基于多尺度特征选择性融合的遥感图像检测算法北大核心CSCD

遥感图像的检测在监察自然环境、军事、国土安全等方面具有极其广阔的应用前景,而遥感图像具有背景复杂、目标面积小、特征提取困难等缺点,进行检测时容易产生小目标漏检问题。本文提出一种基于多尺度特征选择性融合的遥感图像检测算法。所提算法采用改进的Resnet50作为主干网络,将Resnet50第一个卷积替换成动态卷积,并将其ConvBlock模块中的卷积替换成金字塔卷积,提高特征提取能力。同时,为了避免遗漏底层信息,在动态卷积层后加入所提有效空间通道注意力机制模块。最后,选取基于上下文信息的不同尺度特征进行融合,提高了模型对目标物体的定位能力。实验结果表明,本文算法在保证速度的同时提高了对遥感图像的检测精度,在遥感图像公开数据集RSOD和NWPUVHR-10上平均精度均值(mean average precision,mAP)分别达到91.88%和90.23%,检测速度达到33 FPS。     

Face anti-spoofing based on multi-modal and multi-scale features fusion

Face anti-spoofing is used to assist face recognition system to judge whether the detected face is real face or fake face. In the traditional face anti-spoofing methods, features extracted by hand are used to describe the difference between living face and fraudulent face. But these handmade features do not apply to different variations in an unconstrained environment. The convolutional neural network (CNN) for face deceptions achieves considerable results. However, most existing neural network-based methods simply use neural networks to extract single-scale features from single-modal data, while ignoring multi-scale and multi-modal information. To address this problem, a novel face anti-spoofing method based on multi-modal and multi-scale features fusion ( MMFF) is proposed. Specifically, first residual network ( Resnet )-34 is adopted to extract features of different scales from each modality, then these features of different scales are fused by feature pyramid network (FPN), finally squeeze-and-excitation fusion ( SEF) module and self-attention network ( SAN) are combined to fuse features from different modalities for classification. Experiments on the CASIA-SURF dataset show that the new method based on MMFF achieves better performance compared with most existing methods.     

基于强化学习的列车驾驶曲线节能优化算法

在保证安全和准时性的前提下,自动化列车运行可以有效减少列车耗能。为了灵活应对列车运行种的动态变化,提出了一种基于强化学习的方法,可以优化列车控制策略且不采用之前关于列车动力学的知识和设计的列车速度曲线。这个优化模型将列车节能作为目标,把准点到达、列车限速、停车位置作为限制条件。大量的列车运行经验可以被用来训练深度神经网络直到得到最优化行为价值函数,通过对训练过的神经网络输入状态,可以准确输出每个行为的价值,然后再根据行为价值的大小来选择最优的驾驶策略。     

基于序列关联的参差信号分选算法

针对复杂环境下常规直方图信号分选算法对于参差信号分选能力不佳的问题,该文提出一种基于脉冲间隔与单个脉冲关联的直方图算法.该算法根据脉冲间隔与单个脉冲的对应关系建立了脉冲间隔分布矩阵(PIDM),然后通过对PIDM行列的累加计算,得到一种新的直方图,该直方图可避免传统脉冲重复间隔(PRI)变换算法在分选参差信号时对于参差信号帧周期过多抑制的缺陷,且能够通过PIDM对辐射源脉冲串进行序列提取,进而得到参差子序列的周期值.仿真分析结果表明,在不增加计算复杂度的情况下,该算法对存在多部参差辐射源和固定重频辐射源的混合场景仍可保持良好的分选效果.     

一种人与人和机器到机器共存下能效最大化的上行用户分配算法

针对5G超密集异构蜂窝网络中人与人(H2H)和机器到机器(M2M)共存场景下有服务质量(QoS)保障和负载均衡的上行用户分配问题,该文提出一种基于匹配理论的用户分配算法.该算法在用户分配过程中同时考虑不同类型节点的接入控制策略,在最大化网络能效的同时,实现节点QoS保障.仿真结果表明,与其他算法相比,该算法不仅能够在能效、负载均衡以及QoS保障方面获得更好的性能,并且能获得与穷举法相近的性能.此外,所提算法的收敛速度很快且不受节点和基站数目变化的影响,适合解决H2H和M2M共存场景中的用户分配问题.     

基于Landweber迭代算法的欠采样恢复数字预失真技术

传统宽带数字预失真(DPD)为了更好地矫正功率放大器(PA)非线性特性,通常要求反馈通道带宽达到发送信号带宽的5倍,相应地要求更高采样率的模数转换器(ADC),这将导致数字预失真系统面临着硬件成本和能耗问题.针对这一问题,该文提出一种基于Landweber迭代算法的欠采样恢复(USR)数字预失真(Landweber-USR DPD)技术.这种以内外循环的方式进行处理,可将反馈通道带宽从理论要求的5倍降低至2倍,以良好的质量从欠采样的功放输出信号中恢复全频带的输出信号,使还原出的数据更接近真实的功放输出信号,以实现更好的预失真效果.实验选用基于单管氮化镓(GaN)器件的宽带F类功率放大器,在1.8 GHz工作频点下用5 MHz的长期演进(LTE)信号激励,反馈ADC速率分别设置为全采样速率(40 Msps)和欠采样速率(10 Msps).实验结果充分证明了Landweber迭代算法恢复功放数据的可靠性以及Landweber-USR DPD技术的有效性,为宽带通信系统中数字预失真技术的工程实现提供了有效降低ADC采样率的思路和方法.     

Unraveling the Charge Storage and Activity-Enhancing Mechanisms of Zn-Doping Perovskite Fluorides and Engineering the Electrodes and Electrolytes for Wide-Temperature Aqueous Supercabatteries

Herein, a trimetallic Ni–Co–Zn perovskite fluoride (ABF₃) (denoted as KNCZF) electrode material is explored for advanced aqueous supercabatteries (ASCBs), with KNCZF and activated carbon–FeBiCu@reduced graphene oxides (AC–FeBiCu@rGO) as cathode and anode, respectively, which outperform aqueous supercapacitors (ASCs) and batteries (ABs) with AC and FeBiCu@rGO anodes because of the synergistic effect of pseudocapacitive (KNCZF), capacitive (AC), and faradaic (FeBiCu@rGO) responses. One of the important findings is that the KNCZF shows a typical bulk phase conversion mechanism for charge storage in the alkaline media with the transition of ABF₃ perovskite nanocrystals into amorphous metal oxides/(oxy)hydroxides nanosheets, showing the redox-active and redox-inert roles for the Ni/Co and Zn species, respectively, which can be deduced by various ex-situ techniques. Another interesting finding is that the redox-inert Zn species largely enhance the activity of Ni/Co redox-active species in the ABF₃ materials, mainly owing to the promotion of surface electroactive sites, adsorption of OH^?, and charge transfer of surface Ni/Co atoms by Zn-doping, which can be proved by ex-situ characterizations and theoretical calculations. Overall, this study reveals the structure–activity relationship and charge storage mechanisms of Zn-doping ABF₃ materials for advanced ASCBs, showing a great impact on developing advanced electrochemical energy storage.     

Accelerating approximate matrix multiplication for near-sparse matrices on GPUs

The Journal of Supercomputing - Although the matrix multiplication plays a vital role in computational linear algebra, there are few efficient solutions for matrix multiplication of the near-sparse...     

一种面向低采样率的点云数据处理网络

在直接处理点云的三维神经网络中,采样阶段实现了对原始点云中关键点的筛选,对于整个网络的性能及网络的抗噪能力具有重要作用。目前主流的最远点采样（FPS）方法在处理大规模3D点云数据时计算量大且耗时,并且低采样率时经过FPS采样后模型性能下降明显。针对这两个问题,提出一种面向低采样率的点云数据处理网络AS-Net。设计一个新的采样模块代替原backbone中的FPS,其由两个Layer组成,每个Layer基于长短期记忆网络获取原始点云与采样点云之间的联系权重,从而高效提取关键信息,去除冗余信息。在此基础上,利用注意力机制选择特征值较高的原始点云作为采样点,采样点作为后序任务的关键点输入到网络,进一步提高网络模型性能。基于ModelNet40数据集的实验结果表明,在低采样率条件下,AS-Net仍可达到81.6%的分类准确率,与使用FPS作为采样方法的网络模型相比提高52.7%。此外,其对噪声干扰具有很强的鲁棒性,对于大场景的分割时间效率优于同类采样方法。