基于Transformer的实时语义分割网络及应用北大核心CSCD

为提升机器人对电缆隧道场景的环境感知能力,提出了一种适用于电缆隧道场景的实时语义分割网络。使用Transformer结构提取图像全局语义信息,同时使用卷积神经网络捕获图像局部细节信息,将二者结合增强图像分割精度。同时,针对机器人以及隧道特点,将语义分割结果转化为机器人前方目标信息以及机器人偏离情况信息,辅助机器人导航避障。实验结果表明,所提出的网络在公共数据集Cityscapes、CamVid上平均交叉联合度量分别为77.2%、76.8%,相比BiSeNet分别提高了8.8%、8.1%,并能较好地适应于实际电缆隧道环境中,提供有效的场景感知信息,实现自主避障。     

雪崩光电探测器的击穿效应模型分析

基于CMOS工艺制备了空穴触发的Si基雪崩探测器(APD),基于不同工作温度下器件的击穿特性,建立空穴触发的雪崩器件的击穿效应模型。根据雪崩击穿模型和击穿电压测试结果,拟合曲线得到击穿电场与温度的关系参数(dE/dT),器件在250～320 K区间内,击穿电压与温度是正温度系数,器件发生雪崩击穿为主,dV/dT=23.3 mV/K,其值是由倍增区宽度以及载流子碰撞电离系数决定的。在50～140 K工作温度下,击穿电压是负温度系数,器件发生隧道击穿,dV/dT=-58.2 mV/K,其值主要受雪崩区电场的空间延伸和峰值电场两方面因素的影响。     

带异形孔的双稳态压电能量采集器研究

针对传统线性压电悬臂梁能量采集器共振频率高、偏离共振频率时输出电压快速下降的问题,该文设计了一种悬臂梁基板上带异形孔的新型双稳态能量采集器。建立该能量采集器的理论模型,并制作了实验样机,研究了该能量采集器在外界不同正弦激振频率下,磁间距对其输出电压和工作频带的影响。结果表明,随着磁极对间距减小,带异形孔结构的双稳态能量采集器的双稳态效应先增强再减弱,由此确定最佳磁极对间距为12 mm,谐振频率为18 Hz,最大输出均方根电压达到12.01 V,采集器有效工作频率为15.5～22.5 Hz,工作带宽达到7 Hz,带异形孔的双稳态能量采集器具有更宽的采集频带,在低频振动环境下具有更高的输出电压响应。     

Weak signal detection and adaptive synchronous stability of a novel fifth-order memristive circuit system

This paper is devoted to introduce a novel complex fifth-order memristive circuit system and its applications in synchronous stability and weak signal detection. Firstly, the typical dynamical behaviors of the memristive system are discussed by chaotic phase portrait, complexity analysis, one-parameter bifurcation and Lyapunov exponent spectrum. Secondly, the adaptive control method is applied to realize the synchronization between the drive memristive system (DMS) and the response memristive system (RMS). The results indicate that the synchronization method has strong robustness and anti-interference ability. Thirdly, the weak signal detection of the novel five-dimensional memristive system is realized by using the extreme sensitivity of chaotic system to initial values. Finally, the fifth-order memristive circuit is designed by using basic electronic elements and simulated by Multisim software. And the anti-interference ability and sensitivity of the fifth-order memristive circuit are further verified by adding different weak disturbance signals at different positions of the circuit.     

Local Electronic Structure Modulation Enables Fast-Charging Capability for Li-Rich Mn-Based Oxides Cathodes With Reversible Anionic Redox Activity

Anionic and cationic redox chemistries boost ultrahigh specific capacities of Li-rich Mn-based oxides cathodes (LRMO). However, irreversible oxygen evolution and sluggish kinetics result in continuous capacity decay and poor rate performance, restricting the commercial fast-charging cathodes application for lithium ion batteries. Herein, the local electronic structure of LRMO is appropriately modulated to alleviate oxygen release, enhance anionic redox reversibility, and facilitate Li⁺ diffusion via facile surface defect engineering. Concretely, oxygen vacancies integrated on the surface of LRMO reduce the density of states of O 2p band and trigger much delocalized electrons to distribute around the transition metal, resulting in less oxygen release, enhancing reversible anionic redox and the MnO₆ octahedral distortion. Besides, partially reduced Mn and lattice vacancies synchronously stimulate the electrochemical activity and boost the electronic conductivity, Li⁺ diffusion rate, and fast charge transfer. Therefore, the modified LRMO exhibits enhanced cyclic stability and fast-charging capability: a high discharging capacity of 212.6 mAh·g⁻¹ with 86.98% capacity retention after 100 cycles at 1 C is obtained and to charge to its 80%, SOC is shortened to 9.4 min at 5 C charging rate. This work will draw attention to boosting the fast-charging capability of LRMO via the local electronic structure modulation.     

Silicon waveguide-based single cavity Fano resonance temperature sensor

A compact Fano resonant temperature sensor composed of a micro-ring resonator (MRR) coupled double-T-shaped waveguide is developed. The coupling gap and coefficient of the device are optimized by the finite difference time domain (FDTD) method. The maximum slope ratio (SR) of the MRR-coupled single-T-shaped waveguide is ?2.13 dB/nm. The SR of the double-T-shaped waveguide is ?49.69 dB/nm which is 23 times that of the single-T-shaped waveguide. The simulation results show that the temperature sensitivity of optical intensity decreases with increasing temperature in the range from 303.6 K to 343.8 K. The wavelength-temperature sensitivity of the double-T-shaped waveguide microring is 76.5 pm/K. After introducing the double-T-shaped waveguide structure, the device''s performance is greatly improved, and the double-T-shaped waveguide has a good application prospect as a temperature sensor.     

物联网时代下的智能电网发展

电力行业与物联网的关系较为密切,智能电网是电力系统发展到一定阶段的必然产物,物联网是智能电网发展的有利技术手段,在智能电网发展的各环节中,都有物联网的参与。文中对智能电网的发展进行了论述,并重点对物联网在智能电网各环节中的作用进行了探讨,最后介绍了两种面向智能电网应用的物联网设备。     

MXene Functionalized,Highly Breathable and Sensitive Pressure Sensors with Multi-Layered Porous Structure

Breathable, flexible, and highly sensitive pressure sensors have drawn increasing attention due to their potential in wearable electronics for body-motion monitoring, human-machine interfaces, etc. However, current pressure sensors are usually assembled with polymer substrates or encapsulation layers, thus causing discomfort during wearing (i.e., low air/vapor permeability, mechanical mismatch) and restricting their applications. A breathable and flexible pressure sensor is reported with nonwoven fabrics as both the electrode (printed with MXene interdigitated electrode) and sensing (coated with MXene/silver nanowires) layers via a scalable screen-printing approach. Benefiting from the multi-layered porous structure, the sensor demonstrates good air permeability with high sensitivity (770.86–1434.89 kPa⁻¹), a wide sensing range (0–100 kPa), fast response/recovery time (70/81 ms), and low detection limit (≈1 Pa). Particularly, this sensor can detect full-scale human motion (i.e., small-scale pulse beating and large-scale walking/running) with high sensitivity, excellent cycling stability, and puncture resistance. Additionally, the sensing layer of the pressure sensor also displays superior sensitivity to humidity changes, which is verified by successfully monitoring human breathing and spoken words while wearing a sensor-embedded mask. Given the outstanding features, this breathable sensor shows promise in the wearable electronic field for body health monitoring, sports activity detection, and disease diagnosis.     

A Visible and Near-Infrared Light-Fueled Omnidirectional Twist-Bend Crawling Robot

In recent years, although light-driven soft actuators have attracted intense scientific attention and achieved remarkable progress, the design and construction of an intelligent robotic system with maneuverability, self-adaptability, untethered control, and greater freedom of action, in particular the omnidirectional motion capability on a plane, remains challenging. Herein, four types of photo-thermal fillers and an unprecedented twist-bend actuation mode is introduced into a liquid crystal elastomer-based soft robot. The obtained twist-bend crawling robot not only exhibits in situ rotation, four-way turning, and four-way linear motion under light irradiation with four wavelength bands (520, 655, 808, and 980 nm), but also demonstrates the ability to avoid obstacles in complex geographical environments. This work may bring a new perspective for fabrication and development of soft robots that can adapt to dynamic and complex environmental conditions.     

基于BERT和LightGBM的文本关键词提取方法

传统的文本关键词提取方法忽略了上下文语义信息,不能解决一词多义问题,提取效果并不理想。基于LDA和BERT模型,文中提出LDA-BERT-LightG BM(LB-LightG BM)模型。该方法选择LDA主题模型获得每个评论的主题及其词分布,根据阈值筛选出候选关键词,将筛选出来的词和原评论文本拼接在一起输入到BERT模型中,进行词向量训练,得到包含文本主题词向量,从而将文本关键词提取问题通过LightG BM算法转化为二分类问题。通过实验对比了textrank算法、LDA算法、LightG BM算法及文中提出的LB-LightG BM模型对文本关键词提取的准确率P、召回率R以及F1。结果表明,当Top N取3～6时,F1的平均值比最优方法提升3.5%,该方法的抽取效果整体上优于实验中所选取的对比方法,能够更准确地发现文本关键词。