TR-23000 Elazig Turkiye

Super-resolution technologies are one of the tools used in image restoration, which aims to obtain high-resolution content from low-resolution images. Super-resolution technology aims to increase the quality of a low-resolution image by reconstructing it. It is a useful technology, especially in content where low-resolution images need to be enhanced. Super-resolution applications are used in areas such as face recognition, medical imaging, and satellite imaging. deep neural network models used for single-image super-resolution are quite successful in terms of computational performance. In these models, low-resolution images are converted to high resolution using methods such as bicubic interpolation. Since the super-resolution process is performed in the high-resolution area, it adds a memory cost and computational complexity. In our proposed model, a low-resolution image is given as input to a convolutional neural network to reduce computational complexity. In this model, a subpixel convolution layer is presented that learns a series of filters to enhance low-resolution feature maps to high-resolution images. In our proposed model, convolution layers are added to the efficient subpixel convolutional neural network (ESPCN) model, and in order to prevent the lost gradient value, we transfer the feature information of the current layer from the previous layer to the next upper layer. The efficient subpixel convolutional neural network (R-ESPCN) model proposed in this paper is remodeled to reduce the time required for the real-time subpixel convolutional neural network to perform super-resolution operations on images. The results show that our method is significantly improved in accuracy and demonstrates the applicability of deep learning methods in the field of image data processing.

关键词： subpixel convolutional neural network super-resolution residual network image reconstruction

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：