Autonomous underwater vehicles （AUVs） navigating on the sea surface are usually required to complete the communication tasks in complex sea conditions. The movement forms and flow field characteristics of a multi-mo...Autonomous underwater vehicles （AUVs） navigating on the sea surface are usually required to complete the communication tasks in complex sea conditions. The movement forms and flow field characteristics of a multi-moving state AUV navigating in head sea at high speed were studied. The mathematical model on longitudinal motion of the high-speed AUV in head sea was established with considering the hydrodynamic lift based on strip theory, which was solved to get the heave and pitch of the AUV by Gaussian elimination method. Based on this, computational fluid dynamics （CFD） method was used to establish the mathematical model of the unsteady viscous flow around the AUV with considering free surface effort by using the Reynolds-averaged Navier-Stokes （RANS） equations, shear-stress transport （SST） k-w model and volume of fluid （VOF） model. The three-dimensional numerical wave in the computational field was realized through defining the unsteady inlet boundary condition. The motion forms of the AUV navigating in head sea at high speed were carried out by the program source code of user-defined function （UDF） based on dynamic mesh. The hydrodynamic parameters of the AUV such as drag, lift, pitch torque, velocity, pressure and wave profile were got, which reflect well the real ambient flow field of the AUV navigating in head sea at high speed. The computational wave profile agrees well with the experimental phenomenon of a wave-piercing surface vehicle. The force law of the AUV under the impacts of waves was analyzed qualitatively and quantitatively, which provides an effective theoretical guidance and technical support for the dynamics research and shape design of the AUV in real complex environnaent.展开更多
针对暗环境动态特征轮廓模糊、盲区遮挡情况,高效准确地检测跟踪动态目标特征,对灾害救援、搜寻跟踪具有实际意义。为实现暗环境下模糊轮廓特征的有效检测跟踪,提出一种时空关联机制的红外目标实时检测深度学习网络(Spatial Local Dynam...针对暗环境动态特征轮廓模糊、盲区遮挡情况,高效准确地检测跟踪动态目标特征,对灾害救援、搜寻跟踪具有实际意义。为实现暗环境下模糊轮廓特征的有效检测跟踪,提出一种时空关联机制的红外目标实时检测深度学习网络(Spatial Local Dynamic You Only Look Once Version 8,SLD-YOLOv8),设计非局部自适应Non-local模块和空间通道卷积关联模块,对原YOLOv8网络的瓶颈层Bottleneck CSP进行优化。为有效提取深层空间多尺度表征信息,增加用于小目标检测的160×160检测层和动态检测头,较好地提升暗环境中目标跟踪的边界回归性能,并实时有效地推理出目标特征的相对深度位置信息。实验结果表明,改进后的红外目标检测算法对暗环境下的动态特征检测具有较好的鲁棒性和准确性,其平均精度评估指标mAP_0.5和mAP_0.5:0.95比原模型提高了5.6%和4.5%,证明了新算法对暗环境目标跟踪的有效性。展开更多
现有的烟火检测方法主要依赖员工现场巡视,效率低且实时性差,因此,提出一种基于YOLOv5s的复杂场景下的高效烟火检测算法YOLOv5s-MRD(YOLOv5s-MPDIoU-RevCol-Dyhead)。首先,采用MPDIoU(Maximized Position-Dependent Intersection over U...现有的烟火检测方法主要依赖员工现场巡视,效率低且实时性差,因此,提出一种基于YOLOv5s的复杂场景下的高效烟火检测算法YOLOv5s-MRD(YOLOv5s-MPDIoU-RevCol-Dyhead)。首先,采用MPDIoU(Maximized Position-Dependent Intersection over Union)方法改进边框损失函数,以适应重叠或非重叠的边界框回归(BBR),从而提高BBR的准确性和效率;其次,利用可逆柱状结构RevCol(Reversible Column)网络模型思想重构YOLOv5s模型的主干网络,使它具有多柱状网络架构,并在模型的不同层之间加入可逆链接,从而最大限度地保持特征信息以提高网络的特征提取能力;最后,引入Dynamic head检测头,以统一尺度感知、空间感知和任务感知,从而在不额外增加计算开销的条件下显著提高目标检测头的准确性和有效性。实验结果表明:在DFS(Data of Fire and Smoke)数据集上,与原始YOLOv5s算法相比,所提算法的平均精度均值(mAP@0.5)提升了9.3%,预测准确率提升了6.6%,召回率提升了13.8%。可见,所提算法能满足当前烟火检测应用场景的要求。展开更多
车辆检测是智能交通系统和自动驾驶的重要组成部分。然而,实际交通场景中存在许多不确定因素,导致车辆检测模型的准确率低实时性差。为了解决这个问题,提出了一种快速准确的车辆检测算法——YOLOv8-DEL。使用DGCST(dynamic group convol...车辆检测是智能交通系统和自动驾驶的重要组成部分。然而,实际交通场景中存在许多不确定因素,导致车辆检测模型的准确率低实时性差。为了解决这个问题,提出了一种快速准确的车辆检测算法——YOLOv8-DEL。使用DGCST(dynamic group convolution shuffle transformer)模块代替C2f模块来重构主干网络,以增强特征提取能力并使网络更轻量;添加的P2检测层能使模型更敏锐地定位和检测小目标,同时采用Efficient RepGFPN进行多尺度特征融合,以丰富特征信息并提高模型的特征表达能力;通过结合GroupNorm和共享卷积的优点,设计了一种轻量型共享卷积检测头,在保持精度的前提下,有效减少参数量并提升检测速度。与YOLOv8相比,提出的YOLOv8-DEL在BDD100K数据集和KITTI数据集上,mAP@0.5分别提高了4.8个百分点和1.2个百分点,具有实时检测速度(208.6 FPS和216.4 FPS),在检测精度和速度方面实现了更有利的折中。展开更多
基金Project(2006AA09Z235)supported by the National High Technology Research and Development Program of ChinaProject(CX2009B003)supported by Hunan Provincial Innovation Foundation For Postgraduate,China
文摘Autonomous underwater vehicles （AUVs） navigating on the sea surface are usually required to complete the communication tasks in complex sea conditions. The movement forms and flow field characteristics of a multi-moving state AUV navigating in head sea at high speed were studied. The mathematical model on longitudinal motion of the high-speed AUV in head sea was established with considering the hydrodynamic lift based on strip theory, which was solved to get the heave and pitch of the AUV by Gaussian elimination method. Based on this, computational fluid dynamics （CFD） method was used to establish the mathematical model of the unsteady viscous flow around the AUV with considering free surface effort by using the Reynolds-averaged Navier-Stokes （RANS） equations, shear-stress transport （SST） k-w model and volume of fluid （VOF） model. The three-dimensional numerical wave in the computational field was realized through defining the unsteady inlet boundary condition. The motion forms of the AUV navigating in head sea at high speed were carried out by the program source code of user-defined function （UDF） based on dynamic mesh. The hydrodynamic parameters of the AUV such as drag, lift, pitch torque, velocity, pressure and wave profile were got, which reflect well the real ambient flow field of the AUV navigating in head sea at high speed. The computational wave profile agrees well with the experimental phenomenon of a wave-piercing surface vehicle. The force law of the AUV under the impacts of waves was analyzed qualitatively and quantitatively, which provides an effective theoretical guidance and technical support for the dynamics research and shape design of the AUV in real complex environnaent.
文摘针对暗环境动态特征轮廓模糊、盲区遮挡情况,高效准确地检测跟踪动态目标特征,对灾害救援、搜寻跟踪具有实际意义。为实现暗环境下模糊轮廓特征的有效检测跟踪,提出一种时空关联机制的红外目标实时检测深度学习网络(Spatial Local Dynamic You Only Look Once Version 8,SLD-YOLOv8),设计非局部自适应Non-local模块和空间通道卷积关联模块,对原YOLOv8网络的瓶颈层Bottleneck CSP进行优化。为有效提取深层空间多尺度表征信息,增加用于小目标检测的160×160检测层和动态检测头,较好地提升暗环境中目标跟踪的边界回归性能,并实时有效地推理出目标特征的相对深度位置信息。实验结果表明,改进后的红外目标检测算法对暗环境下的动态特征检测具有较好的鲁棒性和准确性,其平均精度评估指标mAP_0.5和mAP_0.5:0.95比原模型提高了5.6%和4.5%,证明了新算法对暗环境目标跟踪的有效性。
文摘现有的烟火检测方法主要依赖员工现场巡视,效率低且实时性差,因此,提出一种基于YOLOv5s的复杂场景下的高效烟火检测算法YOLOv5s-MRD(YOLOv5s-MPDIoU-RevCol-Dyhead)。首先,采用MPDIoU(Maximized Position-Dependent Intersection over Union)方法改进边框损失函数,以适应重叠或非重叠的边界框回归(BBR),从而提高BBR的准确性和效率;其次,利用可逆柱状结构RevCol(Reversible Column)网络模型思想重构YOLOv5s模型的主干网络,使它具有多柱状网络架构,并在模型的不同层之间加入可逆链接,从而最大限度地保持特征信息以提高网络的特征提取能力;最后,引入Dynamic head检测头,以统一尺度感知、空间感知和任务感知,从而在不额外增加计算开销的条件下显著提高目标检测头的准确性和有效性。实验结果表明:在DFS(Data of Fire and Smoke)数据集上,与原始YOLOv5s算法相比,所提算法的平均精度均值(mAP@0.5)提升了9.3%,预测准确率提升了6.6%,召回率提升了13.8%。可见,所提算法能满足当前烟火检测应用场景的要求。
