Impact craters are important for understanding the evolution of lunar geologic and surface erosion rates,among other functions.However,the morphological characteristics of these micro impact craters are not obvious an...Impact craters are important for understanding the evolution of lunar geologic and surface erosion rates,among other functions.However,the morphological characteristics of these micro impact craters are not obvious and they are numerous,resulting in low detection accuracy by deep learning models.Therefore,we proposed a new multi-scale fusion crater detection algorithm(MSF-CDA)based on the YOLO11 to improve the accuracy of lunar impact crater detection,especially for small craters with a diameter of<1 km.Using the images taken by the LROC(Lunar Reconnaissance Orbiter Camera)at the Chang’e-4(CE-4)landing area,we constructed three separate datasets for craters with diameters of 0-70 m,70-140 m,and>140 m.We then trained three submodels separately with these three datasets.Additionally,we designed a slicing-amplifying-slicing strategy to enhance the ability to extract features from small craters.To handle redundant predictions,we proposed a new Non-Maximum Suppression with Area Filtering method to fuse the results in overlapping targets within the multi-scale submodels.Finally,our new MSF-CDA method achieved high detection performance,with the Precision,Recall,and F1 score having values of 0.991,0.987,and 0.989,respectively,perfectly addressing the problems induced by the lesser features and sample imbalance of small craters.Our MSF-CDA can provide strong data support for more in-depth study of the geological evolution of the lunar surface and finer geological age estimations.This strategy can also be used to detect other small objects with lesser features and sample imbalance problems.We detected approximately 500,000 impact craters in an area of approximately 214 km2 around the CE-4 landing area.By statistically analyzing the new data,we updated the distribution function of the number and diameter of impact craters.Finally,we identified the most suitable lighting conditions for detecting impact crater targets by analyzing the effect of different lighting conditions on the detection accuracy.展开更多
针对基于You Only Look Once v2算法的目标检测存在精度低及稳健性差的问题,提出一种车辆目标实时检测的You Only Look Once v2优化算法;该算法以You Only Look Once v2算法为基础,通过增加网络深度,增强特征提取能力,同时,通过添加残...针对基于You Only Look Once v2算法的目标检测存在精度低及稳健性差的问题,提出一种车辆目标实时检测的You Only Look Once v2优化算法;该算法以You Only Look Once v2算法为基础,通过增加网络深度,增强特征提取能力,同时,通过添加残差模块,解决网络深度增加带来的梯度消失或弥散问题;该方法将网络结构中低层特征与高层特征进行融合,提升对小目标车辆的检测精度。结果表明,通过在KITTI数据集上进行测试,优化后的算法在检测速度不变的情况下,提高了车辆目标检测精度,平均精度达到0.94,同时提升了小目标检测的准确性。展开更多
目标检测是计算机视觉的基石,在医疗成像、工业缺陷检测、自动驾驶和农业监测等领域有着广泛的应用。YOLO(You Only Look Once)作为单阶段目标检测的开创性算法,以其卓越的实时性和高效性成为计算机视觉领域的核心技术。从2016年YOLOv1...目标检测是计算机视觉的基石,在医疗成像、工业缺陷检测、自动驾驶和农业监测等领域有着广泛的应用。YOLO(You Only Look Once)作为单阶段目标检测的开创性算法,以其卓越的实时性和高效性成为计算机视觉领域的核心技术。从2016年YOLOv1的首次亮相到2025年YOLOv12的发布,YOLO通过持续优化网络架构、改进损失函数设计和创新数据增强策略,显著提升了检测性能和应用适配性。本文系统回顾了YOLO长达十年的发展历程,分析了其在不同领域的应用,并探讨了其快速发展背后的驱动力,包括深度学习技术的突破、开源社区的广泛协作以及对实时性和边缘计算的迫切需求。与双阶段算法(如Faster R-CNN)和基于Transformer的方法(如DETR)相比,YOLO在速度和工程部署上展现出显著优势,但在小目标检测、密集场景处理和复杂背景下的鲁棒性方面仍面临挑战。未来YOLO或可通过整合轻量级Transformer模块、优化多任务学习框架以及引入生成式AI增强数据多样性,有望进一步提升精度和场景适应能力。展开更多
Rapid and high-precision speed bump detection is critical for autonomous driving and road safety,yet it faces challenges from non-standard appearances and complex environments.To address this issue,this study proposes...Rapid and high-precision speed bump detection is critical for autonomous driving and road safety,yet it faces challenges from non-standard appearances and complex environments.To address this issue,this study proposes a you only look once(YOLO)algorithm for speed bump detection(SPD-YOLO),a lightweight model based on YOLO11s that integrates three core innova-tive modules to balance detection precision and computational efficiency:it replaces YOLO11s’original backbone with StarNet,which uses‘star operations’to map features into high-dimensional nonlinear spaces for enhanced feature representation while maintaining computational efficiency;its neck incorporates context feature calibration(CFC)and spatial feature calibration(SFC)to improve detection performance without significant computational overhead;and its detection head adopts a lightweight shared convolutional detection(LSCD)structure combined with GroupNorm,minimizing computational complexity while preserving multi-scale feature fusion efficacy.Experi-ments on a custom speed bump dataset show SPD-YOLO achieves a mean average precision(mAP)of 79.9%,surpassing YOLO11s by 1.3%and YOLO12s by 1.2%while reducing parameters by 26.3%and floating-point operations per second(FLOPs)by 29.5%,enabling real-time deploy-ment on resource-constrained platforms.展开更多
Fire detection has held stringent importance in computer vision for over half a century.The development of early fire detection strategies is pivotal to the realization of safe and smart cities,inhabitable in the futu...Fire detection has held stringent importance in computer vision for over half a century.The development of early fire detection strategies is pivotal to the realization of safe and smart cities,inhabitable in the future.However,the development of optimal fire and smoke detection models is hindered by limitations like publicly available datasets,lack of diversity,and class imbalance.In this work,we explore the possible ways forward to overcome these challenges posed by available datasets.We study the impact of a class-balanced dataset to improve the fire detection capability of state-of-the-art(SOTA)vision-based models and propose the use of generative models for data augmentation,as a future work direction.First,a comparative analysis of two prominent object detection architectures,You Only Look Once version 7(YOLOv7)and YOLOv8 has been carried out using a balanced dataset,where both models have been evaluated across various evaluation metrics including precision,recall,and mean Average Precision(mAP).The results are compared to other recent fire detection models,highlighting the superior performance and efficiency of the proposed YOLOv8 architecture as trained on our balanced dataset.Next,a fractal dimension analysis gives a deeper insight into the repetition of patterns in fire,and the effectiveness of the results has been demonstrated by a windowing-based inference approach.The proposed Slicing-Aided Hyper Inference(SAHI)improves the fire and smoke detection capability of YOLOv8 for real-life applications with a significantly improved mAP performance over a strict confidence threshold.YOLOv8 with SAHI inference gives a mAP:50-95 improvement of more than 25%compared to the base YOLOv8 model.The study also provides insights into future work direction by exploring the potential of generative models like deep convolutional generative adversarial network(DCGAN)and diffusion models like stable diffusion,for data augmentation.展开更多
To avoid colliding with trees during its operation,a lawn mower robot must detect the trees.Existing tree detection methods suffer from low detection accuracy(missed detection)and the lack of a lightweight model.In th...To avoid colliding with trees during its operation,a lawn mower robot must detect the trees.Existing tree detection methods suffer from low detection accuracy(missed detection)and the lack of a lightweight model.In this study,a dataset of trees was constructed on the basis of a real lawn environment.According to the theory of channel incremental depthwise convolution and residual suppression,the Embedded-A module is proposed,which expands the depth of the feature map twice to form a residual structure to improve the lightweight degree of the model.According to residual fusion theory,the Embedded-B module is proposed,which improves the accuracy of feature-map downsampling by depthwise convolution and pooling fusion.The Embedded YOLO object detection network is formed by stacking the embedded modules and the fusion of feature maps of different resolutions.Experimental results on the testing set show that the Embedded YOLO tree detection algorithm has 84.17%and 69.91%average precision values respectively for trunk and spherical tree,and 77.04% mean average precision value.The number of convolution parameters is 1.78×10^(6),and the calculation amount is 3.85 billion float operations per second.The size of weight file is 7.11MB,and the detection speed can reach 179 frame/s.This study provides a theoretical basis for the lightweight application of the object detection algorithm based on deep learning for lawn mower robots.展开更多
针对少精、弱精患者的家用精子检测仪嵌入式部署需求,提出一种基于改进you only look once(YOLO)v8n的轻量化精子检测算法.通过对YOLOv8n模型进行结构优化,在Neck层引入高效多分支尺度特征金字塔网络(EMBSFPN),在提高精度的同时保证了...针对少精、弱精患者的家用精子检测仪嵌入式部署需求,提出一种基于改进you only look once(YOLO)v8n的轻量化精子检测算法.通过对YOLOv8n模型进行结构优化,在Neck层引入高效多分支尺度特征金字塔网络(EMBSFPN),在提高精度的同时保证了模型的轻量化.在检测头部分采用轻量级共享可变形卷积检测(LSDECD)头替换原来的检测头,大大减少了模型的参数量和运算量.实验结果表明,改进后的算法在精子检测任务上实现了良好的性能,平均精度提高了2.3%,模型运算量减少了26.8%,为嵌入式系统上的精子检测应用提供了一种有效的解决方案.展开更多
For the detection of marine ship objects in radar images, large-scale networks based on deep learning are difficult to be deployed on existing radar-equipped devices. This paper proposes a lightweight convolutional ne...For the detection of marine ship objects in radar images, large-scale networks based on deep learning are difficult to be deployed on existing radar-equipped devices. This paper proposes a lightweight convolutional neural network, LiraNet, which combines the idea of dense connections, residual connections and group convolution, including stem blocks and extractor modules.The designed stem block uses a series of small convolutions to extract the input image features, and the extractor network adopts the designed two-way dense connection module, which further reduces the network operation complexity. Mounting LiraNet on the object detection framework Darknet, this paper proposes Lira-you only look once(Lira-YOLO), a lightweight model for ship detection in radar images, which can easily be deployed on the mobile devices. Lira-YOLO's prediction module uses a two-layer YOLO prediction layer and adds a residual module for better feature delivery. At the same time, in order to fully verify the performance of the model, mini-RD, a lightweight distance Doppler domain radar images dataset, is constructed. Experiments show that the network complexity of Lira-YOLO is low, being only 2.980 Bflops, and the parameter quantity is smaller, which is only 4.3 MB. The mean average precision(mAP) indicators on the mini-RD and SAR ship detection dataset(SSDD) reach 83.21% and 85.46%, respectively,which is comparable to the tiny-YOLOv3. Lira-YOLO has achieved a good detection accuracy with less memory and computational cost.展开更多
为了综合分析YOLO(You Only Look Once)算法在提升交通安全性和效率方面的重要作用,从“人-车-路”3个核心要素的角度,对YOLO算法在交通目标检测中的发展和研究现状进行系统性地总结.概述了YOLO算法常用的评价指标,详细阐述了这些指标...为了综合分析YOLO(You Only Look Once)算法在提升交通安全性和效率方面的重要作用,从“人-车-路”3个核心要素的角度,对YOLO算法在交通目标检测中的发展和研究现状进行系统性地总结.概述了YOLO算法常用的评价指标,详细阐述了这些指标在交通场景中的实际意义.对YOLO算法的核心架构进行概述,追溯了该算法的发展历程,分析各个版本迭代中的优化和改进措施.从“人-车-路”3种交通目标的视角出发,梳理并论述了采用YOLO算法进行交通目标检测的研究现状及应用情况.分析目前YOLO算法在交通目标检测中存在的局限性和挑战,提出相应的改进方法,展望未来的研究重点,为道路交通的智能化发展提供了研究参考.展开更多
目前在铝材表面缺陷检测领域,常用的检测模型存在检测精度不高、实时性不强和参数量大等问题。针对上述问题,对目标检测模型YOLOv8做了改进。首先,该模型使用自研的动态可变形卷积模块取代原有最后一层C2f模块;其次,将RT-DETR(real-time...目前在铝材表面缺陷检测领域,常用的检测模型存在检测精度不高、实时性不强和参数量大等问题。针对上述问题,对目标检测模型YOLOv8做了改进。首先,该模型使用自研的动态可变形卷积模块取代原有最后一层C2f模块;其次,将RT-DETR(real-time detection transformer)模型中的检测头移植到新模型中,利用Transformer中解码器的思想,省去了模型后处理的步骤;最后,将NWD(normalized Wasserstein distance)loss和WIoUv3(wise-intersection over union version 3)loss结合作为改进后模型的回归损失函数,使模型动态筛选锚框,解决IoU对于不同尺寸缺陷类型敏感度不同导致的标签分配不准确的问题。改进之后的模型相比于基线模型YOLOv8,平均精度提高了3.8%,每秒处理帧数提高至92 f/s,同时在钢材表面缺陷检测数据集和太阳能电池板表面缺陷检测数据集上也具有很强的鲁棒性,新模型在实时检测、实际部署中具有很大优势。展开更多
基金the National Key Research and Development Program of China (Grant No.2022YFF0711400)the National Space Science Data Center Youth Open Project (Grant No. NSSDC2302001)
文摘Impact craters are important for understanding the evolution of lunar geologic and surface erosion rates,among other functions.However,the morphological characteristics of these micro impact craters are not obvious and they are numerous,resulting in low detection accuracy by deep learning models.Therefore,we proposed a new multi-scale fusion crater detection algorithm(MSF-CDA)based on the YOLO11 to improve the accuracy of lunar impact crater detection,especially for small craters with a diameter of<1 km.Using the images taken by the LROC(Lunar Reconnaissance Orbiter Camera)at the Chang’e-4(CE-4)landing area,we constructed three separate datasets for craters with diameters of 0-70 m,70-140 m,and>140 m.We then trained three submodels separately with these three datasets.Additionally,we designed a slicing-amplifying-slicing strategy to enhance the ability to extract features from small craters.To handle redundant predictions,we proposed a new Non-Maximum Suppression with Area Filtering method to fuse the results in overlapping targets within the multi-scale submodels.Finally,our new MSF-CDA method achieved high detection performance,with the Precision,Recall,and F1 score having values of 0.991,0.987,and 0.989,respectively,perfectly addressing the problems induced by the lesser features and sample imbalance of small craters.Our MSF-CDA can provide strong data support for more in-depth study of the geological evolution of the lunar surface and finer geological age estimations.This strategy can also be used to detect other small objects with lesser features and sample imbalance problems.We detected approximately 500,000 impact craters in an area of approximately 214 km2 around the CE-4 landing area.By statistically analyzing the new data,we updated the distribution function of the number and diameter of impact craters.Finally,we identified the most suitable lighting conditions for detecting impact crater targets by analyzing the effect of different lighting conditions on the detection accuracy.
文摘针对基于You Only Look Once v2算法的目标检测存在精度低及稳健性差的问题,提出一种车辆目标实时检测的You Only Look Once v2优化算法;该算法以You Only Look Once v2算法为基础,通过增加网络深度,增强特征提取能力,同时,通过添加残差模块,解决网络深度增加带来的梯度消失或弥散问题;该方法将网络结构中低层特征与高层特征进行融合,提升对小目标车辆的检测精度。结果表明,通过在KITTI数据集上进行测试,优化后的算法在检测速度不变的情况下,提高了车辆目标检测精度,平均精度达到0.94,同时提升了小目标检测的准确性。
文摘目标检测是计算机视觉的基石,在医疗成像、工业缺陷检测、自动驾驶和农业监测等领域有着广泛的应用。YOLO(You Only Look Once)作为单阶段目标检测的开创性算法,以其卓越的实时性和高效性成为计算机视觉领域的核心技术。从2016年YOLOv1的首次亮相到2025年YOLOv12的发布,YOLO通过持续优化网络架构、改进损失函数设计和创新数据增强策略,显著提升了检测性能和应用适配性。本文系统回顾了YOLO长达十年的发展历程,分析了其在不同领域的应用,并探讨了其快速发展背后的驱动力,包括深度学习技术的突破、开源社区的广泛协作以及对实时性和边缘计算的迫切需求。与双阶段算法(如Faster R-CNN)和基于Transformer的方法(如DETR)相比,YOLO在速度和工程部署上展现出显著优势,但在小目标检测、密集场景处理和复杂背景下的鲁棒性方面仍面临挑战。未来YOLO或可通过整合轻量级Transformer模块、优化多任务学习框架以及引入生成式AI增强数据多样性,有望进一步提升精度和场景适应能力。
文摘Rapid and high-precision speed bump detection is critical for autonomous driving and road safety,yet it faces challenges from non-standard appearances and complex environments.To address this issue,this study proposes a you only look once(YOLO)algorithm for speed bump detection(SPD-YOLO),a lightweight model based on YOLO11s that integrates three core innova-tive modules to balance detection precision and computational efficiency:it replaces YOLO11s’original backbone with StarNet,which uses‘star operations’to map features into high-dimensional nonlinear spaces for enhanced feature representation while maintaining computational efficiency;its neck incorporates context feature calibration(CFC)and spatial feature calibration(SFC)to improve detection performance without significant computational overhead;and its detection head adopts a lightweight shared convolutional detection(LSCD)structure combined with GroupNorm,minimizing computational complexity while preserving multi-scale feature fusion efficacy.Experi-ments on a custom speed bump dataset show SPD-YOLO achieves a mean average precision(mAP)of 79.9%,surpassing YOLO11s by 1.3%and YOLO12s by 1.2%while reducing parameters by 26.3%and floating-point operations per second(FLOPs)by 29.5%,enabling real-time deploy-ment on resource-constrained platforms.
基金supported by a grant from R&D Program Development of Rail-Specific Digital Resource Technology Based on an AI-Enabled Rail Support Platform,grant number PK2401C1,of the Korea Railroad Research Institute.
文摘Fire detection has held stringent importance in computer vision for over half a century.The development of early fire detection strategies is pivotal to the realization of safe and smart cities,inhabitable in the future.However,the development of optimal fire and smoke detection models is hindered by limitations like publicly available datasets,lack of diversity,and class imbalance.In this work,we explore the possible ways forward to overcome these challenges posed by available datasets.We study the impact of a class-balanced dataset to improve the fire detection capability of state-of-the-art(SOTA)vision-based models and propose the use of generative models for data augmentation,as a future work direction.First,a comparative analysis of two prominent object detection architectures,You Only Look Once version 7(YOLOv7)and YOLOv8 has been carried out using a balanced dataset,where both models have been evaluated across various evaluation metrics including precision,recall,and mean Average Precision(mAP).The results are compared to other recent fire detection models,highlighting the superior performance and efficiency of the proposed YOLOv8 architecture as trained on our balanced dataset.Next,a fractal dimension analysis gives a deeper insight into the repetition of patterns in fire,and the effectiveness of the results has been demonstrated by a windowing-based inference approach.The proposed Slicing-Aided Hyper Inference(SAHI)improves the fire and smoke detection capability of YOLOv8 for real-life applications with a significantly improved mAP performance over a strict confidence threshold.YOLOv8 with SAHI inference gives a mAP:50-95 improvement of more than 25%compared to the base YOLOv8 model.The study also provides insights into future work direction by exploring the potential of generative models like deep convolutional generative adversarial network(DCGAN)and diffusion models like stable diffusion,for data augmentation.
基金the National Natural Science Foundation of China (No.51275223)。
文摘To avoid colliding with trees during its operation,a lawn mower robot must detect the trees.Existing tree detection methods suffer from low detection accuracy(missed detection)and the lack of a lightweight model.In this study,a dataset of trees was constructed on the basis of a real lawn environment.According to the theory of channel incremental depthwise convolution and residual suppression,the Embedded-A module is proposed,which expands the depth of the feature map twice to form a residual structure to improve the lightweight degree of the model.According to residual fusion theory,the Embedded-B module is proposed,which improves the accuracy of feature-map downsampling by depthwise convolution and pooling fusion.The Embedded YOLO object detection network is formed by stacking the embedded modules and the fusion of feature maps of different resolutions.Experimental results on the testing set show that the Embedded YOLO tree detection algorithm has 84.17%and 69.91%average precision values respectively for trunk and spherical tree,and 77.04% mean average precision value.The number of convolution parameters is 1.78×10^(6),and the calculation amount is 3.85 billion float operations per second.The size of weight file is 7.11MB,and the detection speed can reach 179 frame/s.This study provides a theoretical basis for the lightweight application of the object detection algorithm based on deep learning for lawn mower robots.
文摘针对少精、弱精患者的家用精子检测仪嵌入式部署需求,提出一种基于改进you only look once(YOLO)v8n的轻量化精子检测算法.通过对YOLOv8n模型进行结构优化,在Neck层引入高效多分支尺度特征金字塔网络(EMBSFPN),在提高精度的同时保证了模型的轻量化.在检测头部分采用轻量级共享可变形卷积检测(LSDECD)头替换原来的检测头,大大减少了模型的参数量和运算量.实验结果表明,改进后的算法在精子检测任务上实现了良好的性能,平均精度提高了2.3%,模型运算量减少了26.8%,为嵌入式系统上的精子检测应用提供了一种有效的解决方案.
基金supported by the Joint Fund of Equipment Pre-Research and Aerospace Science and Industry (6141B07090102)。
文摘For the detection of marine ship objects in radar images, large-scale networks based on deep learning are difficult to be deployed on existing radar-equipped devices. This paper proposes a lightweight convolutional neural network, LiraNet, which combines the idea of dense connections, residual connections and group convolution, including stem blocks and extractor modules.The designed stem block uses a series of small convolutions to extract the input image features, and the extractor network adopts the designed two-way dense connection module, which further reduces the network operation complexity. Mounting LiraNet on the object detection framework Darknet, this paper proposes Lira-you only look once(Lira-YOLO), a lightweight model for ship detection in radar images, which can easily be deployed on the mobile devices. Lira-YOLO's prediction module uses a two-layer YOLO prediction layer and adds a residual module for better feature delivery. At the same time, in order to fully verify the performance of the model, mini-RD, a lightweight distance Doppler domain radar images dataset, is constructed. Experiments show that the network complexity of Lira-YOLO is low, being only 2.980 Bflops, and the parameter quantity is smaller, which is only 4.3 MB. The mean average precision(mAP) indicators on the mini-RD and SAR ship detection dataset(SSDD) reach 83.21% and 85.46%, respectively,which is comparable to the tiny-YOLOv3. Lira-YOLO has achieved a good detection accuracy with less memory and computational cost.
文摘为了综合分析YOLO(You Only Look Once)算法在提升交通安全性和效率方面的重要作用,从“人-车-路”3个核心要素的角度,对YOLO算法在交通目标检测中的发展和研究现状进行系统性地总结.概述了YOLO算法常用的评价指标,详细阐述了这些指标在交通场景中的实际意义.对YOLO算法的核心架构进行概述,追溯了该算法的发展历程,分析各个版本迭代中的优化和改进措施.从“人-车-路”3种交通目标的视角出发,梳理并论述了采用YOLO算法进行交通目标检测的研究现状及应用情况.分析目前YOLO算法在交通目标检测中存在的局限性和挑战,提出相应的改进方法,展望未来的研究重点,为道路交通的智能化发展提供了研究参考.
文摘目前在铝材表面缺陷检测领域,常用的检测模型存在检测精度不高、实时性不强和参数量大等问题。针对上述问题,对目标检测模型YOLOv8做了改进。首先,该模型使用自研的动态可变形卷积模块取代原有最后一层C2f模块;其次,将RT-DETR(real-time detection transformer)模型中的检测头移植到新模型中,利用Transformer中解码器的思想,省去了模型后处理的步骤;最后,将NWD(normalized Wasserstein distance)loss和WIoUv3(wise-intersection over union version 3)loss结合作为改进后模型的回归损失函数,使模型动态筛选锚框,解决IoU对于不同尺寸缺陷类型敏感度不同导致的标签分配不准确的问题。改进之后的模型相比于基线模型YOLOv8,平均精度提高了3.8%,每秒处理帧数提高至92 f/s,同时在钢材表面缺陷检测数据集和太阳能电池板表面缺陷检测数据集上也具有很强的鲁棒性,新模型在实时检测、实际部署中具有很大优势。
