Pre-Buried ETL with Bottom-Up Strategy Toward Flexible Perovskite Solar Cells with Efficiency Over 23%

With rapid development of photovoltaic technology, flexible perovskite solar cells (f-PSCs) have attracted much attention for their light weight, high flexibility and portability. However, the power conversion efficiency (PCE) achieved so far is not yet comparable to that of rigid devices. This is mainly due to the great challenge of depositing homogeneous and high-quality perovskite films on flexible substrate. In this study, the pre-buried 3-aminopropionic acid hydroiodide (3AAH) additives into the electron transport layer (ETL) and modified the ETL/perovskite (PVK) interface by a bottom-up strategy. 3AAH treatment induced a templated perovskite grain growth and improved the quality of the ETL. By this, the residual stresses generated in PVK during the annealing-cooling process are released and converted into micro-compressive stresses. As a result, the defect density of f-PSCs with pre-buried 3AAH is reduced and the photovoltaic performance is greatly improved, reaching an exceptional PCE of 23.36%. This strategy provides a new idea to bridge the gap between flexible and rigid devices.     

Spring-Like Ammonium Salt Assisting Stress Release for Low-Temperature Deposited FAPbI3 Films Toward Flexible Photovoltaic Application

The substrates of conventional flexible perovskite solar cells (FPSCs) are thermoplastic polymer material polyethylene naphthalate (PEN), which will deform during high temperature annealing process. In addition, lead iodide (PbI₂) permanently formed and the substrate undergoes reversible deformation from 20 °C to 200 °C and back to 20 °C. Therefore, to balance the substrate supporting capacity and the crystalline quality of narrow band gap α-phase formamidinium lead iodide (α-FAPbI₃), an annealing process of 120 °C for 30 minutes is determined. Additionally, there will also be a large number of gaps and lattice strain at the perovskite grain boundaries during the annealing process as the FAPbI₃ phase transition is accompanied by much lattice shrinkage. As a result, 1,6-hexanediammonium diiodide (HADI) is chosen to passivate the defects and release the stress of perovskite film. Therefore, a recorded 1.4% extended stretch rate of the flexible film is attained. Finally, the champion PCE of 21.14% under AM 1.5G and 31.52% under 1062 lux is achieved after HADI treatment, accompanied by a better long-term and mechanical stability. This study provides annealing process optimization and stress relief strategies for the further development of narrow band gap FPSCs.     

Energy Transfer Induced by TADF Polymer Enables the Recycling of Excitons in Perovskite Solar Cells

Formamidinium lead triiodide (FAPbI₃) has been demonstrated as the most efficient perovskite system to date, due to its excellent thermal stability and an ideal bandgap approaching the Shockley-Queisser limit. Whereas, there are intrinsic quantum confinement effects in FAPbI₃, which lead to unwanted non-radiative recombination. Additionally, the black α-phase of FAPbI₃ is unstable under room temperature due to the significant residual tensile stress in the film. To simultaneously address the above issues, a thermally-activated delayed fluorescence polymer P1 is designed in the study to modify the FAPbI₃ film. Owing to the spectral overlap between the photoluminescence of P1 and absorption of the above-bandgap quantum wells of FAPbI₃, the Förster energy transfer occurs at the P1/FAPbI₃ interface, which further triggers the Dexter energy transfer within FAPbI₃. The exciton “recycling” can thus be realized, which reduces the non-radiative recombination losses in perovskite solar cells (PSCs). Moreover, P1 is found to introduce compressive stress into FAPbI₃, which relieves the tensile stress in perovskite. Consequently, the PSCs with P1 treatment achieve an outstanding power conversion efficiency (PCE) of 23.51%. Moreover, with the alleviation of stress in the perovskite film, flexible PSCs (f-PSCs) also deliver a high PCE of 21.40%.     

基于高分辨率网络和图卷积网络的三维人体重建模型

针对单目图像重建人体时出现的头部姿态翻转和图像特征间隐式空间线索缺失的问题,提出了一种基于高分辨率网络（HRNet）和图卷积网络（GCN）的三维人体重建模型。首先利用HRNet和残差块作为主干网络从原始图像中提取丰富的人体特征信息,然后使用GCN来捕获特征之间隐式的空间线索以获得空间精确的特征表示,最后使用此特征来预测多人线性蒙皮模型（SMPL）的参数以得到更加准确的重建结果;同时为了有效解决人体头部姿态翻转的问题,对SMPL的关节点重新进行了定义,在原有关节的基础上增加对头部关节点的定义。实验结果表明,所提模型能够准确地重建出三维人体,在2D数据集LSP上的重建准确率达到了92.41%,在3D数据集MPI-INF-3DHP上的关节误差和重建误差也大幅降低,平均误差仅分别为97.73 mm和64.63 mm,验证了所提模型在人体重建领域的有效性。     

RTDNet：面向高分辨率卫星影像的赤潮探测网络

目的 赤潮是一种常见的海洋生态灾害,严重威胁海洋生态系统安全。及时准确获取赤潮的发生和分布信息可以为赤潮的预警和防治提供有力支撑。然而,受混合像元和水环境要素影响,赤潮分布精细探测仍是挑战。针对赤潮边缘探测的难点,结合赤潮边缘高频特征学习与位置语义,提出了一种计算量小、精度高的网络模型RTD-Net （red tide detection network）。方法 针对赤潮边缘探测不准确的问题,设计了基于RIR （residual-in-residual）结构的网络,以提取赤潮边缘水体的高频特征;利用多感受野结构和坐标注意力机制捕获赤潮水体的位置语义信息,增强赤潮边缘水体的细节信息并抑制无用的特征。结果 在GF1-WFV （Gaofen1 wide field of view）赤潮数据集上的实验结果表明,所提出的RTDNet模型赤潮探测效果不仅优于支持向量机（support vector machine,SVM）、U-Net、Deep-Labv3+及HRNet （high-resolution network）等通用机器学习和深度学习模型,而且也优于赤潮指数法GF1_RI （Gaofen1 red tide index ）以及赤潮探测专用深度学习模型RDU-Net （red tide detection U-Net）,赤潮误提取、漏提取现象明显减少,F1分数在两幅测试图像上分别达到了0.905和0.898,相较于性能第2的模型DeepLabv3+提升了2%以上。而且,所提出的模型参数量小,仅有2.65 MB,约为DeepLabv3+的13%。结论 面向赤潮探测提出一种基于RIR结构的赤潮深度学习探测模型,通过融合多感受野结构和注意力机制提升了赤潮边缘探测的精度和稳定性,同时有效降低了计算量。本文方法展现了较好的应用效果,可适用于不同高分辨率卫星影像的赤潮探测。     

结合全局上下文与融合注意力的干涉相位去噪

目的 干涉相位去噪是合成孔径雷达干涉测量（interferometric synthetic aperture radar,InSAR）技术中的关键环节,其效果对测量精度具有重要影响。针对现有的干涉相位去噪方法大多关注局部特征以及在特征提取方面的局限性,同时为了平衡去噪和结构保持两者之间的关系,提出了一种结合全局上下文与融合注意力的相位去噪网络GCFA-PDNet(global context and fused attention phase denoising network)。方法 将干涉相位分离为实部和虚部依次输入到网络,先从噪声相位中提取浅层特征,再将其映射到由全局上下文提取模块和融合注意力模块组成的特征增强模块,最后通过全局残差学习生成去噪图像。全局上下文提取模块能提取全局上下文信息,具有非局部方法的优势;融合注意力模块既强调关键特征,又能高效提取隐藏在复杂背景中的噪声信息。结果 所提出的方法与对比方法中性能最优者相比,在模拟数据结果的平均峰值信噪比（peak signal to noise ratio, PSNR）和结构相似性（structural similarity,...     

基于改进ResNet-50的图像特征提取网络

为了提高图像的特征质量,保证最后提取到的特征高度精炼,提出了一种新的方法;该方法首先将低分辨率图像经过小波变换分解成高频分量和低频分量,并结合插值法进行插值,最后通过小波逆变换得到高分辨率图像来为后续的特征提取提供高质量的图片输入;接着,选取ResNet-50网络作为基础网络,将Efficient Channel Attention(ECA)模块与ResNet残差结构结合形成一个全新的ECA-ResNet50模块,ECA模块具有的通道级的注意力机制,可以让整个网络更加专注于提取显著特征;经实验测试,该方法对于图像特征提取的质量有着明显的提升,均方误差下降可达6.65;结果表明,该方法可行有效,具有良好的工程应用前景;     

自适应斥力系数的无人机路径规划

使用人工势场法进行无人机路径规划时,往往存在目标不可达、运动轨迹迂回反复和路径长度过长等问题.传统的人工势场法不能根据环境具体信息对斥力系数进行调整,而现有的改进方法不能在自适应调整斥力系数的同时兼顾规划效果和规划时长.针对以上问题,提出了一种基于深度学习的无人机自适应斥力系数路径规划方法.首先通过融合遗传算法与人工势场法找出在特定环境下最合适的斥力系数样本集,其次利用该样本集训练残差神经网络,最后通过残差神经网络计算适应环境的斥力系数,进而使用人工势场法进行路径规划.仿真实验表明,该方法在一定程度上解决了人工势场法规划中目标不可达、运动轨迹迂回反复和路径长度过长等问题,规划效果和规划时长方面均有优异表现,能很好地满足无人机路径规划中对当前环境的自适应要求和快速规划的要求.     

改进残差网络的猴痘病毒皮肤病变分类

目前猴痘病毒在全球范围内传播,这种病毒在临床上与其他皮肤疾病难以区分,特别是天花病毒和水痘病毒.在确定性聚合酶链式反应技术和其他生物检测技术还没有完全成熟的情况下,通过计算机辅助诊断技术检测猴痘病毒皮肤病变是一种可行的方法,因此提出了一种基于残差网络的猴痘病毒皮肤病变分类算法.该算法以残差网络为基本框架,结合深度可分离卷积和轻量化注意力,在降低模型计算量与复杂度的同时,也提高了模型的分类性能.实验结果表明,该算法对猴痘病毒皮肤病变表现出较好的分类性能,对猴痘皮肤病变的分类准确率、召回率和精度分别为97.3%, 96.8%和97.2%,且均优于实验中所对比的常见分类模型和其他研究方法.     

基于深度残差胶囊网络与注意力机制的加密流量识别方法

随着用户安全意识的提高和加密技术的发展,加密流量已经成为网络流量中的重要部分,识别加密流量成为网络流量监管的重要部分。基于传统深度学习模型的加密流量识别方法存在效果差、模型训练时间长等问题。针对上述问题,提出了一种基于深度残差胶囊网络模型（DRCN,deep residual capsule network）的加密流量识别方法。原始胶囊网络通过全连接形式堆叠导致模型耦合系数变小,无法搭建深层网络模型。针对上述问题,DRCN模型采用三维卷积算法（3DCNN）动态路由算法代替全连接动态路由算法,减少了每个胶囊层之间传递的参数,降低了运算复杂度,进而构建深层胶囊网络,提高识别的准确率和效率;引入通道注意力机制为不同的特征赋予不同的权重,减少无用特征对识别结果的影响,进一步增强模型特征提取能力;将残差网络引入胶囊网络层,搭建残差胶囊网络模块缓解了深度胶囊网络的梯度消失问题。在数据预处理方面,截取的数据包前784byte,将截取的字节转化成图像输入到DRCN模型中,该方法避免了人工特征提取,减少了加密流量识别的人工成本。在ISCXVPN2016数据集上的实验结果表明,与效果最好的BLSTM模型相比,DRCN模型的准确率提高了5.54%,模型的训练时间缩短了232s。此外,在小数据集上,DRCN模型准确率达到了94.3%。上述实验结果证明,所提出的识别方案具有较高的识别率、良好的性能和适用性。