检索结果-内蒙古大学图书馆

SqueezeSlimU-Net: An Adaptive and Efficient Segmentation Architecture for real-time UAV Weed Detection

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 5749-5764页

作者： Machidon, Alina L. Krasovec, Andraz Pejovic, Veljko Machidon, Octavian M. Univ Ljubljana Fac Comp & Informat Sci Ljubljana 1000 Slovenia Inst Jozef Stefan Comp Syst Dept Ljubljana 1000 Slovenia

The limited processing capacity of computing equipment that is usually mounted on unmanned aerial vehicles (UAVs) often prevents real-time execution of computer vision tasks, such as image segmentation. In this article, we introduce SqueezeSlimU-Net (SSU-Net), an adaptive and efficient deep learning (DL) model designed to enhance UAV capabilities in performing complex image segmentation tasks under resource constraints, thereby advancing real-time UAV vision-a crucial technology in fields, such as precision agriculture. SSU-Net combines benefits of three specialized DL architectures: the semantic segmentation capabilities of the U-Net architecture, the computational efficiency of SqueezeNet's fire modules, and the dynamic adaptability of slimmable neural networks. This integration allows SSU-Net to adjust its network width in real-time, thus striking the balance between inference accuracy and computational load based on the operational parameters such as task requirements and UAV's battery life. To validate SSU-Net's efficacy, we applied it to a weed detection task using two UAV-collected datasets and tested it on an edge computing platform for UAVs. Our experiments show that SSU-Net can reduce inference energy consumption by up to 65% with only a minimal 2% reduction in accuracy. A comparative evaluation with other state-of-the-art DL image segmentation approaches shows that SSU-Net achieves on par weed detection performance while requiring significantly fewer model parameters. In addition, SSU-Net outperforms state-of-the-art network pruning techniques in balancing accuracy and resource usage. Timing benchmarks show SSU-Net fostering real-time weed detection even on low-resource UAVs, making it ideal for UAV remote sensing applications.

关键词： Adaptive neural networks computational effi- ciency image segmentation precision agriculture precision agriculture real-time unmanned aerial vehicle (UAV) vision real-time unmanned aerial vehicle (UAV) vision weed detection weed detection weed detection

来源：评论

学校读者我要写书评

暂无评论

Application of Soft Computing for Prediction of Peripheral Neuropathy in T2DM-A real-time Data Analysis

引用

IEEE ACCESS 2025年 13卷 29264-29278页

作者： Sheikh, Mehewish Musheer Balachandra, Mamatha Narendra, V. G. Maiya, Arun G. Manipal Acad Higher Educ Manipal Inst Technol Dept Comp Sci & Engn Manipal 576104 India Manipal Coll Hlth Profess Dept Physiotherapy Manipal 576104 India

Type 2 Diabetes mellitus (T2DM) patients are experiencing diabetic foot problems that heavily burden healthcare systems around the globe. Although there have been challenges with diagnosing and treating these problems using traditional approaches, the advent of machine learning technology signals the beginning of a new age in diabetic foot care, with the promise of improved precision and customized treatment plans. Machine learning is a beneficial tool for extracting essential insights from large, complicated datasets to improve the accuracy of diabetic foot diagnosis and therapeutic planning. This research aims to employ artificial intelligence to build a decision support system that will use clinical and demographic variables to predict the likelihood that individuals with mild, moderate, or severe peripheral neuropathy may develop diabetic foot syndrome in T2DM. real-time processing of clinical information is made possible by a customized stacked ensemble model, which offers immediate peripheral neuropathy risk prediction with low computing latency. The system's capacity to transform raw patients' data into valuable insights in milliseconds supports quick clinical decision-making. Additionally, comparison and testing have been conducted on four deep learning algorithms: ResMLP, LSTM, DNN, and 1D-CNN. The predictions produced by the classifiers have been interpreted using three explainers: LIME, Eli5, and SHAP. Using the mutual information feature selection technique, the final stack reached a maximum accuracy of 99%. The three most significant markers that helped predict the onset of diabetic foot syndrome were area under pressure, vpt right, and vpt left;the encouraging findings point to the possibility of predicting diabetic foot condition with a decision system.

关键词： Foot Diabetes Pressure measurement Skin real-time systems Medical diagnostic imaging Machine learning algorithms Feature extraction Amputation Temperature distribution Artificial intelligence deep learning feature extraction image analysis machine learning prediction methods risk analysis

来源：评论

学校读者我要写书评

暂无评论

Infrared thermal radiation and deep learning algorithms for evaluating the warm-up effect of sports training: Thermal imaging monitoring model☆

引用

THERMAL SCIENCE AND ENGINEERING PROGRESS 2025年 58卷

作者： Liu, Yumeng Li, Yunlong Liang, Danqing Li, Cheng Wu, Chuanzhong Jiangsu Univ Sci & Technol Coll Phys Educ Zhenjiang 212100 Peoples R China Shanghai Int Studies Univ Xianda Coll Econ & Humanities Shanghai 202162 Peoples R China Natl Univ Phys Educ & Sports Ukraine UA-02000 Kiev Ukraine

Warm-up activity is an important link to prevent sports injury and improve sports performance. The traditional evaluation methods of warm-up effect mainly rely on the experience judgment of coaches and the subjective feelings of athletes, and lack of objective and quantitative evaluation means. The purpose of this study is to develop an evaluation system based on infrared thermal imaging monitoring model, which can monitor and evaluate the warm-up effect of athletes in real time. The basic principle and acquisition process of infrared thermal imaging technology are introduced. The real-time temperature distribution images of athletes during the warm-up process are obtained by using high-precision infrared thermal imager. The data pre-processing steps include image normalization, enhancement and feature extraction, etc., to ensure that the data quality meets the requirements of model training. In the model training stage, the way of supervised learning is used to train and verify the deep learning model by using the labeled thermal imaging data set. Finally, the performance of the model is evaluated by cross-validation and other methods to ensure that it has good generalization ability. The results show the performance of thermal imaging monitoring model in practical application. The experimental results show that the model can accurately identify the athletes who are under-warmed or over-warmed, and give timely adjustment suggestions. The model can also predict the possible muscle fatigue and potential injury risk during the warm-up process, which provides a scientific basis for coaches to make personalized training plans.

关键词： Infrared thermal radiation deep learning algorithm Physical training Evaluation of warm-up effect Thermal imaging monitoring model

来源：评论

学校读者我要写书评

暂无评论

Intelligent representation method of image flatness for cold rolled strip

引用

Journal of Iron and Steel Research International 2024年第5期31卷 1177-1195页

作者： Yang-huan Xu Dong-cheng Wang Hong-min Liu Bo-wei Duan National Engineering Research Center for Equipment and Technology of Cold Strip Rolling Yanshan UniversityQinhuangdao066004HebeiChina National Key Laboratory of Metal Forming Technology and Heavy Equipment China National Heavy Machinery Research Institute Co.Ltd.Xi’an710018ShaanxiChina State Key Laboratory of Metastable Materials Science and Technology Yanshan UniversityQinhuangdao066004HebeiChina

real flatness images are the bases for flatness detection based on machine vision of cold rolled *** characteristics of a real flatness image are analyzed,and a lightweight strip location detection(SLD)model with deep semantic segmentation networks is *** interference areas in the real flatness image can be eliminated by the SLD model,and valid information can be *** this basis,the concept of image flatness is proposed for the first *** image flatness representation(IFAR)model is established on the basis of an autoencoder with a new *** optimal structure of the bottleneck layer is 16×16×4,and the IFAR model exhibits a good representation ***,interpretability analysis of the representation factors is carried out,and the difference and physical meaning of the representation factors for image flatness with different categories are *** flatness with new defect morphologies(bilateral quarter waves and large middle waves)that are not present in the original dataset are generated by modifying the representation factors of the no wave ***,the SLD and IFAR models are used to detect and represent all the real flatness images on the test *** average processing time for a single image is 11.42 ms,which is suitable for industrial *** research results provide effective methods and ideas for intelligent flatness detection technology based on machine vision.

关键词： Cold rolled strip image flatness Location detection Representation learning Bottleneck layer

来源：评论

学校读者我要写书评

暂无评论

Radar Signal processing and Its Impact on deep learning-Driven Human Activity Recognition

引用

SENSORS 2025年第3期25卷 724-724页

作者： Ayaz, Fahad Alhumaily, Basim Hussain, Sajjad Imran, Muhammad Ali Arshad, Kamran Assaleh, Khaled Zoha, Ahmed Univ Glasgow James Watt Sch Engn Glasgow City G12 8QQ England Ajman Univ Coll Engn & Informat Technol Dept Elect & Comp Engn POB 346 Ajman U Arab Emirates Ajman Univ Artificial Intelligence Res Ctr POB 346 Ajman 346 U Arab Emirates

Human activity recognition (HAR) using radar technology is becoming increasingly valuable for applications in areas such as smart security systems, healthcare monitoring, and interactive computing. This study investigates the integration of convolutional neural networks (CNNs) with conventional radar signal processing methods to improve the accuracy and efficiency of HAR. Three distinct, two-dimensional radar processing techniques, specifically range-fast Fourier transform (FFT)-based time-range maps, time-Doppler-based short-time Fourier transform (STFT) maps, and smoothed pseudo-Wigner-Ville distribution (SPWVD) maps, are evaluated in combination with four state-of-the-art CNN architectures: VGG-16, VGG-19, ResNet-50, and MobileNetV2. This study positions radar-generated maps as a form of visual data, bridging radar signal processing and image representation domains while ensuring privacy in sensitive applications. In total, twelve CNN and preprocessing configurations are analyzed, focusing on the trade-offs between preprocessing complexity and recognition accuracy, all of which are essential for real-time applications. Among these results, MobileNetV2, combined with STFT preprocessing, showed an ideal balance, achieving high computational efficiency and an accuracy rate of 96.30%, with a spectrogram generation time of 220 ms and an inference time of 2.57 ms per sample. The comprehensive evaluation underscores the importance of interpretable visual features for resource-constrained environments, expanding the applicability of radar-based HAR systems to domains such as augmented reality, autonomous systems, and edge computing.

关键词： human activity classification radar domain representations deep learning computational cost transfer learning

来源：评论

学校读者我要写书评

暂无评论

Terahertz image denoising via multiscale hybrid-convolution residual network

引用

CAAI Transactions on Intelligence Technology 2025年第1期10卷 235-252页

作者： Heng Wu Zijie Guo Chunhua He Shaojuan Luo Bofang Song Guangdong Provincial Key Laboratory of Cyber-Physical System School of AutomationGuangdong University of TechnologyGuangzhouChina School of Computer Guangdong University of TechnologyGuangzhouChina School of Chemical Engineering and Light Industry Guangdong University of TechnologyGuangzhouChina Wyant College of Optical Sciences The University of ArizonaTucsonArizonaUSA

Terahertz imaging technology has great potential applications in areas,such as remote sensing,navigation,security checks,and so ***,terahertz images usually have the problems of heavy noises and low *** terahertz image denoising methods are mainly based on traditional image processing methods,which have limited denoising effects on the terahertz *** deep learning-based image denoising methods are mostly used in natural images and easily cause a large amount of detail loss when denoising terahertz ***,a residual-learning-based multiscale hybridconvolution residual network(MHRNet)is proposed for terahertz image denoising,which can remove noises while preserving detail features in terahertz ***,a multiscale hybrid-convolution residual block(MHRB)is designed to extract rich detail features and local prediction residual noise from terahertz ***,MHRB is a residual structure composed of a multiscale dilated convolution block,a bottleneck layer,and a multiscale convolution *** uses the MHRB and global residual learning to achieve terahertz image *** studies are performed to validate the effectiveness of MHRB.A series of experiments are conducted on the public terahertz image *** experimental results demonstrate that MHRNet has an excellent denoising effect on synthetic and real noisy terahertz *** with existing methods,MHRNet achieves comprehensive competitive results.

关键词： image processing multiscale hybrid-convolution residual learning terahertz image denoising

来源：评论

学校读者我要写书评

暂无评论

SLTM Network: Efficient Application of Lightweight image Segmentation Technology in Detecting Drivable Areas for Unmanned Line-Marking Machines

引用

IEEE ACCESS 2024年 12卷 169001-169012页

作者： Wang, Chao Chen, Xiangkai Wang, Bingtao Zhang, Liang Liu, Bing Shandong Univ Sch Mech Elect & Informat Engn Weihai 264209 Peoples R China Shandong Univ Off Acad Affairs Weihai 264209 Peoples R China

image segmentation plays a crucial role in the roadwork operations of autonomous line-painting machines. However, the limited resources of mobile platforms in intelligent line-painting applications pose a dual challenge of ensuring both accuracy and real-time performance in road segmentation. To address this issue, this study introduces a lightweight yet efficient image segmentation model, termed the SLTM Network. Central to this network is the lightweight SLTM module, which significantly reduces the model's parameter count and lowers the computational overhead of the decoder. To enhance the interplay of information at different spatial resolutions, the network incorporates an SE attention-enhanced upsampling module (SAUM) and employs a Spatial Attention Sequence (SAS) unit to improve global environment perception at a low computational cost. Comprehensive experimental evaluations on the Cityscapes dataset demonstrate that the SLTM Network excels in balancing speed and accuracy, achieving an mIoU of 70.5% with only 4.07M parameters and an impressive inference speed of 267.1 FPS. On the embedded device Jetson Xavier NX, it achieves an inference speed of 34.2 FPS. Compared to existing lightweight image segmentation models, the SLTM Network exhibits significant advantages in both processing speed and accuracy, making it particularly suitable for real-time autonomous line-painting machine applications.

关键词： Roads image segmentation Feature extraction Computer architecture Computational efficiency Synthetic aperture sonar Semantics Lightweight image segmentation unmanned line-marking machines deep learning in autonomous driving embedded systems optimization SLTM network real-time road segmentation

来源：评论

学校读者我要写书评

暂无评论

HFD-NET: a real-time deep learning-based visual algorithm with weak feature enhancement for hair follicle detection

引用

SIGNAL image AND VIDEO processing 2025年第5期19卷 1-15页

作者： Zhang, Liyang Tong, Juntao Chen, Yue Wang, Nian Li, Weiang Chen, Mengxin Tian, Jinghong Key Lab Urban Rail Transit Intelligent Operat & Ma Jinhua 321000 Peoples R China Zhejiang Normal Univ Coll Comp Sci & Technol Jinhua 321000 Peoples R China Zhejiang Normal Univ Coll Engn Jinhua 321000 Peoples R China

Hair follicle detection in complex scalp environments remains challenging due to small target sizes, morphological similarities, and background interference. To address these issues, we propose HFD-NET, a novel real-time detection framework that significantly enhances weak feature representation while maintaining computational efficiency. Unlike existing methods, HFD-NET introduces three key innovations: (1) the CSRFConv module, which dynamically fuses spatial-channel features to suppress background noise and amplify discriminative follicle characteristics;(2) the C3k2_IBC module, optimized for multi-scale small-object detection through inverted bottleneck convolutions;and (3) the ADown module, which minimizes information loss during downsampling by preserving critical edge and texture details. Extensive experiments on the FDU_HairFollicleDataset demonstrate HFD-NET's superiority, achieving a 67.1% mAP@0.5-outperforming YOLO11n by 5.5%-with only a 0.1M parameter increase and negligible computational overhead. HFD-NET also generalizes effectively to public datasets (e.g., 71.6% mAP@0.5 on HFDC), surpassing Faster R-CNN, SSD, and recent YOLO variants in accuracy-efficiency trade-offs. This work bridges the gap between high-precision detection and real-time applicability, offering a practical solution for clinical hair transplantation and scalp health monitoring.

关键词： Hair follicle detection YOLO11 deep learning Feature extraction

来源：评论

学校读者我要写书评

暂无评论

MMG-Based Motion Segmentation and Recognition of Upper Limb Rehabilitation Using the YOLOv5s-SE

引用

SENSORS 2025年第7期25卷 2257-2257页

作者： Cao, Gangsheng Jia, Shen Wu, Qing Xia, Chunming East China Univ Sci & Technol Sch Mech & Power Engn Shanghai 200237 Peoples R China Shanghai Univ Engn Sci Sch Mech & Automot Engn Shanghai 201620 Peoples R China

Mechanomyography (MMG) is a non-invasive technique for assessing muscle activity by measuring mechanical signals, offering high sensitivity and real-time monitoring capabilities, and it has many applications in rehabilitation training. Traditional MMG-based motion recognition relies on feature extraction and classifier training, which require segmenting continuous actions, leading to challenges in real-time performance and segmentation accuracy. Therefore, this paper proposes an innovative method for the real-time segmentation and classification of upper limb rehabilitation actions based on the You Only Look Once (YOLO) algorithm, integrating the Squeeze-and-Excitation (SE) attention mechanism to enhance the model's performance. In this paper, the collected MMG signals were transformed into one-dimensional time-series images. After image processing, the training set and test set were divided for the training and testing of the YOLOv5s-SE model. The results demonstrated that the proposed model effectively segmented isolated and continuous MMG motions while simultaneously performing real-time motion category prediction and outputting results. In segmentation tasks, the base YOLOv5s model achieved 97.9% precision and 98.0% recall, while the improved YOLOv5s-SE model increased precision to 98.7% (+0.8%) and recall to 98.3% (+0.3%). Additionally, the model demonstrated exceptional accuracy in predicting motion categories, achieving an accuracy of 98.9%. This method realizes the automatic segmentation of time-domain motions, avoids the limitations of manual parameter adjustment in traditional methods, and simultaneously enhances the real-time performance of MMG motion recognition through image processing, providing an effective solution for motion analysis in wearable devices.

关键词： mechanomyography upper limb motion segmentation pattern recognition target detection YOLOv5s-SE deep learning

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Low-Light image Enhancement Based on Explicit Denoising and Knowledge Distillation

引用

Computers, Materials & Continua 2025年第2期82卷 2537-2554页

作者： Wenkai Zhang Hao Zhang Xianming Liu Xiaoyu Guo Xinzhe Wang Shuiwang Li College of Computer Science and Engineering Guilin University of TechnologyGuilin541006China Guangxi Key Laboratory of Embedded Technology and Intelligent System Guilin University of TechnologyGuilin541004China

Under low-illumination conditions, the quality of image signals deteriorates significantly, typically characterized by a peak signal-to-noise ratio (PSNR) below 10 dB, which severely limits the usability of the images. Supervised methods, which utilize paired high-low light images as training sets, can enhance the PSNR to around 20 dB, significantly improving image quality. However, such data is challenging to obtain. In recent years, unsupervised low-light image enhancement (LIE) methods based on the Retinex framework have been proposed, but they generally lag behind supervised methods by 5–10 dB in performance. In this paper, we introduce the Denoising-Distilled Retine (DDR) method, an unsupervised approach that integrates denoising priors into a Retinex-based training framework. By explicitly incorporating denoising, the DDR method effectively addresses the challenges of noise and artifacts in low-light images, thereby enhancing the performance of the Retinex framework. The model achieved a PSNR of 19.82 dB on the LOL dataset, which is comparable to the performance of supervised methods. Furthermore, by applying knowledge distillation, the DDR method optimizes the model for real-time processing of low-light images, achieving a processing speed of 199.7 fps without incurring additional computational costs. While the DDR method has demonstrated superior performance in terms of image quality and processing speed, there is still room for improvement in terms of robustness across different color spaces and under highly resource-constrained conditions. Future research will focus on enhancing the model’s generalizability and adaptability to address these challenges. Our rigorous testing on public datasets further substantiates the DDR method’s state-of-the-art performance in both image quality and processing speed.

关键词： deep learning low-light image enhancement real-time processing knowledge distillation

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：