检索结果-内蒙古大学图书馆

SST-YOLOv5s: advancing real-time blood cell object detection through multi-headed attention mechanism

SIGNAL IMAGE AND VIDEO PROCESSING 2025年第3期19卷 1-17页

作者： Zhang, Mingyu Chen, Jiaqing Xiamen Univ Wang Yanan Inst Studies Econ Xiamen 361005 Peoples R China Wuhan Univ Technol Coll Math & Stat Wuhan 430070 Peoples R China

An important method of diagnosing diseases in medicine is the examination of blood samples in which the blood cells in the blood sample are examined. To address the challenges of detecting small and variable blood cells in real-time, we introduce a small-target Swin Transformer YOLOv5s model named SST-YOLOv5s in this study. Based on YOLOv5s, the small object detection layer is added to enhance the feature extraction capability of YOLOv5s and improve the detection of small targets in the dataset. In this way, the C3 module is replaced by the Swin Transformer Blocks, which enhances the extraction of global information from the model as well as speeds up the computation of the model. Subsequently, considering the need for the testing speed of the model, Simplify Optimal Transport Assignment is used in this research to reduce the testing time of the model greatly. The experimental results demonstrate that compared with the baseline model of YOLOv5s, the SST-YOLOv5s model achieves a mAP0.5 of 93.0%, a mAP0.5:0.95 of 60.4%, and a Latency of 0.6. To the best of our knowledge, SST-YOLOv5s is the first YOLOv5s to add Swin Transformer and SimOTA to a small-target detection head and use it to detect cells and further proves that SST-YOLOv5s can be better applied in practical medical applications.

关键词： Deep learning Medical image process Blood cell YOLOv5s object detection

来源：评论

学校读者我要写书评

暂无评论

Toward Better Accuracy-Efficiency Tradeoffs for Oriented SAR Ship object detection

引用

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 9666-9681页

作者： Ju, Moran Niu, Buniu Li, Mulin Mao, Tengkai Jin, Si-nian Dalian Maritime Univ Coll Informat Sci & Technol Dalian 116026 Peoples R China

In oriented synthetic aperture radar (SAR) ship detection task, convolutional neural network based detectors have dramatically improved the detection performance, but enormous parameters make it difficult to realize model lightweighting. Recently, DETR and its variants have demonstrated excellent performance in object detection task, while model construction through linear layers has great potential in terms of model lightweighting. However, DETR-based models are rarely applied to oriented object detection task, while the network structure relies on manual experience and cannot be designed automatically. In this article, we propose a novel neural architecture search based lightweight detector in polar coordinate system with DETR as search space for oriented SAR ship detection, where oriented bounding boxes are encoded and decoded in polar coordinate system to cope with boundary discontinuity problems, and the weight entanglement strategy is adopted to realize automatic and lightweight design of DETR. Meanwhile, we design an oriented multiscale attention to alleviate the problem of sampling a large amount of background due to offset learning. Furthermore, we introduce a downsampling feedforward network to significantly reduce network floating point operations. Finally, we transplant FPDDet head as auxiliary head to improve encoder potential ship feature learning and decoder cross-attention learning. Experimental results show that our models not only achieve DETR lightweighting and real-time detection, but also improve detection performance. Our base models achieve state-of-the-art performance on both RSSDD and RSDD datasets compared to previous best models, with 1.36% and 2.28% improvement in mAP with 32.67 and 32.14 GFLOPs, respectively.

关键词： Marine vehicles Detectors Radar polarimetry object detection Decoding Transformers Accuracy Search problems Feature extraction Synthetic aperture radar DETR neural architecture search (NAS) oriented ship detection synthetic aperture radar (SAR)

来源：评论

学校读者我要写书评

暂无评论

SA-LfV: self-annotated labeling from videos for object detection

引用

MACHINE LEARNING 2025年第1期114卷 1-24页

作者： Sivapuram, Arun Kumar Komuravelli, Prashanth Gorthi, Rama Krishna Sai Indian Inst Technol Tirupati Dept Elect Engn Tirupati Andhra Pradesh India Indian Inst Technol Tirupati Dept Comp Sci & Engn Tirupati Andhra Pradesh India

In the realm of object detection, the remarkable strides made by deep neural networks over the past decade have been hampered by challenges such as data labeling and the need to capture natural variations in training samples. Existing benchmark datasets are confined with limited set of classes, and natural variations. This paper presents "SA-LfV", a novel framework designed to streamline object detection from videos with minimal human input. By utilizing basic computer vision tasks, such as image classification and tracking single objects, our method generates pseudo-labels for object detection efficiently. To ensure a rich variety of training samples, we introduce two innovative sampling strategies. The first applies density-based clustering, choosing samples that represent a wide range of scenarios. The second analyzes object movements and their mutual information, capturing diverse behaviors and appearances. The proposed object detection data labeling procedure is demonstrated on object-tracking datasets and custom-downloaded videos. Through these methods, our framework has produced a dataset with 70,000 pseudo-labeled bounding boxes across 13 object classes, significantly diversifying the available data for object detection tasks. Our experiments show that the proposed framework can effectively adapt to unlabelled ImageNet classes, indicating its potential to broaden the capabilities of object detection models. Moreover, integrating our self-annotated dataset with standard benchmark datasets leads to a notable improvement in object detection performance. This new approach not only simplifies the traditionally labor-intensive process of manual labeling but also paves the way for expanding object detection to a wider range of classes and applications.

关键词： object detection Tracking Labeling Clustering Data augmentation

来源：评论

学校读者我要写书评

暂无评论

Enhanced YOLOv10 Framework Featuring DPAM and DALSM for Real-Time Underwater object detection

引用

IEEE ACCESS 2025年 13卷 8691-8708页

作者： Sriram, Suthir Aburvan, P. Kaarthic, T. P. Arun Vijayaraj, Nivethitha Murugan, Thangavel Amrita Vishwa Vidyapeetham Amrita Sch Comp Dept Comp Sci & Engn Chennai 601103 India United Arab Emirates Univ Coll Informat Technol Al Ain U Arab Emirates

Recently, spotting underwater objects has been increasingly difficult due to the complexities of marine environments and varied visibility conditions. YOLOv10 is notable for its effective, robust architecture, featuring significant components: advanced backbone networks and enhanced feature pyramid networks that also deliver anchor-free detection. In delivering YOLOv10, we enhance it with dual partial attention mechanism (DPAM) and dual adaptive label assignment with sun glint removal module (DALSM) along with marine fusion loss (MFL). With DPAM, the latest refinement processes for conservation focus on feature extraction to account for key highlights and in the scene, include temporal context, both critical for interpretation of the dynamic realm below the ocean. The prefix with DALSM involves adaptive dual label assignment and techniques for sun glint removal. The marine fusion loss (MFL) provides an object detection prediction that combines both binary cross-entropy loss and complete intersection over union (CIoU) loss to enhance bounding box localization while also including spatial context to incorporate important underwater features. With the experiments, we attenuate enhancements with device gradient clipping, model checkpointing, and advanced augmentation processes to gain 3.04% improvement in mean Average Precision (mAP). These enhancements mitigate perceived challenges of underwater detection while enhancing knowledge of understanding marine life.

关键词： object detection Accuracy Feature extraction Fish Real-time systems Biological system modeling Sun Attention mechanisms Visualization Lighting Dual partial attention mechanism (DPAM) dual adaptive label assignment with sun glint removal module (DALSM) marine-fusion loss (MFL) YOLOv10

来源：评论

学校读者我要写书评

暂无评论

SIHENet: Semantic Interaction and Hierarchical Embedding Network for 360° Salient object detection

引用

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2025年 74卷

作者： He, Zhentao Shao, Feng Xie, Zhengxuan Chai, Xiongli Ho, Yo-Sung Ningbo Univ Fac Informat Sci & Engn Ningbo 315211 Peoples R China Gwangju Inst Sci & Technol GIST Sch Informat & Commun Gwangju 500712 South Korea

With the application of panoramic imaging instruments in environmental perception and measurement, 360 degrees salient object detection (SOD) has become a concern. However, how to reduce the serious distortion of 360 degrees images is still an open question. In this article, we fully exploit the intrinsic information complementarity of cross-projection and cross-level features and innovatively propose a semantic interaction and hierarchical embedding network (SIHENet) to realize 360 degrees SOD. In terms of fusion of different projection features, considering that the distortion of cubemap projection (CMP) images is relatively small, we propose a multiscale hierarchical embedding (MSHE) module to establish the context of CMP images and embed equirectangular projection (ERP) images in it to reduce distortion. In terms of cross-level feature fusion, in order to solve the feature dilution problem in feature transfer, we propose a multiprojection semantic interaction (MPSI) module to integrate the high-level information of ERP and CMP images. In order to solve the edge ambiguity problem, we propose a multilevel semantic matching (MLSM) module to enhance the edge information at different levels of ERP image. Extensive experiments on three public 360 degrees datasets demonstrate the competitive performance of the proposed model in comparison to state-of-the-art (SOTA) 360 degrees SOD models.

关键词： Feature extraction Semantics Image edge detection Distortion object detection Optical distortion Image segmentation Data mining Deformation Shape 360 degrees omnidirectional images cross-level feature cross-projection feature salient object detection (SOD)

来源：评论

学校读者我要写书评

暂无评论

Remote Sensing Small object detection Based on Multicontextual Information Aggregation

引用

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 8248-8260页

作者： Wang, Jingyu Ma, Mingrui Huang, Pengfei Mei, Shaohui Zhang, Liang Wang, Hongmei Northwestern Polytech Univ Sch Astronaut Xian 710072 Peoples R China Natl Key Lab Air Based Informat Percept & Fus Luoyang 471000 Peoples R China Northwestern Polytech Univ Unmanned Syst Res Inst Xian 710072 Peoples R China Northwestern Polytech Univ Sch Elect & Informat Xian 710129 Peoples R China

Due to wide field of view and background confusion, remote sensing objects are small and densely packed, hence commonly used detection methods detecting small objects are not satisfactory. In this article, we propose the multicontextual information aggregation YOLO (MCIA-YOLO) method, combining three novel modules to effectively aggregate multicontextual information across channels, depths, and pixels. First, the channel-spatial information aggregation module assembles spatial global features pursuant to channel contextual information, increasing the density of key information. Second, the shallow-deep information sparse aggregation module applies a sparse cross self-attention mechanism. By sparsely correlating long-range dependency information across different regions, the representation capability of a small target is enhanced while removing redundant information. Third, to enrich local multiscale features and better identify dense targets, multiscale weighted aggregation module convolves multireceptive field information and performs weighted fusion. Our method demonstrates satisfactory performance on dataset VisDrone2019, UAVDT, and NWPU VHR-10, especially in small objects detection, surpassing several state-of-the-art methods.

关键词： Feature extraction Remote sensing object detection Data mining Accuracy Detectors Aggregates Transformers Shape YOLO Global contextual information multicontextual information multireceptive field enhancement small object detection sparse cross self-attention

来源：评论

学校读者我要写书评

暂无评论

Real-time CNN-based object detection of prohibited items for X-ray security screening

引用

RADIATION PHYSICS AND CHEMISTRY 2025年 232卷

作者： Park, Junsung An, Geunyoung Lee, Byeong-No Seo, Hee Jeonbuk Natl Univ Dept Appl Plasma & Quantum Beam Engn Jeonju South Korea Korea Atom Energy Res Inst Dept Res Radiat Convergence Jeongeup South Korea Jeonbuk Natl Univ Dept Quantum Syst Engn Jeonju South Korea

In homeland security, X-ray security scanners are essential for detecting contraband, including weapons, drugs, and hazardous materials. However, detection accuracy might be limited when items overlap or are concealed in complex baggage configurations. To address this issue, we developed a one-stage convolutional neural network (CNN) object detection model based on the You Only Look Once (YOLO) architecture, which employs a CSPDarknet53 backbone for feature extraction. This model was designed for X-ray security scanners to enhance the detection of prohibited items in carry-on baggage, even under challenging conditions. The model was trained on a dataset comprising 6000 annotated X-ray images of suitcases, simulating real-world baggage scenarios with varying degrees of object overlap. Prohibited items were categorized into 22 classes based on the relevant International Civil Aviation Organization (ICAO) standards and Korean law, covering sharp objects (e.g., scissors, knives), weapons (e.g., guns, shurikens), and hazardous materials (e.g., liquids, flammables). To improve generalization, extensive image augmentation techniques were applied during training. Performance was evaluated using AP@50 (average precision at an IoU threshold of 0.5) and the F1-score, averaging 0.87 and 0.86 across all classes, respectively. Additionally, the average inference time per baggage item was 12.58 ms, making the system suitable for real-time airport security applications. These results demonstrate the potential of artificial intelligence (AI)-based security screening as a practical, high-speed detection solution for real-world airport security applications. This study addresses occlusion challenges in security screening by integrating multi-stage object overlap handling and dual-energy imaging. Compared to existing approaches, it could have higher applicability in complex baggage screening environments. While the model effectively detects structured prohibited items, further improvem

关键词： Transmission X-ray security screening Convolutional neural network object detection Prohibited item

来源：评论

学校读者我要写书评

暂无评论

PFRNet: A Small object detection Method Based on Parallel Feature Extraction and Attention Mechanism

引用

IEEE ACCESS 2025年 13卷 26727-26738页

作者： Lin, Hai Wang, Ji Li, Jingguo Zhanjiang Presch Educ Coll Zhanjiang 524084 Peoples R China Zhanjiang Ocean Univ Zhanjiang 524088 Guangdong Peoples R China

To address the challenge of detecting small objects in aerial and satellite remote sensing images with low-resolution, we propose a high-precision object detection method based on PFRNet. PFRNet incorporates parallel feature extraction branches and a progressive feature refinement mechanism, significantly enhancing the model's ability to perceive detailed features. In addition, PFRNet introduces the spatial pyramid pooling fusion with spatial attention (SPPFSPA) module, which integrates multi-scale features with an attention mechanism, enabling the model to better focus on areas of interest, thereby improving detection performance. Results demonstrate that PFRNet achieves outstanding detection accuracy, markedly outperforming other algorithms, particularly in small object detection. Visualization analysis reveals that the PFR module effectively captures richer and more comprehensive visual features in images, providing robust input for subsequent detection tasks, which is crucial for PFRNet's superior performance. Overall, the proposed PFRNet model makes significant strides in small object detection in UAV aerial and satellite remote sensing images, offering strong support for applications such as intelligent transportation and precision agriculture.

关键词： Feature extraction object detection Attention mechanisms Satellites Remote sensing Convolution Computational modeling Accuracy Neck Semantics UAV aerial imagery satellite remote sensing parallel feature extraction

来源：评论

学校读者我要写书评

暂无评论

Foregroundness-Aware Task Disentanglement and Self-Paced Curriculum Learning for Domain Adaptive object detection

引用

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS 2025年第1期36卷 369-380页

作者： Liu, Yabo Wang, Jinghua Xiao, Linhui Liu, Chengliang Wu, Zhihao Xu, Yong Harbin Inst Technol Shenzhen Key Lab Visual Object Detect & Recognit Shenzhen 518055 Peoples R China Peng Cheng Lab Shenzhen 518055 Peoples R China Harbin Inst Technol Sch Comp Sci & Technol Shenzhen 518055 Peoples R China Chinese Acad Sci Inst Automat Beijing 100190 Peoples R China

Unsupervised domain adaptive object detection (UDA-OD) is a challenging problem since it needs to locate and recognize objects while maintaining the generalization ability across domains. Most existing UDA-OD methods directly integrate the adaptive modules into the detectors. This integration procedure can significantly sacrifice the detection performances, though it enhances the generalization ability. To solve this problem, we propose an effective framework, named foregroundness-aware task disentanglement and self-paced curriculum adaptation (FA-TDCA), to disentangle the UDA-OD task into four independent subtasks of source detector pretraining, classification adaptation, location adaptation, and target detector training. The disentanglement can transfer the knowledge effectively while maintaining the detection performance of our model. In addition, we propose a new metric, i.e., foregroundness, and use it to evaluate the confidence of the location result. We use both foregroundness and classification confidence to assess the label quality of the proposals. For effective knowledge transfer across domains, we utilize a self-paced curriculum learning paradigm to train adaptors and gradually improve the quality of the pseudolabels associated with the target samples. Experiment results indicate that our method achieves state-of-the-art results on four cross-domain object detection tasks.

关键词： Proposals Task analysis Detectors object detection Measurement Training Learning systems Curriculum learning domain adaptive object detection location metric task disentanglement

来源：评论

学校读者我要写书评

暂无评论

CGDINet: A Deep Learning-Based Salient object detection Algorithm

引用

IEEE ACCESS 2025年 13卷 4697-4723页

作者： Hu, Chengyu Guo, Jianxin Xie, Hanfei Zhu, Qing Yuan, Baoxi Gao, Yujie Ma, Xiangyang Chen, Jialu Tian, Juan Xijing Univ Sch Elect Informat Xian 710123 Peoples R China Xijing Univ Xian Key Lab High Precis Ind Intelligent Vis Measu Xian 710123 Peoples R China Beijing Hengyue Intelligent Informat Technol Co Lt Xian 710061 Peoples R China Xijing Univ Sch Humanities & Educ Xian 710123 Peoples R China

Salient object detection (SOD) is a key preprocessing step in computer vision, widely used in object tracking, action recognition, and image retrieval, among other fields. However, traditional SOD algorithms often face issues such as rough object boundaries, incomplete extraction of global image features, and insufficient attention to key areas. To address these problems, an improved significance object detection network-CGDINet (Coordinate Attention-Group Consensus Aggregation Module-Depth Auxiliary Module-Inverse Saliency Pyramid Reconstruction Network)-is proposed. CGDINet introduces the Group Consensus Aggregation Module (GCAM) embedded with the Coordinate Attention (CA) mechanism, named CAGM (Coordinate Attention-Group Consensus Aggregation Module), to enhance global feature capture capabilities and improve the processing of directional features. Additionally, the Depth Auxiliary Module (DAM) is incorporated to enhance the focus on important regions. Experiments were conducted on five public datasets (DUT-OMRON, ECSSD, PASCAL-S, DUTS-TE, and HKU-IS). The results show that CGDINet outperforms other mainstream significance object detection models in evaluation metrics such as maxF(beta) , S-alpha , and MAE, with almost no increase in computational cost (FLOPs) and parameters. The experimental results validate that CGDINet can effectively address the issues of incomplete global feature extraction and insufficient attention to key areas, thereby significantly enhancing the performance of significance object detection

关键词： Feature extraction Transformers Context modeling object detection Computational modeling Accuracy Image edge detection Dams Convolutional neural networks Attention mechanisms Salient object detection deep learning InSPyReNet CAGM DAM

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：