在目标检测领域中,基于交并比(intersection over union, IoU)的系列损失函数存在一定的局限性,使得边界框回归的精度和稳定性有待进一步提升。为此提出了一种基于非线性高斯平方距离的边界框回归损失函数。首先综合考虑了边界框中重叠...在目标检测领域中,基于交并比(intersection over union, IoU)的系列损失函数存在一定的局限性,使得边界框回归的精度和稳定性有待进一步提升。为此提出了一种基于非线性高斯平方距离的边界框回归损失函数。首先综合考虑了边界框中重叠性、中心点距离和长宽比3个因素,将边界框建模为高斯分布;然后提出一种高斯平方距离来衡量概率分布之间的差距;最后设计了符合优化趋势的非线性函数,将高斯平方距离转化为有利于神经网络学习的损失函数。实验结果表明,与IoU损失相比,所提方法在掩膜区域卷积神经网络、一阶全卷积目标检测器和自适应特征选择目标检测器上的平均精度均值分别提高了0.3%、1.1%和2.3%,证明了该方法能有效提升目标检测的性能,同时有利于高精度边界框的回归。展开更多
Ambient backscatter communications(AmBC)is a new ultra-low-power communication paradigm,which holds great promise for enabling energy self-sustainability(ESS)to massive data-intensive Internet of Everything(IoE)device...Ambient backscatter communications(AmBC)is a new ultra-low-power communication paradigm,which holds great promise for enabling energy self-sustainability(ESS)to massive data-intensive Internet of Everything(IoE)devices in 6G.Recent advances improve throughput and reliability by adopting multiple-antenna techniques in conventional backscatter communications(CoBC),but they cannot be directly applied to AmBC devices for high spectral and energy efficiency due to the unknown RF source and minimalist design in backscatter tag.To fill this gap,we propose SM-backscatter,an AmBC-compatible system that greatly improves spectral efficiency while maintaining ultra-low-power consumption.Specifically,the SM-backscatter consists of two novel components:i)a multiple-antenna backscatter tag that adopts spatial modulation(SM),and ii)a joint detection algorithm that detects both backscatter and source signals.To this end,we theoretically obtain an optimal detector and propose two suboptimal detectors with low complexity.Subsequently,we derive the BERs of both the backscatter and source signals to analyze the communication performance by introducing a two-step algorithm.Our simulation results verify the correctness of the theoretical analysis and indicate that our system can significantly outperform existing solutions.展开更多
文摘在目标检测领域中,基于交并比(intersection over union, IoU)的系列损失函数存在一定的局限性,使得边界框回归的精度和稳定性有待进一步提升。为此提出了一种基于非线性高斯平方距离的边界框回归损失函数。首先综合考虑了边界框中重叠性、中心点距离和长宽比3个因素,将边界框建模为高斯分布;然后提出一种高斯平方距离来衡量概率分布之间的差距;最后设计了符合优化趋势的非线性函数,将高斯平方距离转化为有利于神经网络学习的损失函数。实验结果表明,与IoU损失相比,所提方法在掩膜区域卷积神经网络、一阶全卷积目标检测器和自适应特征选择目标检测器上的平均精度均值分别提高了0.3%、1.1%和2.3%,证明了该方法能有效提升目标检测的性能,同时有利于高精度边界框的回归。
基金This work was supported in part by the National Key R&D Program of China with Grant number 2019YFB1803400Young Elite Scientists Sponsorship Program by CAST under Grant number 2018QNRC001National Science Foundation of China with Grant number 91738202,62071194.
文摘Ambient backscatter communications(AmBC)is a new ultra-low-power communication paradigm,which holds great promise for enabling energy self-sustainability(ESS)to massive data-intensive Internet of Everything(IoE)devices in 6G.Recent advances improve throughput and reliability by adopting multiple-antenna techniques in conventional backscatter communications(CoBC),but they cannot be directly applied to AmBC devices for high spectral and energy efficiency due to the unknown RF source and minimalist design in backscatter tag.To fill this gap,we propose SM-backscatter,an AmBC-compatible system that greatly improves spectral efficiency while maintaining ultra-low-power consumption.Specifically,the SM-backscatter consists of two novel components:i)a multiple-antenna backscatter tag that adopts spatial modulation(SM),and ii)a joint detection algorithm that detects both backscatter and source signals.To this end,we theoretically obtain an optimal detector and propose two suboptimal detectors with low complexity.Subsequently,we derive the BERs of both the backscatter and source signals to analyze the communication performance by introducing a two-step algorithm.Our simulation results verify the correctness of the theoretical analysis and indicate that our system can significantly outperform existing solutions.
