改进HSR-FCN的服装图像识别分类算法研究

目前网络上的服装图像数量增长迅猛，对于大量服装图像实现智能分类的需求日益增加。将基于区域的全卷积网络（Region-Based Fully Convolutional Networks，R-FCN）引入到服装图像识别中，针对服装图像分类中网络训练时间长、形变服装图像识别率低的问题，提出一种新颖的改进框架HSR-FCN。新框架将R-FCN中的区域建议网络和HyperNet网络相融合，改变图片特征学习方式，使得HSR-FCN可以在更短的训练时间内达到更高的准确率。在模型中引入了空间转换网络，对输入服装图像和特征图进行了空间变换及对齐，加强了对多角度服装和形变服装的特征学习。实验结果表明，改进后的HSR-FCN模型有效地加强了对形变服装图像的学习，且在训练时间更短的情况下，比原来的网络模型R-FCN平均准确率提高了大约3个百分点，达到96.69%。     

用于疏浚工程泥浆密度测量的电阻层析成像技术

江河湖海疏浚工程对于拓展和保障航运安全、促进生态恢复、维护/改善远海岛礁环境具有不可或缺的作用.目前,在绞吸和耙吸疏浚船上,多采用射线密度计及电磁流量计对输泥密度、产量等进行在线计量,并作为实现疏浚系统自动控制的重要参量.由于射线密度计潜在的人员健康、公共安全和环保风险,在一些国家和地区已建立了放射源的有序退出机制.电阻层析成像使用安装于管壁的电极阵列获得管道内流体的电导率分布,进而推导出管道内的泥浆密度,具有无辐射、全场测量及实时可视化的优点.电阻层析成像测量系统经过特殊设计,可适应从内陆淡水到海水的巨大电导率变化.初步测试表明,电阻层析成像测量数据与射线密度计具有良好一致性,具备了作为在线泥浆密度测量的可行性.此外,ERT可提供实时的流动成像,对于提高对疏浚管道监测能力和降低堵管概率具有重要意义.     

可重构机器人模块间装配误差识别方法研究

基于对可重构机器人配对接口间位姿误差的分析划分,研制模块间装配误差在线测量识别接口,建立配对接口位姿误差与接口几何结构偏差之间的关系模型,进而建立配对接口位姿误差与配对接口内部测距传感器测量值之间的映射关系模型,提出一种基于内部测距传感器位移量的模块间装配误差在线识别、补偿方法.为验证方法的正确性,研制单关节?连杆试验平台.试验结果表明装配误差经在线识别、补偿后,连杆末端的位置误差平均值减少7倍多.     

Crystallographic characteristics and microwave dielectric properties of Ni-modified MgTa2O6 ceramics

Ni²⁺ modified MgTa₂O₆ ceramics with a trirutile phase and space group P4₂/mnm were obtained. The correlations between crystallographic characteristics and microwave dielectric performance of MgTa₂O₆ ceramics were systematically studied based on the chemistry bond theory (PVL theory) for the first time. The results indicate that the introduction of Ni²⁺ causes a change in polarizability and the Mg–O bond ionicity, which contributes to the variation of dielectric constant. Moreover, the lattice energy, and packing fraction, full width at half maximum of the Raman peak of Ta–O bond, as the quantitative characterization of crystallographic parameters, regulate the dielectric loss of MgTa₂O₆ ceramics in GHz frequency band. In addition, the study of sintering behavior shows that the densification and micromorphology are the crucial factors affecting the microwave dielectric performance. Typically, Ni²⁺ doping on the A-site of MgTa₂O₆ can effectively promote the Q × f values to 173,000 GHz (at 7.43 GHz), which ensures its applicability in 5G communication technology.     

含锡废液的安全处理技术研究

论述了两级沉淀法对含锡废液的处理技术,通过实验得到两级沉淀法处置含锡废液的最优条件,实现了含锡废液的安全有效快速处理.     

Investigation of crystal characteristics,Raman spectra,and microwave dielectric properties of Mg1-xZnxTa2O6 ceramics

Mg_(1-x)Zn_xTa₂O₆ (x = 0.00?0.08) dielectric ceramics were synthesized via the traditional solid-state reaction method. We used XRD and Rietveld refinement to demonstrate that a pure Mg_(1-x)Zn_xTa₂O₆ phase with trirutile structure was formed. Zn²⁺ substitution helped to decrease the Raman full width at half width of the A_1g mode at 703 cm^?1, which resulted in an increase in the order and rigidity of the TaO₆ octahedron, this in turn contributed to improving the Q×f values. Additionally, the introduction of Zn²⁺ significantly promoted grain growth and increased the dense, and the molecular polarizability, these factors lead to a higher permittivity. Moreover, enhanced Ta-O bond energy resulted in a more stable TaO₆ octahedron in the Mg_(1?x)Zn_xTa₂O₆ system, which contributed to enhanced τ_f values via substitution of Zn²⁺ doped on the A-site. Correspondingly, the microwave dielectric properties were significantly improved for 0.04-doped samples, obtaining: ε_r = 27, Q × f = 185,000 GHz (at 7.47 GHz), τ_f =32 ppm/°C.     

一种抗运动干扰的实时心率提取方法

针对基于容积脉搏波（PPG）提取运动心率时，传统心率提取算法由于运动噪声干扰使测量结果误差大、实时性不好的问题，提出一种抗运动干扰的实时心率提取方法。该方法通过实时小波去噪，同时结合三轴加速度信号（ACC）对运动进行分类训练，计算各运动状态心率增益，对实时心率值进行补偿。实验结果表明，通过与同时采集的ECG信号计算出的实时心率进行对比，绝对误差率仅为1.2%左右。相比传统心率提取算法，该算法具有抗干扰性强，实时准确的特点。     

Modeling Impact of ROADM Intra-Node Add/Drop Contention on a Single Node Lightpath Blocking Performance

We propose an analytical model to evaluate the lightpath blocking performance for a single ROADM node with intra-node add/drop contention,in which the number of lightpaths that can be added/dropped with the same wavelength is limited by the add/drop contention factor.Different models of traffic load per nodal degree are considered to validate the effectiveness of the analytical model.The simulation results show that the proposed analytical model is effective in predicting the performance for different values of add/drop contention factor C and for variable offered loads at the node.The add/drop contention factor shows an important impact on the lightpath blocking performance and properly raising the contention factor can significantly improve the lightpath blocking performance.When the add/drop contention factor C exceeds a certain level,the performance of a ROADM with intra-node contention is close to that of a contentionless ROADM.