检索结果-内蒙古大学图书馆

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

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

LDFNN: A lightweight dual-feedback neural network for space target component detection

引用

ADVANCES IN SPACE RESEARCH 2025年第7期75卷 5702-5717页

作者： Zhou, Wenbo Wang, Zebin Li, Ligang Liu, Bo Wu, Yijia Chinese Acad Sci Natl Space Sci Ctr Key Lab Elect & Informat Technol Space Syst Beijing 100190 Peoples R China Univ Chinese Acad Sci Sch Comp Sci & Technol Beijing 100049 Peoples R China Shanghai Acad Spaceflight Technol Shanghai 201109 Peoples R China Beijing Univ Chem Technol Beijing 100029 Peoples R China Shandong Univ Finance & Econ Sch Comp Sci & Technol Jinan Peoples R China

As the number of spacecraft and the complexity of their missions increase, space target component detection technology is becoming critical for ensuring spacecraft safety in orbit. An effective detection system not only monitors and identifies potential collision threats in real-time, enabling timely obstacle avoidance or path adjustments, but also provides crucial localization and identification capabilities for autonomous spacecraft maintenance and repair. However, most existing methods overlook image quality degradation resulting from prolonged on-orbit operation. Although deep learning models have improved detection accuracy, they are often burdened by excessive network parameters and high computational complexity, which limits their applicability to low-power, lightweight, and resource- constrained on-board processing. In this study, we propose a lightweight dual-feedback neural network (LDFNN) for space target component detection and recognition. The network is designed to address the challenges posed by the space environment through a dual- feedback mechanism. The outer-feedback mechanism enhances recognition stability and accuracy by incorporating modulation transfer function (MTF) analysis and image quality assessment, while the inner-feedback mechanism is optimized for low-power on-orbit processing, improving both real-time performance and generalization. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

关键词： Space target Component detection Dual-feedback neural network deep learning MTF

来源：评论

学校读者我要写书评

暂无评论

Bangladeshi Vehicle Classification and Detection Using deep Convolutional Neural Networks With Transfer learning

引用

IEEE ACCESS 2025年 13卷 26429-26455页

作者： Md Farid, Dewan Kumer Das, Proshanta Islam, Monirul Sina, Ebna United Int Univ Dept Comp Sci & Engn Dhaka 1212 Bangladesh

Vehicle classification and detection has been a field of application for deep learning and image processing which play a very important role in intelligent transport management and AI-assisted driving systems. In this paper, we have presented a vehicle classification and detection system to detect and classify low-speed and high-speed Bangladeshi vehicles. To begin, we have implemented and tested the performance of the 11 pre-trained deep convolutional neural network (CNN) models: YOLOv8 Classify, MobileNetV2, GoogLeNet, AlexNet, ResNet-50, SqueezeNet, VGG19, DenseNet-121, Xception, InceptionV3, and NASNetMobile on the six vehicle classification and detection datasets: BIT-Vehicle, IDD, DhakaAI, Poribohon-BD, Sorokh-Poth, and VTID2. We have found that YOLOv8 Classify, MobileNetV2, and GoogLeNet models outperform other models in categorising vehicle types (e.g., car, truck, bus) in images where the vehicle is already the main subject. Next, we have customised the LabelImg image annotation tool to improve the following features: (a) Changing Label Font & Border, (b) Detecting Incorrect Labels, (c) Abbreviating Label Names, (d) Improving Crosshair & Bounding Box Guide, (e) Adding Progress Information, and (f) Improving File List Panel. We have collected data from real-world highway conditions in Dhaka city and labelled 54,556 objects from 5,460 images based on 16 Bangladeshi on-road vehicle classes. Furthermore, we have built a Bangladeshi native vehicle detection classifier for locating and identifying vehicles within larger scenes, often with multiple objects using YOLOv8 Detect and SSD-Mobilenet V2 models and later deploying this classifier into NVIDIA Jetson Nano Developer Kit (a small and powerful computer). Finally, we have tested the proposed Bangladeshi vehicle detection system with different timing, lighting, and weather conditions in several areas of Dhaka city. The proposed system can detect and classify low-speed and high-speed vehicles with an average 93%

关键词： Convolutional neural networks deep learning Vehicle detection Accuracy Urban areas Roads Transfer learning real-time systems Automobiles YOLO deep learning transfer learning vehicle classification and detection system

来源：评论

学校读者我要写书评

暂无评论

High-speed system-on-chip-based platform for real-time crop disease and pest detection using deep learning techniques

引用

COMPUTERS & ELECTRICAL ENGINEERING 2025年 123卷

作者： Mallick, Tausif Murty, D. Omkar Pal, Ranita Mandal, Swagata Saha, Himadri Nath Chakrabarti, Amlan Univ Calcutta AKCSIT Kolkata India Neotia Univ Dept Comp Sci & Engn AI & ML SST Sarisha India Jadavpur Univ Dept Elect & Telecommun Engn Kolkata India Jalpaiguri Govt Engn Coll Dept Elect & Commun Kolkata India Univ Calcutta Dept Comp Sci SNEC Kolkata India

Crop diseases significantly threaten global agricultural productivity and food security, leading to economic losses and increased pesticide use, which pollutes soil and water and disrupts ecological balance. Mustard and mung bean crops are particularly affected by various diseases and pests such as Alternaria blight, aphids, charcoal rot, bruchids, and mosaic. timely and accurately identifying these diseases and pests are crucial for effective crop management. This research tackles disease classification in mustard and mung bean crops by employing transfer learning, a MobileNetV3based CNN model, and a System-on-Chip (SoC) computing platform. The processing system and processing logic of SoC enhance computing flexibility. Xilinx deep learning Processor Unit (DPU) intellectual property (IP) accelerates disease classification 24 times compared to software counterparts. At the same time, our proposed design enhances the throughput by around 29% and reduces the power consumption by around 19%. MobileNetV3 achieves classification accuracies of 96.14% on mung bean and 93.25% on mustard datasets, surpassing other state-of-the-art methods. A vital aspect of this research is developing a user-friendly mobile application for image capture, communication with SoC, and result display, making disease and pest detection more convenient and accessible. The SoC-based system is versatile and can be extended to classify various crop varieties beyond mung bean and mustard without hardware modifications.

关键词： Plant disease detection Plant pest detection CNN Transfer learning image classification System on chip

来源：评论

学校读者我要写书评

暂无评论

GSBF-YOLO: a lightweight model for tomato ripeness detection in natural environments

引用

JOURNAL OF real-time image processing 2025年第1期22卷 1-14页

作者： Hao, Fengqi Zhang, Zuyao Ma, Dexin Kong, Hoiio Qilu Univ Technol Key Lab Comp Power Network & Informat Secur Minist EducShandong Comp Sci Ctr Natl Supercomp Ctr JinanShandong Acad Sci Jinan 250014 Shandong Peoples R China City Univ Macau Fac Data Sci Macau 999078 Peoples R China Shandong Fundamental Res Ctr Comp Sci Shandong Prov Key Lab Ind Network & Informat Syst Jinan 250014 Peoples R China Qingdao Agr Univ Commun Coll Qingdao 266109 Peoples R China

Accurate tomato ripeness detection is essential for optimizing harvest timing and maximizing yield. deep learning-based object detection has proven effective in this task. However, many existing algorithms have numerous parameters and substantial computational demands, making them unsuitable for agricultural environments with limited computational resources. Additionally, accurate detection becomes challenging with overlapping fruits, leaf occlusion, or complex backgrounds. To address these issues, this paper proposes a lightweight detection model, GSBF-YOLO. This model designs the GSim module to reduce parameters while maintaining detection accuracy. The C3Ghost module further reduces parameter count by replacing the traditional C3 module. The PANet multi-scale feature fusion network in the neck is replaced with the Bi-directional Feature Pyramid Network (BiFPN), which adjusts weights based on the importance of input features. Lastly, the fine-tuned FocalEIOU Loss function is used to calculate the bounding box regression loss, enhancing the model's ability to adjust the weights of high-quality anchor boxes for better detection of targets in occlusion scenarios. Experimental results show that GSBF-YOLO reduces parameters and computational load by 42% and 45%, respectively, while mean Average Precision (mAP) increases by 1.9% and 1.6% on two datasets. The model achieves 110 Frames Per Second (FPS), meeting real-time detection requirements, and has fewer parameters and higher accuracy compared to models like YOLOv8. The research indicates that the proposed lightweight model can effectively detect tomato ripeness in natural environments.

关键词： Lightweight deep learning Tomato ripeness Object detection real-time detection

来源：评论

学校读者我要写书评

暂无评论

TrioConvTomatoNet-BiLSTM: An Efficient Framework for the Classification of Tomato Leaf Diseases in real time Complex Background images

引用

INTERNATIONAL JOURNAL OF COMPUTATIONAL INTELLIGENCE SYSTEMS 2025年第1期18卷 1-28页

作者： Ledbin Vini, S. Rathika, P. PSN Coll Engn & Technol ECE Dept Tirunelveli 627152 India

Tomatoes are the most valuable vegetable worldwide that suffer from leaf diseases, which affect long-term tomato protection. So, to protect the tomato plants from the leaf diseases, it is essential to perform appropriate control measures through early and accurate categorization of leaf diseases. Recently, automated deep learning-based methods, including convolutional neural networks (CNNs), guaranteed accurate and timely classification of tomato leaf diseases. However, CNNs primarily capture local context features within a limited receptive field, making them effective for uniform background images. To handle complex background images, utilizing local and global context features is essential for accurate classification. To do so, it is essential to hybrid CNN architecture with other deep learning modules. This work suggests the TrioConvTomatoNet-BiLSTM framework, a hybridization of CNN architecture named TrioConvTomatoNet with a sequence module named bidirectional long short-term memory (BiLSTM). The proposed framework integrated both local and global context features for the precise classification of images with complex backgrounds. As a result, the proposed framework achieves remarkable accuracy of 99.65%, 98.83%, and 99.20% in classifying tomato leaf disease images with non-uniform, synthetic, and real-time complex backgrounds against the TrioConvTomatoNet and TrioConvTomatoNet-LSTM frameworks. Despite the fact that it requires a lesser number of training parameters and attained maximum accuracy over other existing hybrid approaches, expresses its superiority, robustness, and practical applicability. These features highlight the potential of the proposed framework in the emerging field of smart agriculture by enabling smartphone-based classification of tomato leaf diseases with real-life scenarios.

关键词： image processing Tomato leaf disease classification Smart agriculture deep learning Hybrid model

来源：评论

学校读者我要写书评

暂无评论

A computational method for real-time roof defect segmentation in robotic inspection

引用

COMPUTER-AIDED CIVIL AND INFRASTRUCTURE ENGINEERING 2025年

作者： Zhao, Xiayu Jebelli, Houtan Univ Illinois Dept Civil & Environm Engn Champaign IL USA

Roof inspections are crucial but perilous, necessitating safer and more cost-effective solutions. While robots offer promising solutions to reduce fall risks, robotic vision systems face efficiency limitations due to computational constraints and scarce specialized data. This study presents real-time roof defect segmentation network (RRD-SegNet), a deep learning framework optimized for mobile robotic platforms. The architecture features a mobile-efficient backbone network for lightweight processing, a defect-specific feature extraction module for improved accuracy, and a regressive detection and classification head for precise defect localization. Trained on the multi-type roof defect segmentation dataset of 1350 annotated images across six defect categories, RRD-SegNet integrates with a roof damage identification module for real-time tracking. The system surpasses state-of-the-art models with 85.2% precision and 76.8% recall while requiring minimal computational resources. Field testing confirms its effectiveness with F1-scores of 0.720-0.945 across defect types at processing speeds of 1.62 ms/frame. This work advances automated inspection in civil engineering by enabling efficient, safe, and accurate roof assessments via mobile robotic platforms.

关键词： image segmentation

来源：评论

学校读者我要写书评

暂无评论

Hybrid ultra-short term solar irradiation forecasting using resource-efficient multi-step long-short term memory

引用

RENEWABLE ENERGY 2025年 247卷

作者： Barancsuk, Lilla Groma, Veronika Kocziha, Barnabas Budapest Univ Technol & Econ Dept Elect Power Engn Muegyet Quay 3 H-1111 Budapest Hungary HUN REN Ctr Energy Res Konkoly Thege Miklos St 29-33 H-1121 Budapest Hungary

Accurate forecasting of solar irradiance is a key tool for optimizing the efficiency and service quality of solar energy systems. In this paper, a novel approach is proposed for multi-step solar irradiation forecasting using deep learning models optimized for low computational resource environments. Traditional forecasting models often lack accuracy, and modern, deep-learning based models, while accurate, require substantial computational resources, making them impractical for real-time or resource-constrained environments. Our method uniquely combines dimensionality reduction via image processing with an LSTM-based architecture, achieving significant input data reduction by a factor of 4250 while preserving essential sky condition information, resulting in a lightweight neural network architecture that balances prediction accuracy with computational efficiency. The forecasts are generated simultaneously for multiple time steps: 1 minute, 5 minutes, 10 minutes and 20 minutes. Models are evaluated against a custom dataset, spanning across more than 3 years, containing 1 min samples encompassing both all-sky imagery and meteorological measurements. The approach is demonstrated to achieve better forecasting accuracy, namely a forecast skill of 10 % compared to persistence, and a significantly reduced computational overhead compared to benchmark ConvLSTM models. Moreover, utilizing the preprocessed image features reduces input size by a factor of 6 compared to the raw images. Our findings suggest that the proposed models are well-suited for deployment in embedded systems, remote sensors, and other scenarios where computational resources are limited.

关键词： Solar irradiation forecast Multistep forecasting deep learning LSTM image features Resource-efficient Total sky imager

来源：评论

学校读者我要写书评

暂无评论

Resource-aware strategies for real-time multi-person pose estimation

引用

image AND VISION COMPUTING 2025年 155卷

作者： Esmail, Mohammed A. Wang, Jinlei Wang, Yihao Sun, Li Zhu, Guoliang Zhang, Guohe Xi An Jiao Tong Univ Sch Microelect Xian 710049 Peoples R China Beijing Inst Technol Beijing 100076 Peoples R China Beijing Res Inst Telemetry Beijing 100076 Peoples R China

When using deep learning applications for human posture estimation (HPE), especially on devices with limited resources, accuracy and efficiency must be balanced. Common deep-learning architectures have a propensity to use a large amount of processing power while yielding low accuracy. This work proposes the implementation of Efficient YoloPose, a new architecture based on You Only Look Once version 8 (YOLOv8)-Pose, in an attempt to address these issues. Advanced lightweight methods like Depthwise Convolution, Ghost Convolution, and the C3Ghost module are used by Efficient YoloPose to replace traditional convolution and C2f (a quicker implementation of the Cross Stage Partial Bottleneck). This approach greatly decreases the inference, parameter count, and computing complexity. To improve posture estimation even further, Efficient YoloPose integrates the Squeeze Excitation (SE) attention method into the network. The main focus of this process during posture estimation is the significant areas of an image. Experimental results show that the suggested model performs better than the current models on the COCO and OCHuman datasets. The proposed model lowers the inference time from 1.1 milliseconds (ms) to 0.9 ms, the computational complexity from 9.2 Giga Floating-point operations (GFlops) to 4.8 GFlops and the parameter count from 3.3 million to 1.3 million when compared to YOLOv8-Pose. In addition, this model maintains an average precision (AP) score of 78.8 on the COCO dataset. The source code for Efficient YoloPose has been made publicly available at [https://***/malareeqi/Efficient-YoloPose].

关键词： deep learning Human pose estimation (HPE) Efficient YoloPose Lightweight techniques Computational efficiency

来源：评论

学校读者我要写书评

暂无评论

Application and optimization of intelligent image identification technology in highway inspection data

引用

INTERNATIONAL JOURNAL OF SYSTEM ASSURANCE ENGINEERING AND MANAGEMENT 2025年 1-13页

作者： Wang, Wenjuan Li, Chengwu Sun, Shouwei Shen, Chao Wang, Yunbin Beijing Yunxuehui Informat Technol Co LTD Beijing 100080 Peoples R China Yunnan Transportat Res Inst Co LTD Kunming 650011 Yunnan Peoples R China

Traditional inspection methods for highways include fixed cameras and automatic detection systems. Although they automate the inspection process to a certain extent, they still face problems such as limited coverage, delayed data processing, and insufficient identification accuracy. These systems may face compatibility challenges when integrated with new technologies. To address these issues, this article explores the application of intelligent image identification technology in highway inspections. High-resolution cameras are used for data acquisition, and convolutional neural networks (CNN) are applied to extract image features and generate feature maps. The faster region-based convolutional neural network (Faster R-CNN) is used to classify and perform the bounding box regression on the features extracted by CNN in these regions, determining the category and precise location of the object. At the same time, the deep learning algorithm Autoencoder is used for anomaly detection, and multi-source data is integrated through Kalman filtering. The model parameters and computational graphs are also optimized to improve system efficiency. A 12-km-long highway was selected for testing. The average processing time of the system was within 400 ms, and the processing time for each frame of the image was about 150 ms. It can be seamlessly integrated with the existing system, meeting the real-time requirements. The results show that intelligent image recognition technology can achieve higher highway inspection efficiency and accuracy while maintaining good compatibility with existing equipment.

关键词： Intelligent image identification technology Convolutional neural network System integration real-time data processing Performance optimization

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：