融合先验信息和特征约束的杆塔螺栓缺陷检测

目的 螺栓是输电线路中数量最多的紧固件,一旦出现缺陷就会影响电力系统的稳定运行。针对螺栓缺陷自动检测中存在的类内多样性和类间相似性挑战,提出了一种融合先验信息和特征约束的Faster R-CNN（fasterregions with convolutional neural network）模型训练方法。方法 在航拍巡检图像预处理阶段,设计了基于先验信息的感兴趣区域提取算法,能够提取被识别目标的上下文区域,从而减少模型训练阶段的数据量,帮助模型在训练阶段关注重点区域,提高其特征提取能力。在模型训练阶段,首先通过费舍尔损失约束Faster R-CNN模型的输出特征生成,使样本特征具有较小的类内距离和较大的类间间隔;然后采用K近邻算法处理样本特征得到K近邻概率,将其作为难易样本的指示以引导模型后续更加关注难样本。结果 在真实航拍巡检图像构建的螺栓数据集上进行测试,与基线模型相比,本文模型使螺栓识别的平均精度均值（mean average precision,mAP）提高了6.4%,其中正常螺栓识别的平均精度（average precision,AP）提高了0.9%,缺陷螺栓识别的平均精度提高了12%。结论 提出的融合先验信息和特征约束的输电杆塔螺栓缺陷检测方法在缺陷螺栓识别上获得了良好的效果,为实现输电线路螺栓缺陷的自动检测奠定了良好的基础。     

基于NS3-gym框架的智能干扰规避系统设计与实现

多智能体构建的无线通信网络在受到外部恶意干扰时,智能体需要与环境进行大量交互来学习干扰规律和优化抗干扰策略。为了有效模拟和验证智能体与外部干扰环境的学习交互过程,需要构建智能干扰规避仿真系统。提出了一种基于NS3-gym框架的智能干扰规避系统,NS3模拟智能通信网络场景并将感知到的网络状态数据作为智能体的输入,智能体对输入数据进行学习分析得到干扰规避决策,并通过gym与NS3之间的交互将其返回到NS3中的仿真网络进行抗干扰策略部署。NS3-gym框架提供了NS3和OpenAI gym之间进行信息交互的接口。在Ubuntu20.04系统下搭建了智能干扰规避系统的仿真平台,分别验证了Q学习算法以及WoLF-PHC算法在扫频干扰、贪婪随机策略干扰、跟随干扰、随机干扰四种场景下的抗干扰性能。仿真结果证明了所提系统架构与仿真平台的正确性和有效性。     

一种3D眼镜虚拟试戴系统的实现

针对人脸图像试戴3D眼镜过程中存在的镜腿遮挡人脸问题,文中提出一种基于人脸图像的3D眼镜虚拟试戴技术。利用构建的人脸形状的三维模型,使其在虚拟试戴中对镜腿起到消隐作用,解决镜腿的遮挡问题。文中对输入的人脸图像进行关键点检测,结合Graham扫描法求得人脸形状的凸多边形,利用平移扫描构建人脸形状的三维模型。此外,文中根据定位人脸图像上的关键点以及姿态估计后对三维眼镜模型的变换,将眼镜模型佩戴到人脸图像上。实验结果表明,该方法对于多视角的人脸图像实现了虚拟试戴效果,解决了多种视角下人脸图像试戴过程中镜腿的遮挡问题,虚拟试戴中镜腿遮挡平均准确率为94.5%,遮挡精度较高。     

Simultaneous wireless information and power transfer with fixed and adaptive modulation

Activating Wireless Power Transfer (WPT) in Radio-Frequency (RF) to provide on-demand energy supply to widely deployed Internet of Everything devices is a key to the next-generation energy self-sustainable 6G network. However, Simultaneous Wireless Information and Power Transfer (SWIPT) in the same RF bands is challenging. The majority of previous studies compared SWIPT performance to Gaussian signaling with an infinite alphabet, which is impossible to implement in any realistic communication system. In contrast, we study the SWIPT system in a well-known Nakagami-m wireless fading channel using practical modulation techniques with finite alphabet. The attainable rate-energy-reliability tradeoff and the corresponding rationale are revealed for fixed modulation schemes. Furthermore, an adaptive modulation-based transceiver is provided for further expanding the attainable rate-energy-reliability region based on various SWIPT performances of different modulation schemes. The modulation switching thresholds and transmit power allocation at the SWIPT transmitter and the power splitting ratios at the SWIPT receiver are jointly optimized to maximize the attainable spectrum efficiency of wireless information transfer while satisfying the WPT requirement and the instantaneous and average BER constraints. Numerical results demonstrate the SWIPT performance of various fixed modulation schemes in different fading conditions. The advantage of the adaptive modulation-based SWIPT transceiver is validated.     

回路压降冲激法的研究

常见的电路分析方法有很多,如叠加定理、戴维南定理和回路电流分析法等。为简化电路分析的复杂程度和减少求解电路方程的运算量,在回路电流法和线性叠加定理的基础上,提出一种简便的电路分析方法——回路压降冲激法。在分析线性网络的电路中,通过实验对比分析得出该电路分析方法可以简化分析过程,并且运算量明显减少。     

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.     

Multicolor Photo‐Crosslinkable AIEgens toward Compact Nanodots for Subcellular Imaging and STED Nanoscopy

Aggregation induced emission (AIE) has attracted considerable interest for the development of fluorescence probes. However, controlling the bioconjugation and cellular labeling of AIE dots is a challenging problem. Here, this study reports a general approach for preparing small and bioconjugated AIE dots for specific labeling of cellular targets. The strategy is based on the synthesis of oxetane‐substituted AIEgens to generate compact and ultrastable AIE dots via photo‐crosslinking. A small amount of polymer enriched with oxetane groups is cocondensed with most of the AIEgens to functionalize the nanodot surface for subsequent streptavidin bioconjugation. Due to their small sizes, good stability, and surface functionalization, the cell‐surface markers and subcellular structures are specifically labeled by the AIE dot bioconjugates. Remarkably, stimulated emission depletion imaging with AIE dots is achieved for the first time, and the spatial resolution is significantly enhanced to ≈95 nm. This study provides a general approach for small functional molecules for preparing small sized and ultrastable nanodots.     

Effects of Ge<Superscript>4+</Superscript> acceptor dopant on sintering and electrical properties of (K<Subscript>0.5</Subscript>Na<Subscript>0.5</Subscript>)NbO<Subscript>3</Subscript> lead-free piezoceramics

Lead-free (K_0.5Na_0.5)(Nb_1-xGe _x )O₃ (KNN-xGe, where x = 0-0.01) piezoelectric ceramics were prepared by conventional ceramic processing. The effects of Ge⁴⁺ cation doping on the phase compositions, microstructure and electrical properties of KNN ceramics were studied. SEM images show that Ge⁴⁺ cation doping improved the sintering and promoted the grain growth of the KNN ceramics. Dielectric and ferroelectric measurements proved that Ge⁴⁺ cations substituted Nb⁵⁺ ions as acceptors, and the Curie temperature (T_C) shows an almost linear decrease with increasing the Ge⁴⁺ content. Combining this result with microstructure observations and electrical measurements, it is concluded that the optimal sintering temperature for KNN-xGe ceramics was 1020°C. Ge⁴⁺ doping less than 0.4 mol.%can improve the compositional homogeneity and piezoelectric properties of KNN ceramics. The KNN-xGe ceramics with x = 0.2% exhibited the best piezoelectric properties: piezoelectric constant d₃₃ = 120 pC/N, planar electromechanical coupling coefficient k_p = 34.7%, mechanical quality factor Q_m = 130, and tanδ = 3.6%.     

Construction of point-line-plane (0-1-2 dimensional) Fe<Subscript>2</Subscript>O<Subscript>3</Subscript>-SnO<Subscript>2</Subscript>/graphene hybrids as the anodes with excellent lithium storage capability

The assembly of hybrid nanomaterials has opened up a new direction for the construction of high-performance anodes for lithium-ion batteries (LIBs). In this work, we present a straightforward, eco-friendly, one-step hydrothermal protocol for the synthesis of a new type of Fe₂O₃-SnO₂/graphene hybrid, in which zero-dimensional (0D) SnO₂ nanoparticles with an average diameter of 8 nm and one-dimensional (1D) Fe₂O₃ nanorods with a length of ~150 nm are homogeneously attached onto two-dimensional (2D) reduced graphene oxide nanosheets, generating a unique point-line-plane (0D-1D-2D) architecture. The achieved Fe₂O₃-SnO₂/graphene exhibits a well-defined morphology, a uniform size, and good monodispersity. As anode materials for LIBs, the hybrids exhibit a remarkable reversible capacity of 1,530 mA·g^?1 at a current density of 100 mA·g^?1 after 200 cycles, as well as a high rate capability of 615 mAh·g^?1 at 2,000 mA·g^?1. Detailed characterizations reveal that the superior lithium-storage capacity and good cycle stability of the hybrids arise from their peculiar hybrid nanostructure and conductive graphene matrix, as well as the synergistic interaction among the components.

Open image in new window