检索结果-内蒙古大学图书馆

deep learning-Based Defects Detection in Keyhole TIG Welding with Enhanced Vision

MATERIALS 2024年第15期17卷 3871页

作者： Zhang, Xuan Zhao, Shengbin Wang, Mingdi Soochow Univ Sch Mech & Elect Engn Suzhou 215137 Peoples R China

Keyhole tungsten inert gas (keyhole TIG) welding is renowned for its advanced efficiency, necessitating a real-time defect detection method that integrates deep learning and enhanced vision techniques. This study employs a multi-layer deep neural network trained on an extensive welding image dataset. Neural networks can capture complex nonlinear relationships through multi-layer transformations without manual feature selection. Conversely, the nonlinear modeling ability of support vector machines (SVM) is limited by manually selected kernel functions and parameters, resulting in poor performance for recognizing burn-through and good welds images. SVMs handle only lower-level features such as porosity and excel only in detecting simple edges and shapes. However, neural networks excel in processing deep feature maps of "molten pools" and can encode deep defects that are often confused in keyhole TIG. Applying a four-class classification task to weld pool images, the neural network adeptly distinguishes various weld states, including good welds, burn-through, partial penetration, and undercut. Experimental results demonstrate high accuracy and real-time performance. A comprehensive dataset, prepared through meticulous preprocessing and augmentation, ensures reliable results. This method provides an effective solution for quality control and defect prevention in keyhole TIG welding process.

关键词： keyhole TIG welding deep learning defects detection

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

EFFECTIVE FEATURE I NFORMATION IDENTIFICATION OF SUGARCANE BASED ON HYBRID deep learning MODELS

引用

APPLIED ENGINEERING IN AGRICULTURE 2024年第3期40卷 243-257页

作者： Pan, Mingzhang Gou, Xuanyuan Zeng, Yue Wang, Zongrun Yuan, Leyi Liang, Ke Guangxi Univ State Key Lab Conservat & Utilizat Biol Resources Nanning Guangxi Peoples R China Guangxi Univ Nanning Guangxi Peoples R China

The efficiency of intelligent sugarcane harvesters in harvesting depends on the effectiveness of identifying and locating the sugarcane during the harvesting process. In the actual harvesting process, accurately extracting valid features of sugarcane amidst the dense and interwoven sugarcane becomes a challenging task. To address this issue, we propose a hybrid deep learning approach to extract sugarcane stem contours and internal stem node feature information from sugarcane efficiently in the context of a complex harvest. Firstly, this study combined the MobileNetV3 and U-Net networks to segment overall images that contain information about the external contours of the sugarcane stem. Then, the extracted overall profile images were optimized using a variety of image processing techniques to meet the requirements of harvesting. Lastly, the improved YOLOX model was utilized to identify the internal stem node features of sugarcane from the optimized overall images. The experimental results on a real sugarcane dataset show that the proposed external sugarcane stem segmentation model achieves a high mean intersection over union (MIoU) of 91.68% with an average segmentation time of just 0.025 seconds. Moreover, the proposed model for internal stem node recognition in sugarcane achieves an average precision (AP) of 96.19% with an average detection time of 0.026 seconds. Additionally, this study compares image segmentation models such as PSPNet and deepLabv3+ with target detection models such as YoloV5 and YoloV7. The experimental results show that the sugarcane feature extraction models proposed in this article all exhibit high accuracy and robustness.

关键词： Keywords. deep learning image segmentation Machine vision Sugarcane identification Target detection.

来源：评论

学校读者我要写书评

暂无评论

BSMO: Enhancing Multitask learning Through Batch Swapping Optimization

BSMO: Enhancing Multitask Learning Through Batch Swapping Op...

引用

Conference on real-time image processing and deep learning

作者： Spencer, Zephaniah Schirner, Gunar Tabkhi, Hamed UNC Charlotte 9201 Univ City Blvd Charlotte NC 28223 USA Northeastern Univ 360 Huntington Ave Boston MA 02115 USA

ISBN: (纸本)9781510673878;9781510673861

With the advent of deep learning, there has been an ever-growing list of applications to which deep Convolutional Neural Networks (DCNNs) can be applied. The field of Multi-Task learning (MTL) attempts to provide optimizations to many-task systems, improving performance by optimization algorithms and structural changes to these networks. However, we have found that current MTL optimization algorithms often impose burdensome computation overheads, require meticulously labeled datasets, and do not adapt to tasks with significantly different loss distributions. We propose a new MTL optimization algorithm: Batch Swapping with Multiple Optimizers (BSMO). We utilize single-task labeled data to train on a multi-task hard parameter sharing (HPS) network through swapping tasks at the batch level. This dramatically increases the flexibility and scalability of training on an HPS network by allowing for per-task datasets and augmentation pipelines. We demonstrate the efficacy of BSMO versus current SO TA algorithms by benchmarking across contemporary benchmarks & networks.

关键词： deep learning Multi-Task learning real-time Vision

来源：评论

学校读者我要写书评

暂无评论

Rcf-yolo: an underwater object detection algorithm based on improved YOLOv10n

引用

JOURNAL OF real-time image processing 2025年第2期22卷 1-18页

作者： Liang, Xiuman Zhang, Teng Yu, Haifeng Liu, Zhendong North China Univ Sci & Technol Sch Elect Engn Tangshan 063000 Peoples R China

Object detection is a key technology for marine exploration. The detection effect is not ideal because of factors such as the biodiversity and overlapping shadows in the underwater environment. Therefore, a new underwater object detection algorithm called RCF-YOLO is proposed. First, a coordinate enhancement (CE) attention module is designed. Depth-separable convolutions are used to extract the location information of the channel and combine it with spatial information to improve the model's ability to infer global features. Second, we have redesigned the neck with the BiFPN concept, which enhances feature interaction capabilities and optimizes the inference structure. The convolutional operation in the neck path is improved to enhance cross-scale connections, effectively integrating shallow and deep features, achieving a good balance between efficiency and accuracy. Finally, the receptive field convolution (RFAConv) is introduced to solve the parameter sharing problem in complex convolution processing, making the model more flexible in adjusting the convolution kernel weights and more effectively capturing the information in the image. The proposed model was compared with several sets of experiments on the URPC, DUO, and ROUD datasets. With a decrease in both the number of parameters and the complexity of the calculation, the accuracy reached 85.3%, 87.9%, and 84.9%. The experimental results show that the RCF-YOLO model has excellent performance in the underwater detection task.

关键词： deep learning Underwater object detection Attention mechanism YOLOv10n

来源：评论

学校读者我要写书评

暂无评论

Identification and characteristic analysis of internal defects in rock-filled concrete based on deep learning method

引用

MATERIALS TODAY COMMUNICATIONS 2024年 41卷

作者： Ren, Tao Zhang, Yufan Yang, Tao Chen, Qiaolin Yu, Lei Wang, Guoji Song, Meizhou Guizhou Univ Coll Civil Engn Guiyang 550025 Peoples R China Guiyang Shougang Real Estate Dev Co Guiyang 550025 Peoples R China

The distribution characteristics and geometric morphology characteristics of defects within RFC are important factors affecting the strength properties and rupture morphology of RFC. However, the excessive size of commonly used aggregates for RFC leads to difficulties in conducting in-depth experimental studies indoors. Based on the improved U-Net and image processing technology, this research establishes an integrated model for the identification, classification, and extraction of defects inside the RFC, quantitatively counts and analyzes the acquired defect distribution characteristics and geometrical morphology characteristics, and establishes a defect characteristic distribution function that can be used for the numerical reconstruction of defects. In order to realize the acceleration of U-Net training using training weights, use VGG-16 with the fully connected layer removed instead of the Encoder part of the U-Net. The integrated model in this research can realize automatic identification, classification, and extraction of multiple types of defects at the same time, and the established distribution function of defect characteristics provides a data basis and new ideas for the establishment of RFC three-dimensional numerical models containing real defects.

关键词： deep learning improved U -Net image processing technology defect characteristic function rock-filled concrete

来源：评论

学校读者我要写书评

暂无评论

Fast Monocular Measurement via deep learning-Based Object Detection for real-time Gas-Insulated Transmission Line Deformation Monitoring

引用

ENERGIES 2025年第8期18卷

作者： Yang, Guiyun Yang, Wengang Li, Entuo Wang, Qinglong Han, Huilong Sun, Jie Wang, Meng North China Elect Power Univ Dept Mech Engn Baoding 071003 Peoples R China North China Elect Power Univ Hebei Engn Res Ctr Adv Mfg & Intelligent Operat & Baoding 071003 Peoples R China North China Elect Power Univ Hebei Key Lab Elect Machinery Hlth Maintenance & F Baoding 071003 Peoples R China Northeastern Univ State Key Lab Rolling & Automat Shenyang 110819 Peoples R China

Deformation monitoring of Gas-Insulated Transmission Lines (GILs) is critical for the early detection of structural issues and for ensuring safe power transmission. In this study, we introduce a rapid monocular measurement method that leverages deep learning for real-time monitoring. A YOLOv10 model is developed for automatically identifying regions of interest (ROIs) that may exhibit deformations. Within these ROIs, grayscale data is used to dynamically set thresholds for FAST corner detection, while the Shi-Tomasi algorithm filters redundant corners to extract unique feature points for precise tracking. Subsequent subpixel refinement further enhances measurement accuracy. To correct image tilt, ArUco markers are employed for geometric correction and to compute a scaling factor based on their known edge lengths, thereby reducing errors caused by non-perpendicular camera angles. Simulated experiments validate our approach, demonstrating that combining refined ArUco marker coordinates with manually annotated features significantly improves detection accuracy. Our method achieves a mean absolute error of no more than 1.337 mm and a processing speed of approximately 0.024 s per frame, meeting the precision and efficiency requirements for GIL deformation monitoring. This integrated approach offers a robust solution for long-term, real-time monitoring of GIL deformations, with promising potential for practical applications in power transmission systems.

关键词： gas-insulated transmission lines (GILs) deep learning YOLO monocular measurement

来源：评论

学校读者我要写书评

暂无评论

Overcoming annotation bottlenecks in underwater fish segmentation: a robust self-supervised learning approach

引用

SIGNAL image AND VIDEO processing 2025年第4期19卷 1-10页

作者： Saleh, Alzayat Sheaves, Marcus Jerry, Dean Azghadi, Mostafa Rahimi James Cook Univ Coll Sci & Engn Townsville Qld Australia James Cook Univ ARC Res Hub Supercharging Trop Aquaculture Genet S Townsville Qld Australia

Accurate fish segmentation in underwater videos is challenging due to low visibility, variable lighting, and dynamic backgrounds, making fully-supervised methods that require manual annotation impractical for many applications. This paper introduces a novel self-supervised learning approach for fish segmentation using deep learning. Our model, trained without manual annotation, learns robust and generalizable representations by aligning features across augmented views and enforcing spatial-temporal consistency. We demonstrate its effectiveness on three challenging underwater video datasets: deepFish, Seagrass, and YouTube-VOS, surpassing existing self-supervised methods and achieving segmentation accuracy comparable to fully-supervised methods without the need for costly annotations. Trained on deepFish, our model exhibits strong generalization, achieving high segmentation accuracy on the unseen Seagrass and YouTube-VOS datasets. Furthermore, our model is computationally efficient due to its parallel processing and efficient anchor sampling technique, making it suitable for real-time applications and potential deployment on edge devices. We present quantitative results using Jaccard Index and Dice coefficient, as well as qualitative comparisons, showcasing the accuracy, robustness, and efficiency of our approach for advancing underwater video analysis.

关键词： Computer vision Convolutional neural networks Underwater videos deep learning Transformer Self-supervised learning

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

Comparison of deep generative models for real-time generation of synthesized defective wafer maps

Comparison of deep generative models for real-time generatio...

引用

Conference on real-time image processing and deep learning

作者： Alam, Lamia Kehtarnavaz, Nasser Univ Texas Dallas Dept Elect & Comp Engn Richardson TX 75080 USA

ISBN: (纸本)9781510673878;9781510673861

Modern wafer inspection systems in Integrated Circuit (IC) manufacturing utilize deep neural networks. The training of such networks requires the availability of a very large number of defective or faulty die patterns on a wafer called wafer maps. The number of defective wafer maps on a production line is often limited. In order to have a very large number of defective wafer maps for the training of deep neural networks, generative models can be utilized to generate realistic synthesized defective wafer maps. This paper compares the following three generative models that are commonly used for generating synthesized images: Generative Adversarial Network (GAN), Variational Auto-Encoder (VAE), and CycleGAN which is a variant of GAN. The comparison is carried out based on the public domain wafer map dataset WM-811K. The quality aspect of the generated wafer map images is evaluated by computing the five metrics of peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), inception score (IS), Frechet inception distance (FID), and kernel inception distance (KID). Furthermore, the computational efficiency of these generative networks is examined in terms of their deployment in a real-time inspection system.

关键词： Synthesized Wafer Maps Generative Adversarial Network Variational Autoencoder CycleGAN real-time image Synthesis

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：