The unique advantage of x-ray ghost imaging(XGI)is its potential in low dose radiology.One of the practical ways to reduce the radiation exposure is to reduce the measurements while remaining sufficient image quality....The unique advantage of x-ray ghost imaging(XGI)is its potential in low dose radiology.One of the practical ways to reduce the radiation exposure is to reduce the measurements while remaining sufficient image quality.Synthetic aperture x-ray ghost imaging(SAXGI)is invented to achieve megapixel XGI with limited measurements,which is expected to implement XGI simultaneously with large field of view and low radiation exposure.In this paper,we experimentally investigate the effect of measurements reduction on the spatial resolution and image quality of SAXGI with standard sample and biomedical specimen.The results with a resolution chart demonstrated that at 360 measurements,SAXGI successfully retrieved the sample image of 1960×1960 pixels with spatial resolution of 4μm.With measurement reduction,the spatial resolution deteriorates but the sparser structures are still discernable.Even with measurements reduced to 10,a spatial resolution of 10μm can still be achieved by SAXGI.A biomedical sample of a fish specimen is employed to evaluate the method and the fish image of 2000×1000 pixels with an SSIM of 0.962 is reconstructed by SAXGI with 770measurements,corresponding to an accumulative exposure reduction of more than 2 times.With the measurements reduced to 10 which corresponds to 1/160 of the accumulative radiation exposure for conventional radiology,bulky structure like the fish skeleton can still be definitely discerned and the SSIM for the reconstructed image still retained 0.9179.Results of this paper demonstrate that measurements reduction is practicable for the radiation exposure reduction of the sample,which implicates that SAXGI with limited measurements is an efficient solution for low dose radiology.展开更多
Accurate vehicle detection is essential for autonomous driving,traffic monitoring,and intelligent transportation systems.This paper presents an enhanced YOLOv8n model that incorporates the Ghost Module,Convolutional B...Accurate vehicle detection is essential for autonomous driving,traffic monitoring,and intelligent transportation systems.This paper presents an enhanced YOLOv8n model that incorporates the Ghost Module,Convolutional Block Attention Module(CBAM),and Deformable Convolutional Networks v2(DCNv2).The Ghost Module streamlines feature generation to reduce redundancy,CBAM applies channel and spatial attention to improve feature focus,and DCNv2 enables adaptability to geometric variations in vehicle shapes.These components work together to improve both accuracy and computational efficiency.Evaluated on the KITTI dataset,the proposed model achieves 95.4%mAP@0.5—an 8.97% gain over standard YOLOv8n—along with 96.2% precision,93.7% recall,and a 94.93%F1-score.Comparative analysis with seven state-of-the-art detectors demonstrates consistent superiority in key performance metrics.An ablation study is also conducted to quantify the individual and combined contributions of GhostModule,CBAM,and DCNv2,highlighting their effectiveness in improving detection performance.By addressing feature redundancy,attention refinement,and spatial adaptability,the proposed model offers a robust and scalable solution for vehicle detection across diverse traffic scenarios.展开更多
为了降低服装目标检测模型的参数量和浮点型计算量,提出一种改进的轻量级服装目标检测模型——GYOLOv5s.首先使用Ghost卷积重构YOLOv5s的主干网络;然后使用DeepFashion2数据集中的部分数据进行模型训练和验证;最后将训练好的模型用于服...为了降低服装目标检测模型的参数量和浮点型计算量,提出一种改进的轻量级服装目标检测模型——GYOLOv5s.首先使用Ghost卷积重构YOLOv5s的主干网络;然后使用DeepFashion2数据集中的部分数据进行模型训练和验证;最后将训练好的模型用于服装图像的目标检测.实验结果表明,G-YOLOv5s的mAP达到71.7%,模型体积为9.09 MB,浮点型计算量为9.8 G FLOPs,与改进前的YOLOv5s网络相比,模型体积压缩了34.8%,计算量减少了41.3%,精度仅下降1.3%,方便部署在资源有限的设备中使用.展开更多
文摘目的鉴于传统的半监督光刻热点检测方法逐渐无法满足集成电路制造对检测精度的要求,且难以解决因数据集不平衡引起的精度损失问题,提出一种新的半监督光刻热点检测模型GSSL。方法在该模型中,将卷积注意力模块(Convolutional Block Attention Module,CBAM)引入到Ghost模块的线性变化中,设计了Ghost_CBAM模块;将该模块与压缩激励网络(Squeeze-and-Excitation,SE)结合设计了GhostNeck模块,实现特征图先降维再升维,建立各个通道之间的关联性;再通过GhostNeck构建整个光刻热点检测模型GSSL,实现逐步引入无标记数据进入训练的半监督学习方式;通过集成数据增强方法对数据集中的热点版图进行数据增强,缓解数据不平衡问题;并应用加权交叉熵损失函数,进一步提升模型对于热点类别的关注度。结果在ICCAD(The International Conference on Computer-Aided Design)2012竞赛基准数据集上进行评估,在标记数据占比为10%~50%的情况下预测热点的平均准确率为91.73%,平均误报为680。结论与其他传统方法相比,GSSL可以有效应对数据集不平衡的问题,提升光刻热点检测精度的同时,显著降低了误报率,在光刻热点检测上具有一定的应用价值。
基金Project supported by the National Key Research and Development Program of China(Grant Nos.2022YFA1603601,2021YFF0601203,and 2021YFA1600703)。
文摘The unique advantage of x-ray ghost imaging(XGI)is its potential in low dose radiology.One of the practical ways to reduce the radiation exposure is to reduce the measurements while remaining sufficient image quality.Synthetic aperture x-ray ghost imaging(SAXGI)is invented to achieve megapixel XGI with limited measurements,which is expected to implement XGI simultaneously with large field of view and low radiation exposure.In this paper,we experimentally investigate the effect of measurements reduction on the spatial resolution and image quality of SAXGI with standard sample and biomedical specimen.The results with a resolution chart demonstrated that at 360 measurements,SAXGI successfully retrieved the sample image of 1960×1960 pixels with spatial resolution of 4μm.With measurement reduction,the spatial resolution deteriorates but the sparser structures are still discernable.Even with measurements reduced to 10,a spatial resolution of 10μm can still be achieved by SAXGI.A biomedical sample of a fish specimen is employed to evaluate the method and the fish image of 2000×1000 pixels with an SSIM of 0.962 is reconstructed by SAXGI with 770measurements,corresponding to an accumulative exposure reduction of more than 2 times.With the measurements reduced to 10 which corresponds to 1/160 of the accumulative radiation exposure for conventional radiology,bulky structure like the fish skeleton can still be definitely discerned and the SSIM for the reconstructed image still retained 0.9179.Results of this paper demonstrate that measurements reduction is practicable for the radiation exposure reduction of the sample,which implicates that SAXGI with limited measurements is an efficient solution for low dose radiology.
文摘Accurate vehicle detection is essential for autonomous driving,traffic monitoring,and intelligent transportation systems.This paper presents an enhanced YOLOv8n model that incorporates the Ghost Module,Convolutional Block Attention Module(CBAM),and Deformable Convolutional Networks v2(DCNv2).The Ghost Module streamlines feature generation to reduce redundancy,CBAM applies channel and spatial attention to improve feature focus,and DCNv2 enables adaptability to geometric variations in vehicle shapes.These components work together to improve both accuracy and computational efficiency.Evaluated on the KITTI dataset,the proposed model achieves 95.4%mAP@0.5—an 8.97% gain over standard YOLOv8n—along with 96.2% precision,93.7% recall,and a 94.93%F1-score.Comparative analysis with seven state-of-the-art detectors demonstrates consistent superiority in key performance metrics.An ablation study is also conducted to quantify the individual and combined contributions of GhostModule,CBAM,and DCNv2,highlighting their effectiveness in improving detection performance.By addressing feature redundancy,attention refinement,and spatial adaptability,the proposed model offers a robust and scalable solution for vehicle detection across diverse traffic scenarios.
文摘为了降低服装目标检测模型的参数量和浮点型计算量,提出一种改进的轻量级服装目标检测模型——GYOLOv5s.首先使用Ghost卷积重构YOLOv5s的主干网络;然后使用DeepFashion2数据集中的部分数据进行模型训练和验证;最后将训练好的模型用于服装图像的目标检测.实验结果表明,G-YOLOv5s的mAP达到71.7%,模型体积为9.09 MB,浮点型计算量为9.8 G FLOPs,与改进前的YOLOv5s网络相比,模型体积压缩了34.8%,计算量减少了41.3%,精度仅下降1.3%,方便部署在资源有限的设备中使用.
