A Normalizing Flow-Based Bidirectional Mapping Residual Network for Unsupervised Defect Detection

Unsupervised methods based on density representation have shown their abilities in anomaly detection,but detection performance still needs to be improved.Specifically,approaches using normalizing flows can accurately evaluate sample distributions,mapping normal features to the normal distribution and anomalous features outside it.Consequently,this paper proposes a Normalizing Flow-based Bidirectional Mapping Residual Network(NF-BMR).It utilizes pre-trained Convolutional Neural Networks(CNN)and normalizing flows to construct discriminative source and target domain feature spaces.Additionally,to better learn feature information in both domain spaces,we propose the Bidirectional Mapping Residual Network(BMR),which maps sample features to these two spaces for anomaly detection.The two detection spaces effectively complement each other’s deficiencies and provide a comprehensive feature evaluation from two perspectives,which leads to the improvement of detection performance.Comparative experimental results on the MVTec AD and DAGM datasets against the Bidirectional Pre-trained Feature Mapping Network(B-PFM)and other state-of-the-art methods demonstrate that the proposed approach achieves superior performance.On the MVTec AD dataset,NF-BMR achieves an average AUROC of 98.7%for all 15 categories.Especially,it achieves 100%optimal detection performance in five categories.On the DAGM dataset,the average AUROC across ten categories is 98.7%,which is very close to supervised methods.

ASYMPTOTICS OF THE RESIDUALS DENSITY ESTIMATION IN NONPARAMETRIC REGRESSION UNDER m(n)-DEPENDENT SAMPLE

Let Y_i=M(X_i)+ei, where M(x)=E(Y|X=x) is an unknown realfunction on B(? R), {(X_1,Y_i)} is a stationary and m(n)-dependent sample from(X, Y), the residuals {e_i} are independent of {X_i} and have unknown common densityf(x). In [2] a nonparametric estimate f_n(x) for f(x) has been proposed on the basisof the residuals estimates. In this paper, we further obtain the asymptotic normalityand the law of the iterated logarithm of f_n(x) under some suitable conditions. Theseresults together with those in [2] bring the asymptotic theory for the residuals densityestimate in nonparametric regression under m(n)-dependent sample to completion.

Experimental Studies on Cyclic Shear Behavior of Steel-Clay Interface Under Constant Normal Load

The degradation of the shear stress between pile-clay interface caused by undrained cyclic jacking affects the jacking force.A series of large displacement monotonic shear,cyclic shear and post-cyclic monotonic steel plate-clay interface shear te sts were performed under the constant normal load(CNL)condition to inve stigate the effects of normal stre ss,cyclic amplitude,and number of cycles on a steel plate-clay interface using the GDS multi-function interface shear tester.Based on the experimental results,in monotonic shear tests,change of shear stress took place in the specimen,the shear stress rapidly reached the peak value at shear displacement of 1 mm,and then abruptly decreased to the residual value.In cyclic shear te sts,accumulated displacement was a better parameter to describe the soil degradation characteristics,and the degradation degree of shear stress became greater with the increasing of normal stress and accumulated displacement.Shear stress in post-cyclic monotonic shear tests did not generate a peak value and was lower than that in monotonic shear tests under the same normal stress.The soil was completely disturbed and reached the residual strength when the cumulative displacement approached 6 m.An empirical equation to evaluate shear stress degradation mechanism was formulated and the procedure of parameter identification was presented.

融合Inception V1-CBAM-CNN的轴承剩余寿命预测模型

针对现有的滚动轴承剩余寿命(RUL)预测方法精度低、轴承健康指标(HI)构建困难等问题,提出了一种基于卷积神经网络(CNN)并融合Inception V1模块和卷积注意力机制模块(CBAM)的滚动轴承RUL预测模型。首先,在CNN中添加了CBAM机制,并进行了加权处理,在通道和空间维度对重要特征进行了强化,对次要特征进行了抑制,通过添加改进的InceptionV1模块,提高了CNN通道间信息交互水平,全面提取了退化特征;然后,进行了网络优化,采用全局最大池化(GMP)方法对模型进行了简化,采用Dropout和批量归一化(BN)方法,避免了过拟合,提高了精度,且克服了训练时出现的梯度消失问题;最后,对数据进行了处理,将降噪后的信号重组为三维张量,将其作为HI,构建了退化标签,引入了评价指标,采用PHM2012轴承数据集进行了实验验证,在3种工况下将其与深度神经网络(DNN)、CNN方法、结合注意力机制的残差网络方法(ResNet)进行了对比。研究结果表明:该方法在变负载条件下的平均RMSE为0.033,较其他方法的RMSE值分别降低了86%、78%和69%,在预测精度和泛化能力方面具有明显优势。

一种基于多跳注意残差网络的调制识别算法

为了进一步提升通信信号调制识别的准确率,在ResNet网络的基础上提出一种基于多跳注意残差网络(Multi-skip Attention Residual Network,MARN)的调制识别方法。该方法利用提取不同特征的卷积核进行多跳连接构建3种残差块,进而构建多跳残差网络,提取信号的时域特征;加入CBAM(Convolutional Block Attention Module)注意力机制自适应地调整通道权重,加强信号特征的表征能力;采用自适配归一化(Switchable Normalization,SN)加速网络收敛;加入丢弃率为0.3的AlphaDropout层,提高算法的拟合能力,最终实现对通信信号端到端的分类识别。在RadioML2018.01a数据集上仿真实验,结果表明在信噪比为-10~15 dB下,MARN网络平均识别率达到63.3%,较ResNet网络的平均识别率提升3.7%。

自硬-微波加热复合硬化硅酸盐粘结剂砂性能研究

针对自硬硅酸盐粘结剂砂用于铸铁件时存在硬化速度慢、对环境湿度敏感、砂型(芯)强度低、溃散性差四个问题,本文选择一种新型硅酸盐粘结剂,以陶瓷砂为原砂,在采用固化剂自硬的基础上,还通过微波加热以及添加粉末促硬剂,进行了复合硬化工艺的研究。首先,研究了微波加热工艺参数对型砂试样性能的影响,确定了优化微波加热工艺:加热功率为900 W、加热时间为180 s;而后,选用了三种粉末促硬剂进行优化,得到的最佳粉末添加剂优化方案为粘结剂2.2%、微硅粉0.2%、氧化铝0.2%、石英粉0.06%。结果表明,该组优化方案相比较不加入粉末添加剂,使砂型(芯)1 h强度提高至1.86 MPa,24 h强度提高至2.17 MPa,98%RH吸湿强度提高至1.45 MPa,800℃残留强度降低至0.75 MPa,其常温强度值和残留强度值分别达到和接近自硬树脂砂的水平。

新型硅酸盐自硬粘结剂砂性能的研究

以有机酯为固化剂的普通水玻璃自硬砂具有生产环境好、产生的有害气体少的优势,目前广泛应用于铸钢件的生产。但以硅砂为原砂的水玻璃自硬砂用于生产结构复杂、质量要求高的铸铁件,与树脂砂工艺相比,还存在硬化速度慢,型芯砂强度较低,使用性能对环境湿度敏感等问题。为此,本文以一种新型硅酸盐粘结剂为对象,选用球形陶瓷砂,在固化剂自硬工艺基础上,添加促硬剂和多种粉末改性剂,重点研究了常温强度和高温残留强度的变化规律。正交试验获得的复合粉末改性剂优化配方为:陶瓷砂∶促硬剂∶氧化锆∶氮化硅=1000∶4∶2∶1,粘结剂加入量2.2%,固化剂占粘结剂质量的20%,试样1h强度为0.98MPa,24h强度1.59MPa,800℃残留强度0.62MPa,其常温强度已达到自硬树脂砂的性能指标。

Identifying the best common factor model via exploratory eactor analysis

Currently,there is no solid criterion for judging the quality of the estimators in factor analysis.This paper presents a new evaluation method for exploratory factor analysis that pinpoints an appropriate number of factors along with the best method for factor extraction.The proposed technique consists of two steps:testing the normality of the residuals from the fitted model via the Shapiro-Wilk test and using an empirical quantified index to judge the quality of the factor model.Examples are presented to demonstrate how the method is implemented and to verify its effectiveness.

Impact of climate and human activity on NDVI of various vegetation types in the Three-River Source Region, China

The Three-River Source Region(TRSR)in China holds a vital position and exhibits an irreplaceable strategic importance in ecological preservation at the national level.On the basis of an in-depth study of the vegetation evolution in the TRSR from 2000 to 2022,we conducted a detailed analysis of the feedback mechanism of vegetation growth to climate change and human activity for different vegetation types.During the growing season,the spatiotemporal variations of normalized difference vegetation index(NDVI)for different vegetation types in the TRSR were analyzed using the Moderate Resolution Imaging Spectroradiometer(MODIS)-NDVI data and meteorological data from 2000 to 2022.In addition,the response characteristics of vegetation to temperature,precipitation,and human activity were assessed using trend analysis,partial correlation analysis,and residual analysis.Results indicated that,after in-depth research,from 2000 to 2022,the TRSR's average NDVI during the growing season was 0.3482.The preliminary ranking of the average NDVI for different vegetation types was as follows:shrubland(0.5762)>forest(0.5443)>meadow(0.4219)>highland vegetation(0.2223)>steppe(0.2159).The NDVI during the growing season exhibited a fluctuating growth trend,with an average growth rate of 0.0018/10a(P<0.01).Notably,forests displayed a significant development trend throughout the growing season,possessing the fastest rate of change in NDVI(0.0028/10a).Moreover,the upward trends in NDVI for forests and steppes exhibited extensive spatial distributions,with significant increases accounting for 95.23%and 93.80%,respectively.The sensitivity to precipitation was significantly enhanced in other vegetation types other than highland vegetation.By contrast,steppes,meadows,and highland vegetation demonstrated relatively high vulnerability to temperature fluctuations.A further detailed analysis revealed that climate change had a significant positive impact on the TRSR from 2000 to 2022,particularly in its northwestern areas,accounting for 85.05%of the total area.Meanwhile,human activity played a notable positive role in the southwestern and southeastern areas of the TRSR,covering 62.65%of the total area.Therefore,climate change had a significantly higher impact on NDVI during the growing season in the TRSR than human activity.

Impacts of climate change and human activities on vegetation dynamics on the Mongolian Plateau, East Asia from 2000 to 2023

The Mongolian Plateau in East Asia is one of the largest contingent arid and semi-arid areas of the world.Under the impacts of climate change and human activities,desertification is becoming increasingly severe on the Mongolian Plateau.Understanding the vegetation dynamics in this region can better characterize its ecological changes.In this study,based on Moderate Resolution Imaging Spectroradiometer(MODIS)images,we calculated the kernel normalized difference vegetation index(kNDVI)on the Mongolian Plateau from 2000 to 2023,and analyzed the changes in kNDVI using the Theil-Sen median trend analysis and Mann-Kendall significance test.We further investigated the impact of climate change on kNDVI change using partial correlation analysis and composite correlation analysis,and quantified the effects of climate change and human activities on kNDVI change by residual analysis.The results showed that kNDVI on the Mongolian Plateau was increasing overall,and the vegetation recovery area in the southern region was significantly larger than that in the northern region.About 50.99%of the plateau showed dominant climate-driven effects of temperature,precipitation,and wind speed on kNDVI change.Residual analysis showed that climate change and human activities together contributed to 94.79%of the areas with vegetation improvement.Appropriate human activities promoted the recovery of local vegetation,and climate change inhibited vegetation growth in the northern part of the Mongolian Plateau.This study provides scientific data for understanding the regional ecological environment status and future changes and developing effective ecological protection measures on the Mongolian Plateau.

基于水下场景先验的水下图像增强方法研究

针对水体光线吸收与散射作用引起的图像模糊、低对比度和颜色失真等问题,提出一种基于水下场景先验的水下图像增强方法。首先利用水下场景的先验知识,结合水下成像物理模型和水下场景的光学特性,利用10种预定义衰减系数合成涵盖不同类型和退化水平的水下图像数据集;然后利用初始残差和密集级联,设计一类轻量级卷积神经网络(convolutional neural networks,CNN)模型增强水下图像,结合基于轻量级的网络结构和有效的训练数据,可减少增强模型的计算量并有效改善水下退化图像的视觉质量;最后采用归一化的后处理过程进一步提升图像增强的效果。仿真实验结果表明,所提方法可行有效,可应用到不同的真实水下场景,具有较强的鲁棒性与有效性。

基于Faster-RCNN的绝缘子缺陷检测

针对传统人工检测绝缘子缺陷效率低的问题,提出一种基于Faster-RCNN的绝缘子缺陷检测方法。首先对航拍的绝缘子缺陷图片进行数据增强,其次算法中使用残差网络结构并引入注意力机制,提升检测效果的同时降低了模型复杂性,使用组归一化方式代替批归一化方式,最后用Soft-NMS代替NMS进行结果优化。试验结果表明,改进后算法的精确率达到90.3%,与改进前相对比精确率提升了14.7%,使绝缘子缺陷检测的有效性与可靠性得到了提升。

基于数理统计方法的水质总氮校准曲线残差值检验

(目的)探讨水质总氮校准曲线残差值的正态性、独立性、同方差性。(方法)一年内对不同浓度的硝酸盐氮标准使用液进行了6次测定,对测定结果进行校准曲线拟合,计算残差值。通过QQ图及Anderson—Darling(AD)法检验残差值的正态性;通过图示法及DW法检验残差值的独立性;通过残差图及等级相关系数法检验残差值的同方差性。结果表明,该曲线残差值满足正态性,不满足独立性和同方差性。(结论)使用普通最小二乘法对数据进行曲线拟合需建立在一系列假定条件上,因此实际工作中不能盲目默认相关假定条件成立,应将数理统计理论与化验检测实际相结合,保障曲线拟合的可靠性。

气候变化和人类活动对华北地区植被NDVI的影响研究

归一化植被指数(Normalized Differential Vegetation Index,NDVI)是反映植被生长状态的重要指标,是反映陆地生态环境状况的“指示器”。华北地区地处我国的政治和文化中心,土地覆盖类型复杂多样,是我国重要的粮食生产地,同时受到气候暖干化及加剧的人类活动影响,华北地区的植被生态变得十分脆弱。本研究基于卫星资料NOAACDRAVHRR NDVI和气象数据资料,采用趋势分析、偏相关分析和残差趋势分析等方法,探究了1982-2019年华北地区NDVI的时空变异特征及其对气候变化和人类活动的响应。研究结果表明:(1)1982-2019年华北地区春季、夏季、秋季和生长季的植被NDVI呈显著上升趋势,空间异质性强,其中夏季和生长季的增长速率最快为0.024(10a)^(-1),显著增加的区域面积占比分别为57.35%和58.10%。(2)华北地区春季、夏季和生长季NDVI与降水呈显著正相关关系,秋季NDVI主要受气温的影响,夏季NDVI同时受到气温、降水和相对湿度的积极影响。(3)气候变化和人类活动对华北地区植被的生长影响具有区域差异性,在植被改善区,气候变化的相对作用为45.64%,人类活动的相对作用为54.36%;在植被退化区,气候变化的相对作用为32.66%,人类活动的相对作用为67.34%。(4)不同土地利用类型中,华北地区森林和农田的植被生长较快,其植被改善主要受人类活动的影响,人类活动的相对作用分别为66.07%和60.82%,草地植被的退化也主要受人类活动的影响,相对作用为69.48%,人类活动对华北地区植被的重要影响主要源于我国近几十年来三北防护林等人类重大生态工程的建设以及城市扩张、人口激增的影响,该研究成果也对华北地区生态屏障的建设以及生态环境保护提供了重要理论支撑。

多参数流式细胞术识别与急性髓系白血病微小残留病具有类似表型的再生细胞

目的探讨如何利用多参数流式细胞术(MFC)鉴别与急性髓系白血病(AML)微小残留病(MRD)具有类似表型的正常细胞。方法回顾性分析2020年3月-2022年4月在我院利用MFC检测的AML患者MRD骨髓标本157例次,识别容易被误认为是MRD的正常细胞。结果AML患者治疗后再生的骨髓样本中会出现如下易误判为MRD的正常细胞群:CD117 dim CD56^(+)CD7^(+)CD45^(str)自然杀伤(NK)细胞、CD19 dim CD56^(+)CD7^(+)CD45^(str) NK细胞、CD300e^(+)HLA-DR^(+)CD14^(part) CD64^(part)非经典单核细胞;这些细胞群占骨髓有核细胞比例的中位值分别是0.084%(范围:0~0.6200%)、0%(范围:0~0.2134%)、0.1549%(范围:0~2.0940%)。结论MFC检测AML患者MRD过程中,应避免将治疗后骨髓再生的正常细胞,误判为残留病。

基于人工神经网络的高血压预测模型

为准确预测高血压患者,文章提出了一种基于人工神经网络(artificial neural network,ANN)的高血压预测模型.该模型在原始的ANN模型中引入了批归一化层(batch normalization,BN)和残差连接(residual connection),以改进原始ANN模型所存在的缺陷.实验表明,该模型的收敛速度显著高于原始模型,且可有效加快模型的训练过程.研究结果可为高血压的早期预测和干预提供参考.

基于状态空间和NR-LMS的结构参数辨识方法

提出一种基于系统状态空间模型和归一化鲁棒最小均方根(NR-LMS,Normalized Robust Least Mean Square)理论的动力学结构参数辨识方法.利用系统的输入-输出数据建立其Hankel-Toeplitz模型,利用NR-LMS算法得到该模型参数的估计并求得系统的Hankel矩阵,对Hankel矩阵进行奇异值分解即可确定系统的阶次,进而确定系统状态空间模型的参数.仿真研究和实验结果表明,此方法可以准确、快速地提取出结构的参数,且抗噪能力较强.

NRCNN与角度度量融合的人脸识别方法

常见的卷积神经网络通常使用分类损失来进行可分离的特征学习,在某些情况下存在特征的可区分性不足的问题,而一些改进的方法复杂度较高.为了在较低的复杂性下仍能保证较高的准确率,提出了一种基于嵌套残差卷积神经网络与角度度量的人脸识别方法.首先,设计了一种新颖的基于嵌套残差模块的人脸特征提取网络,通过多特征图融合的方式提取更丰富的特征;其次,使用了一种基于权值标准化的角度度量方法,通过对最后一个全连接层的权值进行标准化的操作来增强特征区分性.在网络训练时,结合上述两种方法使得学习到的特征满足最大类内距离小于最小类间距离的原则。实验表明,该方法在人脸标记数据库上测试准确率达到99.03%,相较于使用分类损失和其他度量学习的方法,该方法仅使用了单个网络并能在保证较高准确率的情况下付出更小的计算代价.

Mechanical Analysis of Viscous-Elastic Fluid Acting on Residual Oil in the Micro Pore

In order to analyze the normal deviatoric stress that viscous-elastic fluid acting on the residual oil under the situation of different flooding conditions and different permeabilities, Viscous-elastic fluid flow equation is established in the micro pore by choosing the continuity equation, motion equation and the upper-convected Maxwell constitutive equation, the flow field is computed by using numerical analysis, the forces that driving fluid acting on the residual oil in micro pore are got, and the influence of flooding conditions, pore width and viscous-elasticity of driving fluid on force is compared and analyzed. The results show that: the more viscous-elasticity of driving fluid increases, the greater the normal deviatoric stress acting on the residual oil increases;using constant pressure gradient flooding, the lager the pore width is, the greater normal deviatoric stress acting on the residual oil will be.

彭水地区残留向斜常压页岩气地震采集实践

中国南方常压页岩气主要分布于四川盆地外围的志留系残留向斜,资源潜力大。区内地质条件复杂,必须利用高信噪比地震资料精细刻画地下的构造特征并准确描述优质页岩的分布规律,提高水平井优质页岩钻遇率与钻井效率。自2011年起,在彭水地区持续开展了地震采集探索与实践。通过系统梳理总结已实施项目的方法、效果及不足,区域噪声特征分析,道距对静校正影响研究,三维地震观测系统退化分析,认为:①二维地震采集剖面五峰—龙马溪组页岩反射波组较清晰,支撑了常压页岩气的选区评价工作;②提出宽方位、低炮点密度、高横向覆盖次数以及中近偏移距信息丰富的三维观测系统设计原则;③三维地震处理剖面信噪比高、波组特征清楚,经实钻验证三维构造成像准确、地层深度预测误差<1%,有力支撑勘探开发。该技术体系可在类似常压页岩气探区推广。