Fine-Grained Action Recognition Based on Temporal Pyramid Excitation Network

Mining more discriminative temporal features to enrich temporal context representation is considered the key to fine-grained action recog-nition.Previous action recognition methods utilize a fixed spatiotemporal window to learn local video representation.However,these methods failed to capture complex motion patterns due to their limited receptive field.To solve the above problems,this paper proposes a lightweight Temporal Pyramid Excitation(TPE)module to capture the short,medium,and long-term temporal context.In this method,Temporal Pyramid(TP)module can effectively expand the temporal receptive field of the network by using the multi-temporal kernel decomposition without significantly increasing the computational cost.In addition,the Multi Excitation module can emphasize temporal importance to enhance the temporal feature representation learning.TPE can be integrated into ResNet50,and building a compact video learning framework-TPENet.Extensive validation experiments on several challenging benchmark(Something-Something V1,Something-Something V2,UCF-101,and HMDB51)datasets demonstrate that our method achieves a preferable balance between computation and accuracy.

Treatable focal region modulated by double excitation signal superimposition to realize platform temperature distribution during transcranial brain tumor therapy with high-intensity focused ultrasound

Recently, the phase compensation technique has allowed the ultrasound to propagate through the skull and focus into the brain. However, the temperature evolution during treatment is hard to control to achieve effective treatment and avoid over-high temperature. Proposed in this paper is a method to modulate the temperature distribution in the focal region.It superimposes two signals which focus on two preset different targets with a certain distance. Then the temperature distribution is modulated by changing triggering time delay and amplitudes of the two signals. The simulation model is established based on an 82-element transducer and computed tomography(CT) data of a volunteer's head. A finitedifference time-domain(FDTD) method is used to calculate the temperature distributions. The results show that when the distances between the two targets respectively are 7.5–12.5 mm on the acoustic axis and 2.0–3.0 mm in the direction perpendicular to the acoustic axis, a focal region with a uniform temperature distribution(64–65℃) can be created.Moreover, the volume of the focal region formed by one irradiation can be adjusted(26.8–266.7 mm^3) along with the uniform temperature distribution. This method may ensure the safety and efficacy of HIFU brain tumor therapy.

Recognition of multi-carrier OFDM and single-carrier with alpha-stable distribution noise

In order to identify the multi-carrier orthogonal frequency division multiplexing(OFDM) and the single-carrier signal in the non-Gaussian noise environment, different features of the two signals are analyzed in terms of five parameters: generalized normalized fourth-order cumulant, the maximum value of the instantaneous amplitude power spectral density, absolute standard deviation of instantaneous phase on the section with weak signals, and position and numbers of the generalized cyclic spectrum's peak. The recognition method of the multi-carrier OFDM and single-carrier signal is proposed in the environment with alpha-stable distribution noise. Simulation results show that the recognition rate of the multi-carrier OFDM can reach 100% when the mixed signal to noise ratio(MSNR) is greater than-5 dB and the recognition rate can reach 90% for the single-carrier when the MSNR is greater than 2 dB.

Photo-Thermal Induced Optical Scattering Modulation Sensor for Malaria Diagnosis

Malaria is one of the leading killer diseases in sub-Saharan Africa. Although the disease is curable, early and accurate diagnosis is key to effective therapy. Existing malaria diagnostic techniques have low detection accuracy especially when the parasite load in the blood is low. In this paper, we report on a simple photo-thermal based technique for detection of the Plasmodium parasites’ biomarker (hemozoin) in blood smear samples. The technique has demonstrated 100% Plasmodium detection sensitivity and specificity from the ten blood smear test samples used.

On the Characteristic of Orthogonality for Multi-Carrier-Based Systems

The dominate factors caused by reasons of losing the orthogonality in fading channel including FBC (fading branch correlation), CFO (carrier frequency offset), and even the CPN (carrier phase noise) are novel discussion in this letter. The first one factor causes the ISI (inter-symbol interference), and, however, the latter two will lead to the phenomenon of CPE (common phase error) and ICI (inter-carrier interference). On the other hand, they will lead to the loss of orthogonality for the radio system with multi-carrier modulating schemes, i.e., both them mainly deteriorate a wireless communication system. Eventually, in this letter not only the analytical expressions are derived, but a three dimension numerical results from the analysis involve the three parameters also illustrated is.

基于双阶段特征解耦网络的单幅图像去雨方法

针对现有的单幅图像去雨方法无法有效地平衡背景图像细节恢复与有雨分量去除问题,提出一种基于双阶段特征解耦网络的单幅图像去雨方法,采用渐进式的学习方式从粗到细分阶段进行单幅图像去雨.首先构建压缩激励残差模块,实现背景图像与有雨分量的初步分离;然后设计全局特征融合模块,其中特别引入特征解耦模块分离有雨分量和背景图像的特征,实现细粒度的图像去雨;最后利用重构损失、结构相似损失、边缘感知损失和纹理一致性损失构成的复合损失函数训练网络,实现高质量的无雨图像重构.实验结果表明,在Test100合成雨图数据集上,所提方法峰值信噪比为25.57dB,结构相似性为0.89;在100幅真实雨图上,所提方法的自然图像质量评估器为3.53,无参考图像空间质量评估器为20.16;在去雨后的RefineNet目标分割任务中,平均交并比为29.41%,平均像素精度为70.06%;视觉效果上,该方法能保留更多的背景图像特征,有效地辅助下游的目标分割任务的开展.

基于改进Deformable DETR的无人机视频流车辆目标检测算法

针对无人机视频流检测中小目标数量多、因图像传输质量较低而导致的上下文语义信息不充分、传统算法融合特征推理速度慢、数据集类别样本不均衡导致的训练效果差等问题,提出一种基于改进Deformable DETR的无人机视频流车辆目标检测算法。在模型结构方面,该算法设计了跨尺度特征融合模块以增大感受野,提升小目标检测能力,并采用针对object_query的挤压-激励模块提升关键目标的响应值,减少重要目标的漏检与错检率;在数据处理方面,使用了在线困难样本挖掘技术,改善数据集中类别样本分布不均的问题。在UAVDT数据集上进行了实验,实验结果表明,改进后的算法相较于基线算法在平均检测精度上提升了1.5%,在小目标检测精度上提升了0.8%,并在保持参数量较少增长的情况下,维持了原有的检测速度。

Exercise-induced adaptation of neurons in the vertebrate locomotor system

Vertebrate neurons are highly dynamic cells that undergo several alterations in their functioning and physiologies in adaptation to various external stimuli.In particular,how these neurons respond to physical exercise has long been an area of active research.Studies of the vertebrate locomotor system’s adaptability suggest multiple mechanisms are involved in the regulation of neuronal activity and properties during exercise.In this brief review,we highlight recent results and insights from the field with a focus on the following mechanisms:(a)alterations in neuronal excitability during acute exercise;(b)alterations in neuronal excitability after chronic exercise;(c)exercise-induced changes in neuronal membrane properties via modulation of ion channel activity;(d)exercise-enhanced dendritic plasticity;and(e)exercise-induced alterations in neuronal gene expression and protein synthesis.Our hope is to update the community with a cellular and molecular understanding of the recent mechanisms underlying the adaptability of the vertebrate locomotor system in response to both acute and chronic physical exercise.

基于模态分析的双定子电励磁场调制电机弹性模量调整方法

准确获得电机结构的模态分析结果是研究电磁噪声的前提条件。该文以一台双定子电励磁场调制电机为研究对象,首先,利用有限元软件对影响其模态分析的弹性模量展开研究,探索弹性模量变化对各阶固有频率的相对影响率大小,找到外定子、电枢绕组、机壳和端盖的弹性模量变化对模态分析结果的影响规律,由此明确参数调整的方向性。然后,采用锤击法进行模态实验,并获得外定子、外定子组件以及整机模态实验结果与仿真结果二者误差的大小。最后,根据二者误差大小及参数调整的方向性,选择合适的弹性模量调整值将其误差均调整到4%以内,从而得到精确的模态仿真结果,进而有利于后续该电机电磁噪声的分析。

基于改进残差网络的拉丝机减速箱故障诊断

减速箱对拉丝辊的转速固定有重要作用,由于拉丝设备结构紧密,内部零件的运行状态不易于观察,因此减速箱轮齿故障导致的转速配比异常很难被及时发现,针对拉丝机减速箱存在的故障诊断环节缺失问题,提出一种遗传算法与优化注意力模块改进的残差网络的故障诊断方法。首先,通过小波包分解与带通滤波的混合方法清洗数据,依照生产车间实际情况提出综合评价指标,并按照指标需求选择小波包分解层数;其次,针对残差网络与注意力模块进行改进;最后,将经过连通域分析与二值化后的特征图送入改进后的模型进行诊断。结果表明,该方法的诊断准确率比注意力-残差网络模型(Squeeze-and-Excitation-ResNet,SE-ResNet)提升了7.32%,比卷积神经网络-极限学习机模型(Convolutional Neural Network-Extreme Learning Machine,CNN-ELM)提升了8.81%,针对注意力模块(Squeeze-and-Excitation Module,SE)的改进将模型的单次诊断时间在原来的基础上缩短0.92 s,对塑编拉丝车间中减速箱的维护具有较大的实用价值。

基于全局注意力机制的Robust-PointPillars三维目标检测

提出了一种基于全局注意力机制的Robust-PointPillars三维目标检测方法,在智能驾驶的应用中,提高了目标检测的精度和鲁棒性。PointPillars等神经网络通过使用点云柱表示点云,具有实现三维目标检测的潜力。首先介绍了空间和通道双重注意力模块,以增强有学习价值的点云特征,解决了PointPillars缺乏点云柱内部学习机制和特征提取不足的问题;挤压与激励网络(SENet)模块的引入,使PointPillars对特征信息的学习理解能力得到进一步提高;最终,对受到干扰或缺失的传感器信号进行抑制,并利用全局注意力算法来提高鲁棒性。基于KITTI数据集上的目标检测结果,本文算法具有良好的目标检测精度和鲁棒性。

Theoretical Studies of the Output Pulse with Variation of the Pumping Pulse for RF Excited CO_2 Pulsed Waveguide Laser

The behavior of a RF-excited waveguide CO2 laser in the pulse regime is studied theoretically. The output pulse evolution is studied by applying three types of pulses namely the square, sine and the triangular ones as the excitation pulses. The frequency dependence behavior of the output pulse is also presented.

Photo-Doped Active Electrically Controlled Terahertz Modulator

We demonstrate an electric-controlled terahertz(THz) modulator which can be used to realize amplitude modulation of terahertz waves with slight photo-doping. The THz pulse transmission was efficiently modulated by electrically controlling the monolayer silicon-based device. The modulation depth reached 100% almost when the applied voltage was 7V at an external laser intensity of 0.6W/cm2. The saturation voltage reduced with the increase of the photo-excited intensity. In a THz continuous wave(CW)system, a significant fall in both THz transmission and reflection was also observed with the increase of applied voltage. This reduction in the THz transmission and reflection was induced by the absorption for electron injection. The results show that a high-efficiency and high modulation depth broadband electric-controlled terahertz modulator in a pure Si structure has been realized.

同步磁阻电动机新型位置估计与电感参数辨识

针对同步磁阻电动机电流导数和基本脉宽调制励磁(FPE)的方法数据处理量大的问题,研究了当前广泛使用的主流无传感器估计技术,从而提出了一种改进的估计方法并将其应用在同步磁阻电机的控制系统中。所提方法在每个脉宽调制(PWM)周期内仅需对两个有功电压矢量和一个零电压矢量作用下的单相电流进行采样、求导并计算,相比于传统方法中需要对两个有功电压矢量和两个零电压矢量作用下的三相电流进行采样、求导并计算而言,该创新估计算法不仅可以节约硬件成本,同时也大大降低了数据处理量。并且创新的估计算法在不降低精度和增加算法复杂程度的基础上,同时对电感参数实现精准辨识,使得该估计算法辨识的参数变多。使用仿真平台对同步磁阻电动机无传感器控制系统进行了仿真搭建。最后搭建了系统相应的硬件实验平台,通过实验验证了所提估计算法的优越性。

改进YOLOv5s的无人机目标检测算法

无人机在情报、侦察和监视领域,目标自动检测可为侦察等任务提供准确的目标位置及类别,为地面指挥人员提供详尽的目标信息。针对无人机图像背景复杂、分辨率高、目标尺度差异大等特点,提出一种改进YOLOv5s目标检测算法。将压缩-激励模块引入到YOLOv5s算法中,提高网络的特征提取能力;引入双锥台特征融合(bifrustum feature fusion,BFF)结构,提高算法对较小目标的检测检测精度;将CIoU Loss替换GIoU Loss作为算法的损失函数,在提高边界框回归速率的同时提高定位精度。实验结果表明,改进后的YOLOv5s取得了86.3%的平均均值精度(mAP),比原算法YOLOv5s提高了16.8个百分点,在复杂背景下仍能显著提升无人机图像目标检测性能。

基于U-Net多尺度自校准注意力视网膜分割算法

针对视网膜细小血管分割精度低的问题,提出一种融合可伸缩级联模块、Transformer和自校准注意力的改进U-Net算法以提高细小血管分割精度。首先在编码阶段利用可伸缩级联模块,先行学习复杂多变的视网膜血管拓扑结构。然后在解码阶段提出一种自校准注意力机制,利用多尺度挤压激励模块,自适应对特征图通道和空间之间特征重要性进行校准,增强目标区域特征响应,抑制背景噪声。最后使用Transformer特征提取块,提高特征空间映射能力。基于DRIVE和CHASEDB1数据集的实验结果表明,所提算法准确率分别为96.49%和96.67%,灵敏度分别为83.75%和83.30%,特异性分别为98.28%和98.01%,AUC分别为0.987 1和0.987 2,所提算法的整体性能优于现有算法,各模块能够有效提高细小血管分割能力。

集群金刚石NV色心量子自旋定向激发磁传感增强方法

为满足工业磁传感器高灵敏度、高信噪比(SNR)测量需求,基于自制的共聚焦磁检测系统,提出一种基于脉冲幅度调制技术的集群金刚石氮空位(NV)色心量子自旋定向激发(QSDE)方法。利用任意波形发生器构建sin型QSDE脉冲光探测磁共振序列,抑制传感体系中对微波敏感的Ns杂质与NV色心产生耦合效应带来的线宽展宽与噪声,提升谱线对比度的同时降低了线宽与噪声,突破微波强度对灵敏度和SNR的限制。经实验验证,相较连续波光探测磁共振(CW-ODMR)方法,该方法可将磁传感灵敏度和信噪比分别大约提升5倍和20 dB,为量子精密测量领域开辟了全新的技术路径和研究思路。

基于通道注意力和边缘融合的伪装目标分割方法

伪装目标分割(COS)的目标是从背景中分离出隐藏的目标对象。近年来,基于卷积神经网络(CNN)的伪装目标检测(COD)发展迅速,然而仍存在无法从前/背景高度相似的场景中准确地检测出完整目标对象的问题。针对上述问题,提出一种基于通道注意力(CA)和边缘融合的COS方法CANet(Network based on Channel Attention and edge fusion),以得到伪装目标的边缘细节更清晰的完整分割结果。首先,引入压缩和激励(SE)注意力模块,以提取更丰富的高级语义特征;其次,提出一个边缘融合模块,抑制低级特征中的干扰,并充分利用图像的边缘细节信息;最后,设计了基于深度可分离卷积的通道注意力模块,以自上而下的方式逐步融合跨级的多尺度特征,进一步地提升检测精度和效率。在多个公开的COD数据集上的实验结果表明,相较于SINet(Search Identification Net)、TINet(Textureaware Interactive guidance Network)和C2FNet(Context-aware Cross-level Fusion Network)等8种主流的方法,CANet表现更佳,且能够获取到丰富的伪装目标内部及边缘细节信息,而且在具有挑战性的COD10K数据集上结构度量指标相较于SINet提升了2.6个百分点。CANet性能优越,适用于医学上检测与人体组织相似的病灶区域、军事领域检测隐蔽目标等相关领域。

面向新型电力系统的变速抽蓄电站调频策略研究

针对变速抽蓄机组调频能力的高效利用问题,提出一种考虑混合抽蓄调频容量的协同调频策略研究。首先,以变分模态分解的调频信号高、低频分量作为混合抽蓄的初始调频参考;其次,利用2种抽蓄机组调频容量的差异对混合抽蓄的调频功率进行修正;最后,依据变速抽蓄机组转差率与调频容量关系,分配变速抽蓄机组之间的调频功率,实现混合抽蓄的有序调频。仿真结果表明了所提的变速抽蓄电站调频可以有效提高抽蓄电厂的调频响应速度。

基于压缩激励模块的工业产品缺陷分类算法

针对传统的工业产品表面缺陷检测技术检测效率低下、检测速度慢等问题,在以前工作的基础上,提出了基于压缩激励模块的工业产品缺陷分类算法。主要是在原有的并行模块与并行非对称卷积模块基础上增加了压缩和激励机制,通过显式的建模特征通道之间的相互依赖关系来提升网络的表征能力。在偏光片图像数据集上的实验结果表明,在不大量增加模型占用内存大小以及单张图片测试时间的基础上,显著提高了缺陷分类的正确率。