针对雨雾等复杂天气下无人机图像质量下降导致目标检测效果不佳的问题,提出基于上下文引导和提示学习的目标检测算法CGP-YOLO(context-guided and prompt-based YOLOv8)。构建一个多任务联合学习的检测网络,通过双分支结构达到平衡图像...针对雨雾等复杂天气下无人机图像质量下降导致目标检测效果不佳的问题,提出基于上下文引导和提示学习的目标检测算法CGP-YOLO(context-guided and prompt-based YOLOv8)。构建一个多任务联合学习的检测网络,通过双分支结构达到平衡图像检测和恢复的任务。提出基于提示学习的跨层注意力加权图像去噪分支,指导网络利用退化提示重构清晰的图像;模型主干设计基于上下文的残差采样模块,集成卷积注意力机制,综合目标的局部和全局信息;采用可分离大核多尺度特征提取模块,处理网络多尺度特征;引入小目标的专用检测头,增强小目标的检测精度。实验结果表明,在参数量仅为基线模型60%的情况下,该模型的检测精度提高了2.4个百分点,平均精度(mAP)提高了2.04个百分点,模型检测效果优于其他经典模型,具备卓越的性能。展开更多
Electrostatic levitation technique and molecular dynamics simulation were performed to investigate the thermophysical properties,liquid structure and crystal growth dependence on undercooling of Ti_(85)Ni_(10)Al_(5) a...Electrostatic levitation technique and molecular dynamics simulation were performed to investigate the thermophysical properties,liquid structure and crystal growth dependence on undercooling of Ti_(85)Ni_(10)Al_(5) alloy.The liquid Ti_(85)Ni_(10)Al_(5) alloy was substantially undercooled up to 335 K(0.18T_(L)).As undercooling increased,the potential energy of the liquid alloy decreased and the alloy entered into a high metastable state.At this state,the atoms tended to bond with each other and the clusters were inclined to convert into high-coordinated clusters,as confirmed by the fraction of the high-coordinated clusters variation.The enlarged clusters and enhanced local structure stability contributed to the increase of the thermophysical parameters and crystal growth velocity,and eventually dendrite refinement.The density,the specific heat and the surface tension of liquid alloy exhibited a linear relation with temperature and the shear viscosity of liquid alloy showed exponential variation which showed good agreement with the calculation results by molecular dynamics simulation.The growth velocity first increased slowly and then dramatically once the undercooling exceeded the threshold.The achieved maximum crystal growth velocity was 12.4 m s^(−1) and it was up to 326 times of the value at 94 K undercooling.展开更多
文摘针对雨雾等复杂天气下无人机图像质量下降导致目标检测效果不佳的问题,提出基于上下文引导和提示学习的目标检测算法CGP-YOLO(context-guided and prompt-based YOLOv8)。构建一个多任务联合学习的检测网络,通过双分支结构达到平衡图像检测和恢复的任务。提出基于提示学习的跨层注意力加权图像去噪分支,指导网络利用退化提示重构清晰的图像;模型主干设计基于上下文的残差采样模块,集成卷积注意力机制,综合目标的局部和全局信息;采用可分离大核多尺度特征提取模块,处理网络多尺度特征;引入小目标的专用检测头,增强小目标的检测精度。实验结果表明,在参数量仅为基线模型60%的情况下,该模型的检测精度提高了2.4个百分点,平均精度(mAP)提高了2.04个百分点,模型检测效果优于其他经典模型,具备卓越的性能。
基金supported by the National Natural Science Foundation of China(Grant Nos.U1806219,52073232,52088101 and U1660108)the Science Fund for Distinguished Young Scholars of Shaanxi Province(Grant No.2020JC-11)the Science Fund for Scientific and Technological Innovation Team of Shaanxi Province(Grant No.2021TD-14)。
文摘Electrostatic levitation technique and molecular dynamics simulation were performed to investigate the thermophysical properties,liquid structure and crystal growth dependence on undercooling of Ti_(85)Ni_(10)Al_(5) alloy.The liquid Ti_(85)Ni_(10)Al_(5) alloy was substantially undercooled up to 335 K(0.18T_(L)).As undercooling increased,the potential energy of the liquid alloy decreased and the alloy entered into a high metastable state.At this state,the atoms tended to bond with each other and the clusters were inclined to convert into high-coordinated clusters,as confirmed by the fraction of the high-coordinated clusters variation.The enlarged clusters and enhanced local structure stability contributed to the increase of the thermophysical parameters and crystal growth velocity,and eventually dendrite refinement.The density,the specific heat and the surface tension of liquid alloy exhibited a linear relation with temperature and the shear viscosity of liquid alloy showed exponential variation which showed good agreement with the calculation results by molecular dynamics simulation.The growth velocity first increased slowly and then dramatically once the undercooling exceeded the threshold.The achieved maximum crystal growth velocity was 12.4 m s^(−1) and it was up to 326 times of the value at 94 K undercooling.
