针对无人机视角下的目标存在多尺度、目标小、被遮挡与背景复杂等问题,提出了一种基于动态样本注意力尺度序列的YOLOv8改进算法BDAD-YOLO。通过引入BiFormer的思想来改造原模型骨干结构,提高模型对关键信息的关注度,更好地保留目标细粒...针对无人机视角下的目标存在多尺度、目标小、被遮挡与背景复杂等问题,提出了一种基于动态样本注意力尺度序列的YOLOv8改进算法BDAD-YOLO。通过引入BiFormer的思想来改造原模型骨干结构,提高模型对关键信息的关注度,更好地保留目标细粒度细节信息。由于目标存在大小、位置等多变性,传统卷积并不能很好地处理这一情况,因此基于DCN(deformable convolutional network)的思想,设计了一种可以增强对小目标特征提取的C2_DCf模块,从而进一步提高颈部网络中小目标层对特征信息的融合。提出一种基于动态样本的注意力尺度序列融合框架AFD(attention-scale sequence fusion framework based on dynamic samples),使用轻量化动态点采样并通过融合不同尺度的特征图来增强网络提取多尺度信息的能力。使用WIoU损失函数,改善小目标低质量数据对梯度的不利影响,以加快网络收敛速度。实验结果表明,在VisDrone数据集中的val集与test集上平均精度(mAP@0.5)分别提升了4.6个百分点、3.7个百分点,在DOTA数据集上平均精度(mAP@0.5)提升了2.4个百分点,证明了改进算法的有效性和普适性。展开更多
During the storage of composite propellants,the migration of plasticizers and other unbonded additives at the interfaces of liner adhesives has garnered significant attention in understanding liner failure mechanisms,...During the storage of composite propellants,the migration of plasticizers and other unbonded additives at the interfaces of liner adhesives has garnered significant attention in understanding liner failure mechanisms,aging processes,and safety performance.However,there is currently no non-destructive and quantitative detection method for migration of plasticizers in propellant liner.In this study,we developed a HTPB sensing liner by incorporating conductive fillers-namely carbon black(CB),carbon nanotubes(CNTs),and graphene nanoplatelets(GNP)-into the HTPB matrix.The synergistic interaction between CNTs and GNP facilitates the formation of a tunneling conductive network that imparts electrical conductivity to the HTPB liner.To elucidate the functional relationship between conductivity and nitroglycerin(NG)migration,we applied the HTPB sensing liner onto double base propellant surfaces and measured both the conductivity of the sensing layer and NG migration during a 71℃accelerated aging experiment.The results shows that when CNTs/GNP content reaches 3wt%,there is an exponential correlation between conductivity and NG migration with a fitting degree of 0.9652;the average response sensitivity of DR/R0 relative to NG migration is calculated as 41.69,with an average deviation of merely 5.67%between NG migrations derived from conductivity fittings compared to those obtained via TGA testing results.Overall,this sensing liner exhibits excellent capabilities for detecting NG migration nondestructively and quantitatively while offering a novel approach for assessing interfacial component migrations as well as debonding defects in propellantsda promising avenue for future self-monitoring strategies regarding propellant integrity.展开更多
文摘针对无人机视角下的目标存在多尺度、目标小、被遮挡与背景复杂等问题,提出了一种基于动态样本注意力尺度序列的YOLOv8改进算法BDAD-YOLO。通过引入BiFormer的思想来改造原模型骨干结构,提高模型对关键信息的关注度,更好地保留目标细粒度细节信息。由于目标存在大小、位置等多变性,传统卷积并不能很好地处理这一情况,因此基于DCN(deformable convolutional network)的思想,设计了一种可以增强对小目标特征提取的C2_DCf模块,从而进一步提高颈部网络中小目标层对特征信息的融合。提出一种基于动态样本的注意力尺度序列融合框架AFD(attention-scale sequence fusion framework based on dynamic samples),使用轻量化动态点采样并通过融合不同尺度的特征图来增强网络提取多尺度信息的能力。使用WIoU损失函数,改善小目标低质量数据对梯度的不利影响,以加快网络收敛速度。实验结果表明,在VisDrone数据集中的val集与test集上平均精度(mAP@0.5)分别提升了4.6个百分点、3.7个百分点,在DOTA数据集上平均精度(mAP@0.5)提升了2.4个百分点,证明了改进算法的有效性和普适性。
基金funded by Zhijian Laboratory Open Fund,Rocket Force University of Engineering(Grant No.2023-ZJSYS-KF01-03).
文摘During the storage of composite propellants,the migration of plasticizers and other unbonded additives at the interfaces of liner adhesives has garnered significant attention in understanding liner failure mechanisms,aging processes,and safety performance.However,there is currently no non-destructive and quantitative detection method for migration of plasticizers in propellant liner.In this study,we developed a HTPB sensing liner by incorporating conductive fillers-namely carbon black(CB),carbon nanotubes(CNTs),and graphene nanoplatelets(GNP)-into the HTPB matrix.The synergistic interaction between CNTs and GNP facilitates the formation of a tunneling conductive network that imparts electrical conductivity to the HTPB liner.To elucidate the functional relationship between conductivity and nitroglycerin(NG)migration,we applied the HTPB sensing liner onto double base propellant surfaces and measured both the conductivity of the sensing layer and NG migration during a 71℃accelerated aging experiment.The results shows that when CNTs/GNP content reaches 3wt%,there is an exponential correlation between conductivity and NG migration with a fitting degree of 0.9652;the average response sensitivity of DR/R0 relative to NG migration is calculated as 41.69,with an average deviation of merely 5.67%between NG migrations derived from conductivity fittings compared to those obtained via TGA testing results.Overall,this sensing liner exhibits excellent capabilities for detecting NG migration nondestructively and quantitatively while offering a novel approach for assessing interfacial component migrations as well as debonding defects in propellantsda promising avenue for future self-monitoring strategies regarding propellant integrity.
