声纹信号-图形差分场增强和多头自注意力机制的变压器工作状态辨识方法*

为提升电力变压器工作状态的智能监测水平,提出声纹信号-图形差分场增强和多头自注意力机制的变压器工作状态辨识方法。基于图形差分场技术将声纹信号映射为二维图像,再借助多头注意力机制的视觉转换器实现图像信息的深层挖掘与状态辨识,采用梯度加权类激活映射实现分类结果的可解释性分析。搭建了包含变压器4种典型工作状态下的实验模拟测试系统平台,实验结果表明：所提方法不仅能够有效表征变压器声纹信号的状态特征,且分类辨识精度相较于“时频图+引入多头注意力机制的变换网络”与“图形差分场+引入残差模块的卷积神经网络”的常规方法有显著提升,提升约6%,同时也具备较好的鲁棒性,可为电气设备的故障检测研究提供一定参考。     

Hepatoprotective Effects of Ixeris chinensis on Nonalcoholic Fatty Liver Disease Induced by High-Fat Diet in Mice: An Integrated Gut Microbiota and Metabolomic Analysis

Ixeris chinensis (Thunb.) Nakai (IC) is a folk medicinal herb used in Mongolian medical clinics for the treatment of hepatitis and fatty liver diseases even though its pharmacological mechanism has not been well characterized. This study investigated the hepatoprotective mechanism of IC on mice with nonalcoholic fatty liver disease (NAFLD) by integrating gut microbiota and metabolomic analysis. A high-fat diet (HFD) was used to develop nonalcoholic fatty liver disease, after which the mice were treated with oral IC (0.5, 1.5 and 3.0 g/kg) for 10 weeks. HFD induced NAFLD and the therapeutic effects were characterized by pathological and histological evaluations, and the serum indicators were analyzed by ELISA. The gut microbial and metabolite profiles were studied by 16S rRNA sequencing and untargeted metabolomic analysis, respectively. The results showed that the administration of IC resulted in significant decreases in body weight; liver index; serum biomarkers such as ALT, TG, and LDL-C; and the liver inflammatory factors IL-1β, IL-6, and TNF-α. The 16S rRNA sequencing results showed that administration of IC extract altered both the composition and abundance of the gut microbiota. Untargeted metabolomic analysis of liver samples detected a total of 212 metabolites, of which 128 were differentially expressed between the HFD and IC group. IC was found to significantly alter the levels of metabolites such as L-glutamic acid, pyridoxal, ornithine, L-aspartic acid, D-proline, and N4-acetylaminobutanal, which are involved in the regulation of glutamine and glutamate, Vitamin B6 metabolism, and arginine and proline metabolic pathways. Correlation analysis indicated that the effects of the IC extract on metabolites were associated with alterations in the abundance of Akkermansiaceae, Lachnospiraceae, and Muribaculaceae. Our study revealed that IC has a potential hepatoprotective effect in NAFLD and that its function might be linked to improvements in the composition of gut microbiota and their metabolites.     

Interpretation of Strengthening Mechanism of Densified Wood from Supramolecular Structures

In this study, densified wood was prepared by hot pressing after partial lignin and hemicellulose were removed through alkaline solution cooking. The tensile strength and elastic modulus of densified wood were improved up to 398.5 MPa and 22.5 GPa as compared with the original wood, and the characterization of its supramolecular structures showed that the crystal plane spacing of the densified wood decreased, the crystallite size increased, and the maximum crystallinity (CI) of cellulose increased by 15.05%; outstandingly, the content of O(6)H⋯O(3′) intermolecular H-bonds increased by approximately one-fold at most. It was found that the intermolecular H-bond content was significantly positively correlated with the tensile strength and elastic modulus, and accordingly, their Pearson correlation coefficients were 0.952 (p < 0.01) and 0.822 (p < 0.05), respectively. This work provides a supramolecular explanation for the enhancement of tensile strength of densified wood.     

Role of Natural Compounds and Target Enzymes in the Treatment of Alzheimer’s Disease

Alzheimer’s disease (AD) is a progressive neurological condition. The rising prevalence of AD necessitates the rapid development of efficient therapy options. Despite substantial study, only a few medications are capable of delaying the disease. Several substances with pharmacological activity, derived from plants, have been shown to have positive benefits for the treatment of AD by targeting various enzymes, such as acetylcholinesterase (AChE), butyrylcholinesterase (BuChE), β-secretase, γ-secretase, and monoamine oxidases (MAOs), which are discussed as potential targets. Medicinal plants have already contributed a number of lead molecules to medicine development, with many of them currently undergoing clinical trials. A variety of medicinal plants have been shown to diminish the degenerative symptoms associated with AD, either in their raw form or as isolated compounds. The aim of this review was to provide a brief summary of AD and its current therapies, followed by a discussion of the natural compounds examined as therapeutic agents and the processes underlying the positive effects, particularly the management of AD.     

基于知识蒸馏与改进ViT网络的花卉图像细粒度分类

由于自然条件下拍摄的花卉图像背景复杂,而且其存在类内差异性大和类间相似性高的问题,现有主流方法仅依靠卷积模块提取花卉的局部特征难以实现准确的细粒度分类。针对上述问题,本文提出了1种高精度、轻量化的花卉分类方法(ConvTrans-ResMLP),通过结合Transformer模块和残差MLP(multi-layer perceptron)模块实现对花卉图像的全局特征提取,并在Transformer模块中加入卷积计算使得模型仍保留提取局部特征的能力;同时,为了进一步将花卉分类模型部署到边缘设备中,本研究基于知识蒸馏技术实现对模型的压缩与优化。实验结果表明,本文所提出的方法在Oxford 17、Oxford 102和自制的Flowers 32数据集上的准确率分别达98.62%、97.61%和98.40%;知识蒸馏后本文的轻量化模型的大小约为原来的1/18,而准确率仅下降2%左右。因此,本研究能较好地提升边缘设备下花卉细粒度分类的效率,对促进花卉培育的自动化发展具有切实意义。     

融合注意力线性特征多样化的DR分级模型

糖尿病视网膜病变(diabetic retinopathy, DR)是目前人类的主要致盲疾病之一。针对DR数据集中样本类间差异小和类分布不均衡等制约分级性能提高的问题,本文提出一种融合注意力线性特征多样化(fusion of attention linear feature diversification, FALFD)的分级算法。该算法首先用改进的Res2Net残差网络作为模型骨干来增大感受野,进一步提高网络捕捉特征信息的能力;其次引入自适应特征多样化模块(adaptive feature diversification module, AFDM)对眼底图像可分辨的微小病理特征进行识别,获得具有高语义信息的局部特征,避免单一特征区域的限制,进而提高分级准确度;再后利用双线性注意力融合模块(bilinear attention fusion module, BAFM)增加可判别区域特征的网络权重占比;最后采用正则化焦点损失(focal loss, FL)进一步提升算法的分类性能。在IDRID数据集上,灵敏度和特异性分别为94.20%和97.05%,二次加权系数为87.83%;在APTO...     

基于多尺度一致性与注意力机制的视网膜血管分割网络

针对现有的视网膜血管分割方法存在对微血管和毛细血管的分割能力不足,导致血管断连和末端血管漏分,造成视网膜血管分割性能不佳的问题,本文提出一种基于多尺度一致性与注意力机制的视网膜血管分割网络(multi-scale consistency and attention mechanism U-Net, MCAU-Net)。首先,该网络在瓶颈特征层嵌入注意力细化模块(attention refinement module, ARM),能有效细化瓶颈层冗余的特征,抑制背景等无关像素的权值。其次,将上下文特征融合模块(context fusion module, CFM)与传统的跳跃连接相结合,以此补充在特征提取过程中逐渐丢失的信息,加强网络对微血管和毛细血管的构建能力。最后,基于网络的多尺度输出设计了一种多尺度一致性的训练方式,以增强网络对不同尺度特征的敏感性。在DRIVE和CHASE＿DB1公开数据集上进行的对比实验表明本文网络具有良好的分割性能。     

基于SOANet的圆锥角膜辅助诊断

亚临床期圆锥角膜(subclinical keratoconus, subkc)发病隐匿,现有医疗设备诊断存在局限性,所以提出一种诊断亚临床期圆锥角膜的检测方法是十分必要的。有研究发现圆锥角膜(Keratoconus, kc)力学性能的改变早于形态学,因此从角膜生物力学的角度筛查亚临床期的圆锥角膜更加符合临床实际。本文运用角膜生物力学特征,以点云数据作为网络输入数据,将SO-Net (self-organizing network)和自注意力(self-attention, SA)机制结合构建SOANet,对圆锥角膜、亚临床期圆锥角膜和正常角膜进行分类。首先,利用可视化生物力学分析仪(corneal visualization Scheimpflug technology, Corvis ST)拍摄角膜受力形变视频,对其进行处理得到点云数据集,接着对点云数据进行增强处理,使3种类型的角膜数据量分布均衡。然后按照3∶1的比例划分训练集和测试集,分别对角膜进行二分类和三分类。最终模型在二分类和三分类测试集上的准确率分别达到98.3%和91.26%,即有效识别亚临床期圆锥角膜和圆锥角膜。实验...     

融合注意力机制的轻量级摔倒检测

摔倒检测大都依靠传感器设备,此类方法受设备自身和环境因素影响较大,常常无法发挥该有的作用,同时,基于视觉的方法往往实时性较差,鲁棒性不强。针对上述问题,本文提出了一种鲁棒性强、能有效部署在嵌入式设备上的轻量化摔倒检测算法。以YOLOv5为基准模型,首先,融合轻量级注意力机制模块,使网络更关注要识别的目标区域,增强网络的识别精度。其次,使用模型压缩方法对模型进行剪枝,减小模型体积和计算量,使模型轻量化,以提高推理速度和便于部署在嵌入式设备中。最后,对剪枝后的模型进行知识蒸馏,在不提升模型复杂度的前提下提升模型的检测精度。实验结果表明：本文模型相较于基准模型,mAP增加了1.7%,召回率提高了1.2%,模型体积减小了79.1%,浮点运算量降低了70.9%。将本文模型部署在嵌入式设备Jetson Nano上,检测速率达到13.2 frame/s,基本满足实时性摔倒检测的要求。     

用户初始服务密码管理方式

本文对用户初始服务密码管理困难的原因进行了深入的分析,提出了将初始服务密码印制在SIM卡涂层下的新型管理方式。