基于YOLO改进算法的安全帽和口罩佩戴自动同时检测

针对工地、危险区域等场景需要实现同时佩戴安全帽与口罩的自动检测问题,提出一种改进的YOLOv3算法以提高同时检测安全帽和口罩佩戴的准确率。首先,对网络模型中的聚类算法进行优化,使用加权核K-means聚类算法对训练数据集聚类分析,选取更适合小目标检测的Anchor Box,以提高检测的平均精度和速度;然后,优化YOLO网络内部的Darknet特征网络层,将4倍降采样提取的特征图进行2倍上采样,再与2倍降采样进行卷积融合,与4倍降采样、8倍降采样以及16倍降采样一同输送到后续网络中,来达到降低小目标的漏检概率。实验结果表明:改进后的算法同时检测安全帽和口罩佩戴的平均准确率比原算法提高了11.3%。     

Calculation and Test of Hitting Force of Firing Bullet on the Bulletproof Helmet

1.INTRODUCTIONThebulletproofhelmetismadebyultrahighstrengthsteelplatethroughhotstamping.Theforceanddeformationarecomplexatfiringbullet.Itrelatestoplasticitydynamics,hotelasticity-plasticitymechanicsandcompoundmaterialdeformationmechanics.Therehasnotaccuratelytheorycomputation.Inthispapertaccordingtothetestdevicethatthebulletproofhelmetcoversthewoodheadmodel,wegettwosystemswhichcalculatethehittingforceatfiringbulletusingmomenttheoremandenergybalanceprinciple,andwealsocomparewiththetestresu…     

Development and application research of light weight heat treated B-grade bulletproof steel

The light weight heat treated B-grade bulletproof steel was developed through composition design and optimization based on multiplex alloying, multiplex micro-alloying design ideas and complex phase structure strengthening theory. The puzzle how to avoid the quenching deformation problem of super high strength thin sheet was solved through heat treatment in a die with a set of cooling system. Such B-grade bulletproof steel plate has fine tempered lath martensite structure. The shooting and certification test results showed that the shoot resistance of B- grade bulletproof steel plate can meet the protection demand of Protection specification for cash carrying vehicles （GA 164--2005）. In comparison with B-grade bulletproof steel plate made by one of the companies in Sweden, the weight of the developed B-grade bulletproof steel plate can be decreased by 8 % under the same shoot resistance condition. It will be meaningful for cash truck and anti-hijacking vehicle to realize light weight, energy conservation and emission reduction.     

方差分量估计在三维七参数坐标转换中的应用

将方差分量估计应用于三维七参数坐标转换模型参数求解,并通过实例计算与传统的解算方法进行精度比较,实验结果表明,将方差分量估计用于三维七参数坐标转换模型参数求解,可有效提高坐标转换的精度.     

复杂交通环境下二轮机动车乘员头盔检测算法

针对现有二轮机动车乘员头盔检测算法在目标密集分布、随机遮挡等情况下效果较差且难以在边缘设备上应用的问题,制作了具有针对性的数据集,对比现有模型后,以YOLOv7为参考提出一种复杂交通环境下二轮机动车乘员头盔检测算法.首先,采用EfficientNet-B3作为主干网络,可提高特征提取能力且更为轻量化;其次,将增大感受野模块(RFB)引入特征融合结构中,以增大模型感受野,提升小目标头盔检测能力;最后,在检测头嵌入SimAM机制,在不增加参数的前提下提高算法精度.结果表明：相较于YOLOv7,文中算法的准确率、召回率和平均准确率分别提高了2.84%,2.26%和3.26%,参数量和运算量分别为YOLOv7的33.1%,23.5%,可实现当前主流模型算法的最佳检测性能和效率;在NVIDIA Jetson Nano开发板上的处理速度达到47.58 F·s^-1,可满足边缘设备部署需求.     

轻量化YOLOv4的安全帽佩戴检测方法

为了平衡深度学习网络在网络深度、模型效果和易用性之间的关系,针对安全帽佩戴检测,提出一种基于轻量化模块的改进YOLOv4网络。在主干网络中,采用深度可分卷积和压缩激励机制所组成的模块替换残差块,在不进行内部堆叠的情况下,减少网络的深度和参数量;在聚合网络中,采用反向注意力机制模块增加目标物体的显著信息,使网络获取的特征包含更丰富的细节信息。在公开的安全帽佩戴图像数据集上进行的实验表明,提出方法的模型大小为106.4 M,相比原网络减少了149.9 M,网络检测的平均精确度均值比原网络有所提升,验证了轻量化目标检测方法及其在安全帽检测应用中的有效性。     

计及弹型对防弹钢板抗侵彻能力的数值模拟

采用ANSYS/LS-DYNA有限元分析软件对B900FD-1型防弹钢板抗侵彻能力进行数值模拟。对比数值模拟结果与试验结果,表明采用塑性随动强化模型和Johnson-Cook模型能够较好地模拟子弹对金属类延性材料的侵彻能力。进一步采用数值分析方法分析了弹型、弹速及弹的长径比对钢板的侵彻能力的影响。     

Numerical analysis on the shoot resistance of heat treated light weight B-grade bulletproof steel plates

Bulletproof steel plates are widely used for the safety of special vehicles. This paper mainly researches on the shoot resistance of heat treated light weight B-grade bulletproof steel plates through numerical analysis. Based on the flow behavior of bulletproof steel plates and bullet at various high strain rates, finite element （FE） model has been set up using ANSYS/LS-DYNA software. The simulation results are compared with the shooting results, which show a good consistency and a high reliability. Therefore, the simulation results are efficient approaches and strategies to decide and select the mechanical property and thickness of bulletproof steel plates, saving a lot of work and the cost of experiments.     

一种6URHS头颈部外骨骼系统动力学影响因素的分析

头颈部外骨骼系统是一种为虚拟现实交互设备—头盔显示器提供主动控制的系统。而该系统执行机构—6URHS平台的动力学特性是否能满足头部运动的要求,将会直接影响到整个系统的使用效果。针对执行机构结构的特殊性,建立了动力学模型,分析了影响外骨骼系统动力学特性的主要影响因素,设计了动平台质量等6因素3水平共18组正交试验,利用极差分析法对影响因素进行了定量比较分析。结果表明外筒惯量对6URHS并联平台的动力学特性影响最为明显,而外筒质量与外筒惯量的极差和比小于1/8,其对6URHS并联平台的影响最弱。     

计算冷挤压成型钢盔挤压力的滑移线法

阐述钢制头盔冷挤压成型的塑性变形和钢制椭圆薄板冷挤压塑性成型过程的塑性流线,并由此建立滑移线分布,用滑移线计算钢制头盔的挤压力。最后给出应用算例。