Advancing thermoelectrics by suppressing deep-level defects in Pb-doped AgCrSe_(2) alloys

AgCrSe2-based compounds have attracted much attention as an environmentally friendly thermoelectric material in recent years due to the intriguing liquid-like properties.However,the ultra-low carrier concentration and the high Ag_(Cr)deep-level defects limit the overall thermoelectric performance.Here,we successfully introduced Pb into Ag-deficient Ag_(0.97)CrSe_(2) alloys to tune the carrier concentration across a broad temperature range.The Pb^(2+) as an acceptor dopant preferentially occupies Cr sites,boosting the hole carrier concentration to 1.77×10^(19) cm^(-3) at room temperature.Furthermore,the Pb strongly inhibits the creation of intrinsic Ag_(Cr) defects,weakens the increased thermal excited ionization with the increasing temperature and slowed the rising trend of the carrier concentration.The designed carrier concentration matches the theoretically predicted optimized one over the entire temperature range,leading to a remarkable enhancement in power factor,especially the maximum power factor of ~500 μW·m^(-1)·K^(-2) at 750 K is superior to most previous results.Additionally,the abundant point defects promote phonon scattering,thus reducing the lattice thermal conductivity.As a result,the maximum figure of merit zT(~0.51 at 750 K) is achieved in Ag_(0.97)Cr_(0.995)Pb_(0.005)Se_(2).This work confirms the feasibility of manipulating deep-level defects to achieve temperature-dependent optimal carrier concentration and provides a valuable guidance for other thermoelectric materials.

The relationship between the deep-level structure in crust and brewing of strong earthquakes in Xingtai area

In this paper, crust medium is treated as Maxwell medium, and crust model includes hard inclusion, soft inclusion,deep-level fault. The stress concentration and its evolution with time are obtained by using three-dimensional finite element method and differential method. The conclusions are drawn as follows:①The average stress concentration and maximum shear stress concentration caused by non-heterogeneous of crust are very high in hard inclusion and around the deep fault. With the time passing by, the concentration of average stress in the model gradually trends to uniform. At the same time, the concentration of maximum shear stress in hard inclusion increases gradually. This character is favorable to transfer shear strain energy from soft inclusion to hard inclusion. ② When the upper mantle beneath the inclusion upheave at a certain velocity of 1 cm/a the changes of average stress concentration with time become complex, and the boundary of the hard and soft inclusion become unconspicuous, but the maximum shear stress concentration increases much more in the hard inclusion with time at a higher velocity. This feature make for transformation of energy from the soft inclusion to the hard inclusion. @ The changes of average stress concentration and maximum shear stress concentration with time around the deep-level fault result in further accumulation of maximum shear stress concentration and finally cause the deep-level fault instable and accelerated creep along fault direction. ④ The changes of vertical displacement on the surface of the model, which is caused by the accelerated creep of the deep-level fault, is similar to that of the observation data before Xingtai strong earthquake.

Influence of a deep-level-defect band formed in a heavily Mg-doped GaN contact layer on the Ni/Au contact to p-GaN

The influence of a deep-level-defect（DLD） band formed in a heavily Mg-doped GaN contact layer on the performance of Ni/Au contact to p-GaN is investigated. The thin heavily Mg-doped GaN（p^＋＋-GaN） contact layer with DLD band can effectively improve the performance of Ni/Au ohmic contact to p-GaN. The temperature-dependent I–V measurement shows that the variable-range hopping（VRH） transportation through the DLD band plays a dominant role in the ohmic contact. The thickness and Mg/Ga flow ratio of p^＋＋-GaN contact layer have a significant effect on ohmic contact by controlling the Mg impurity doping and the formation of a proper DLD band. When the thickness of the p^＋＋-GaN contact layer is 25 nm thick and the Mg/Ga flow rate ratio is 10.29%, an ohmic contact with low specific contact resistivity of 6.97×10^-4Ω·cm^2 is achieved.

Numerical simulation of deep-level rockburst in Fuxin coalfield

On the basis of ANSYS finite element model(FEM) software, the deep-level rockburst in Fuxin coalfield was simulated numerically. Based on Haizhou Mine and Wulong Mine as two typical deep-level rockburst examples in Fuxin coalfield, the rules and characteristics of the deep-level rockburst were analyzed. And the models were es- tablished. For Haizhou mine, the relationship between mining distance and rockburst was presented when 100, 300, 600 m were mined in 3313 working face. When 300 m were mined, the rockburst began to emerge. When 600 m were mined, the rockburst was the most possible to happen and the compression stress of the working face reached to the maximum value. The effect of tectonic stress on synclinal axis is also a key factor to rockburst occurrence. This was verified by the rockburst happened when 496 m were mined. For Wulong mine, based on the 311 working face as an example, the contours of Y stress in the roof and floor were obtained when the mining distance were 100, 200, 300 and 400 m. When 100 and 400 m were mined, the high stress con- centration regions occurred in the front of working face. This shows the rockburst is easy to happen. It is confirmed by the rockburst when 91m were mined in 311 working plane. The above indicates that the numerical simulation has instructive rule to study the deep-level rockburst in Fuxin coalfield.

领导—员工职业定位匹配对青年新员工早期组织适应的影响研究

对于初入职场的青年员工而言,他们与领导之间可能存在多元的职业定位,“以组织为中心”和“以个人为中心”的观念发生碰撞,为他们融入集体带来了新挑战。对此,本研究基于个人—环境匹配理论,聚焦于当前青年新员工与领导的职业定位匹配情况,采用多项式回归和响应面分析方法,对235位主管领导匹配问卷数据进行实证检验,探究青年新员工早期的组织社会化机制,并验证感知深层差异的中介作用与包容型氛围的调节作用。研究发现:青年新员工在上下级职业定位不匹配的情况下,会产生较高水平的感知深层差异,难以顺利完成组织融入,早期组织适应表现较差;良好的包容型组织氛围可减轻青年新员工感知上的深层差异,并促进他们的早期组织适应。研究为理解职业定位对青年新员工早期组织适应产生影响的具体机制提供了理论基础与实践建议,启发管理者将包容型氛围作为促进青年新员工早期组织适应的重要手段。

基于交替注意力机制和图卷积网络的方面级情感分析模型

方面级情感分析旨在预测给定文本中特定目标的情感极性。针对忽略方面词和上下文之间的句法关系和平均池化带来的注意力差异性变小的问题,提出一种基于交替注意力(AA)机制和图卷积网络(GCN)的方面级情感分析模型(AA-GCN)。首先,利用双向长短期记忆(Bi-LSTM)网络对上下文和方面词进行语义建模;其次,通过基于句法依存树的GCN学习位置信息和依赖关系,再利用AA机制进行多层次交互学习,自适应地调整对目标词的关注度;最后,拼接修正后的方面特征和上下文特征,得到最终的分类依据。相较于基于目标依赖的图注意力网络(TDGAT),所提模型在4个公开数据集上准确率提升了1.13%~2.67%,在5个公开数据集上F1值提升了0.98%~4.89%,验证了利用句法关系和提升关键词关注度的有效性。

县级融媒体中心深度发展的几个关键问题

县级融媒体中心是地方党委政府“治县理政”的抓手与平台,作为一种战略安排,融媒体中心有其技术、经济、政治与文化时代背景,数字信息技术不断革新下的挑战与机遇,使得县级传统媒体必然转型为融媒体中心,因而其生产者构成、职能部门构成、渠道构成、产品构成、受众构成、社会构成等必然有新的规律和特点,但本质上依然是一种新闻机构。这就要求我们从宏观、中观、微观等层面来把握这种战略安排,对融媒体中心深度发展的逻辑遵循、核心构成、地方价值、舆论引导和文化建设的理解,就成为管理者、研究者和从业者所有工作的重中之重。

复杂场景下无水尺水位的影像水位反演智能检测方法

实现精细化水务管控和洪涝灾害预警,需要实时、准确感知水位突变事件。现有技术不能满足夜晚、雾霾、雨天、雪天、漂浮物遮挡及阴影等复杂恶劣环境下的水位识别需求。为此,本文提出一种融合改进YOLOv5与卡尔曼滤波原理的无水尺水位智能检测技术:①引入YOLOv5对水位线(水岸分界线)进行检测,并利用线性拟合方法获得实际水位线;②针对水位线在延伸方向无限大而在其法向无限小特点,提出强化中尺度特征的多层级特征融合方法改进原YOLOv5算法;③利用卡尔曼滤波引入水位历史信息作为先验知识,提高本技术对复杂恶劣环境的泛化性能;④将图像中事先标定的固定的标志物加入到深度学习网络中训练,根据标志位真实尺寸解算实际水位高程,实现无水尺检测方案。相关试验和实践表明,改进的YOLOv5更加轻量化;本文所述水位智能检测技术斜率准确性为97.3%,较原算法提高了2.4%;截距准确性为99.3%,较原算法提高了0.5%;在夜晚、雾霾、雨天、雪天、漂浮物遮挡及阴影等复杂恶劣环境下可以自动、准确识别出水位高程,误差小于0.1 m。

基于深度学习的PBX裂纹像素级识别方法

高聚物粘结炸药(Polymer Bonded Explosive,PBX)内部裂纹对其性能及可靠性具有明显影响,裂纹的精确识别是PBX内部质量检测的关键。针对PBX内部裂纹识别,基于U-Net和SegNet两种像素级图像识别网络,建立了5种深度学习网络,对比研究了网络连接方式和预训练模型对裂纹识别的影响。基于CT获取的PBX裂纹图建立了数据集,对5种网络进行了训练,采用准确率(A)、F1值(F1)和平均交并比(MIoU)指标对网络进行了评估,在此基础上选择细节识别性能最优的网络用于PBX裂纹的识别和量化分析。结果表明,针对裂纹像素级识别,U-Net型网络优于Seg-Net型网络,网络中Concatenate操作比Pooling Indices操作保留更多图像细节信息,采用预训练模型MobileNet和ResNet可以提高网络训练速度,但导致其裂纹像素级识别性能降低。利用建立的识别方法开展PBX裂纹识别研究,实现了对PBX裂纹的像素级识别,裂纹检出率0.9570,单像素识别准确率为0.9936,MIoU为0.9873,相对裂纹面积为0.7585。

基于ECSDNN的航空安全事件风险等级预测

航空安全事件风险等级预测是主动风险管理的重要手段。考虑海量航空安全事件数据呈现的高维复杂、类不平衡等特性,提出一种基于集成代价敏感深度神经网络(ECSDNN)的航空安全事件风险等级预测方法。采用分类型属性嵌入特征编码和数值型属性拼接的方法实现航空安全事件数据的特征表示;综合考虑错分比例和固定代价设计代价敏感矩阵和代价敏感损失函数,构建基于代价敏感深度神经网络(CSDNN)的基分类器模型;采用硬投票方法,集成多个参数不同、性能各异的基分类器,构建航空安全事件风险等级预测模型。在航空安全事件报告系统(ASRS)数据集上的实验结果表明:相比基准算法,所提ECSDNN模型的预测准确率提升了4.51%;相比单个CSDNN基分类器,所提ECSDNN模型的预测准确率提升了3.17%。验证了基于ECSDNN的航空安全事件风险等级预测方法的有效性。

The effects of deep-level defects on the electrical properties of Cd0.9Zn0.1Te crystals

The deep-level defects of Cd Zn Te（CZT）crystals grown by the modified vertical Bridgman（MVB）method act as trapping centers or recombination centers in the band gap,which have significant effects on its electrical properties.The resistivity and electron mobility–lifetime product of high resistivity Cd（0.9）Zn（0.1）Te wafer marked CZT1 and low resistivity Cd（0.9）Zn（0.1）Te wafer marked CZT2 were tested respectively.Their deep-level defects were identified by thermally stimulated current（TSC）spectroscopy and thermoelectric effect spectroscopy（TEES）respectively.Then the trap-related parameters were characterized by the simultaneous multiple peak analysis（SIMPA）method.The deep donor level（EDD/dominating dark current was calculated by the relationship between dark current and temperature.The Fermi-level was characterized by current–voltage measurements of temperature dependence.The width of the band gap was characterized by ultraviolet-visible-infrared transmittance spectroscopy.The results show the traps concentration and capture cross section of CZT1 are lower than CZT2,so its electron mobility–lifetime product is greater than CZT2.The Fermi-level of CZT1 is closer to the middle gap than CZT2.The degree of Fermi-level pinned by EDDof CZT1 is larger than CZT2.It can be concluded that the resistivity of CZT crystals increases as the degree of Fermi-level pinned near the middle gap by the deep donor level enlarges.

自适应特征融合与cosIoU-NMS的目标检测算法

针对经典的有锚框检测算法RetinaNet、无锚框检测算法FCOS等目标检测算法中存在漏检以及重复检测的问题,提出一种自适应特征融合与cosIoU-NMS的目标检测算法.首先采用自适应特征融合模块对多尺度特征中相邻3层特征加权融合,获取丰富的上下文信息和空间信息;然后采用cosIoU计算检测框之间的余弦相似度与重叠面积,使目标定位更准确;最后使用cosIoU-NMS代替Greedy-NMS抑制置信度分数较高的冗余框,保留更准确的检测结果.以RetinaNet和FCOS为基准,在PASCAL VOC数据集上的实验结果表明,所提算法的检测精度达到81.3%和82.3%,分别提升2.8个百分点和1.2个百分点;在MSCOCO数据集上检测精度达到36.8%和38.0%,分别提升1.0个百分点和0.7个百分点;该算法能够增强特征表征能力,筛除多余的检测框,有效地提高检测性能.

视觉Transformer在低级视觉领域的研究综述

Transformer是一种革命性的神经网络模型架构,最初为自然语言处理而设计,但其由于卓越的性能,在计算机视觉领域获得了广泛的应用。虽然关于Transformer在自然语言处理领域的应用有大量的研究和文献,但针对低级视觉任务的综述相对匮乏。简要介绍了Transformer的原理并分析归纳了几种变体。在低级视觉任务的应用方面,将重点放在图像恢复、图像增强和图像生成这三个关键领域。通过详细分析不同模型在这些任务中的表现,探讨了它们在常用数据集上的性能差异。对Transformer在低级视觉领域的发展趋势进行了总结和展望,提出了未来的研究方向,以进一步推动Transformer在低级视觉任务中的创新和发展。这一领域的迅猛发展将为计算机视觉和图像处理领域带来更多的突破,为实际应用提供更加强大和高效的解决方案。

三空间注意力的残差U-Net在视网膜血管分割应用

针对视网膜图像末端微小血管对比度低、分割不精确问题,提出一种融合多层次残差与三空间注意力机制的U型网络用于视网膜眼底血管分割。该网络在编码部分为了减少图像特征的丢失,引入原始图像后添加多层次残差模块。此外,为防止网络深层产生梯度弥散与特征数据冗余问题,在残差模块中进一步加入批量归一化与Dropout功能。在解码部分,采用三空间注意力机制来赋予类原始图像、下采样图像和上采样图像特征不同的权重,以增强特征纹理和位置信息,并实现微小血管的精确分割。实验结果显示,在公开的彩色眼底图像数据集上,与现有算法相比,本文算法的准确率、特异性、灵敏度和AUC分别为0.985、0.991、0.829和0.985,与金标准分割结果进行对比得到的血管图具有重要的临床参考价值。

基于自适应时序表征和多级注意力的超短期风电功率预测

针对风电功率数据包含的多尺度时间信息难以描述、现有方法未充分考虑气象因素对于风电功率动态耦合的影响而导致的预测性能下降等问题,提出了一种基于自适应时序表征和多级注意力的超短期风电功率预测方法。采用时序嵌入层对风电功率序列进行表征以获取其周期、非周期模式,并引入自注意力捕捉高维风电功率序列的自相关性;利用交叉注意力重构风电功率与气象因素,形成包含两者耦合关系的多维特征序列;利用一维卷积神经网络沿时间、特征方向分别挖掘多维特征序列的时间相关性和空间相关性,进而利用长短期记忆网络提取相应的时序特征,并将所得时序特征经全局注意力去噪和门控机制融合后输入全连接层,分别进行点预测和区间预测。实验结果表明,所提方法能够获得准确的点预测值和可靠的预测区间。

基于改进YOLOX的城市河道智能水位测量算法

针对目前基于深度学习水位测量算法存在特征信息提取不充分问题,提出一种基于改进YOLOX的城市河道水位智能测量算法。为了提高YOLOX对多类别密集目标的识别率,在特征融合网络中引入CBAM注意力机制,并采用基于计算目标框信息的损失函数D-IoU加快模型收敛。该算法利用改进后的YOLOX对水尺刻度进行识别与统计,并计算出水位值。试验表明提出的新算法对水尺刻度和数字的平均识别率分别达98.62%和92.23%,最终计算水位的平均误差为1.16 cm,较其他图像识别水位测量算法的平均误差减少了1.76 cm,可实现高精度智能测量城市河道的水位值。

DetSegNet:一种基于检测和分割的高精度水尺水位检测网络

为了克服传统水位测量方法中水位边缘粗糙和检测精度不足的问题,提出了一种高精度的水尺水位检测网络——DetSegNet。DetSegNet在网络结构、损失函数等方面对YOLOv4和DeepLabv3+算法进行了改进,并将水尺刻度识别、水体区域分割、水位线检测和水位值计算等模块进行有效结合,实现了对水尺与水体交界区域的高效定位和精确分割。实验结果表明,DetSegNet在水尺图像数据集上的检测精度和速度均优于传统的检测方法;现场测试表明,DetSegNet的水位检测误差小于1 cm,满足水文监测的精度要求。

纤锌矿结构砷化镓纳米线中的深能级缺陷研究

深能级缺陷是造成纤锌矿结构砷化镓纳米线中持续光电导效应的主要原因。采用基于高斯分布的缺陷复合动力学方程分析了纤锌矿结构砷化镓纳米线的光电导衰减曲线,得到平均载流子捕获势垒为60.2 meV。通过分析光照下光电导瞬态行为,提出了基于缺陷光电离模型来提取特定缺陷能级水平特征的方法。通过拟合纤锌矿结构砷化镓纳米线的光离化谱,得到了0.69 eV的光电离能。光离化能和热捕获能之间的较大能量差异意味着缺陷与晶格有强耦合作用,这与闪锌矿结构砷化镓中EL2中心的行为相似。

基于FPGA的便携心电智能诊断加速器及优化选芯方案

心电图(electrocardiogram,ECG)是诊断与心脏相关疾病的关键工具,可穿戴心电监护仪Holter是院外检测的重要手段,小型化、便携性、实时检测是优化方向。人工智能技术应用于包括心电诊断的各个领域,但存在参数量大、难于小型化、计算速度慢的问题,不满足便携心电监护仪的要求,而可编程逻辑门器件(Field-Programmable Gate Array,FPGA)有并行加速的特性。在AI智能算法硬件化的工程应用上,存在成本、速度、资源利用率的权衡,需要进行科学的芯片选型。开发了一种基于1D-CNN的、用于心电诊断的BeatNet,对于4分类的检测任务,该模型具有98.5%的分类准确率。在FPGA上实现BeatNet硬件化的实验,经测试在Altera公司的DE1-SoC开发板上部署RTL级硬件网络电路,最高频率为69 MHz,单个心拍诊断需要221个周期,总计算时间为3.31μs,满足实时标准。针对速度、功耗和成本指标评估了不同部署策略的性能,开发了一个选芯综合指数,可以在仅损失0.2%精度、5μs内完成单心拍的条件下,将FPGA计算芯片成本控制在200美元以内,满足高速度、低成本、低功耗的要求,对工程优化有实用价值。

基于大模型的深度课堂变革研究

大模型技术推动课堂教学走向“师—生—机”的三元互动,也倒逼教师转变角色。传统教学中,教师多以问题和任务促进深度课堂的实施。智能技术支持下,传统教学的“四何”问题开始转变,很大部分成为机器眼里的“是何”问题,如何重构课堂让学习者获得深度认知成为关键。基于此,本研究构建了以“四何”问题为内核的驱动任务、以“师—生—机”为三元主体的深度课堂角色转换机制,形成支持深度课堂目标达成的变革路径,并以B大学“学习科学”课程的研究生为对象开展实践。研究发现,学习者具备较高的学习动机、技术接受度和较低的认知负荷;相比低分组,高分组认知负荷低,能更好地处理技术和复杂任务带来的压力。研究结论可为大模型支持的深度课堂实施提供启示。