Feature extraction of gear-localized defect using adaptive lifting scheme and local gradient maps

In this paper,the adaptive lifting scheme (ALS) and local gradient maps (LGM) are proposed to isolate the transient feature components from the gearbox vibration signals. Based on entropy minimization rule,the ALS is employed to change properties of an initial wavelet and design adaptive wavelet. Then LGM is applied to characterize the transient feature components in detail signal of decomposition results using ALS. In the present studies, the orthogonal Daubechies 4 (Db 4) wavelet is used as the initial wavelet. The proposed method is applied to both simulated signals and vibration signals acquired from a gearbox for periodic impulses detection. The two conventional methods (cepstrum analysis and Hilbert envelope analysis) and the orthogonal Db4 wavelet are also used to analyze the same signals for comparison. The results demonstrate that the proposed method is more effective in extracting transient components from noisy signals.

Object-Based Burned Area Mapping with Extreme Gradient Boosting Using Sentinel-2 Imagery

The Sentinel-2 satellites are providing an unparalleled wealth of high-resolution remotely sensed information with a short revisit cycle, which is ideal for mapping burned areas both accurately and timely. This paper proposes an automated methodology for mapping burn scars using pairs of Sentinel-2 imagery, exploiting the state-of-the-art eXtreme Gradient Boosting (XGB) machine learning framework. A large database of 64 reference wildfire perimeters in Greece from 2016 to 2019 is used to train the classifier. An empirical methodology for appropriately sampling the training patterns from this database is formulated, which guarantees the effectiveness of the approach and its computational efficiency. A difference (pre-fire minus post-fire) spectral index is used for this purpose, upon which we appropriately identify the clear and fuzzy value ranges. To reduce the data volume, a super-pixel segmentation of the images is also employed, implemented via the QuickShift algorithm. The cross-validation results showcase the effectiveness of the proposed algorithm, with the average commission and omission errors being 9% and 2%, respectively, and the average Matthews correlation coefficient (MCC) equal to 0.93.

Mapping landslide susceptibility at the Three Gorges Reservoir, China, using gradient boosting decision tree,random forest and information value models

This work was to generate landslide susceptibility maps for the Three Gorges Reservoir(TGR) area, China by using different machine learning models. Three advanced machine learning methods, namely, gradient boosting decision tree(GBDT), random forest(RF) and information value(InV) models, were used, and the performances were assessed and compared. In total, 202 landslides were mapped by using a series of field surveys, aerial photographs, and reviews of historical and bibliographical data. Nine causative factors were then considered in landslide susceptibility map generation by using the GBDT, RF and InV models. All of the maps of the causative factors were resampled to a resolution of 28.5 m. Of the 486289 pixels in the area,28526 pixels were landslide pixels, and 457763 pixels were non-landslide pixels. Finally, landslide susceptibility maps were generated by using the three machine learning models, and their performances were assessed through receiver operating characteristic(ROC) curves, the sensitivity, specificity,overall accuracy(OA), and kappa coefficient(KAPPA). The results showed that the GBDT, RF and In V models in overall produced reasonable accurate landslide susceptibility maps. Among these three methods, the GBDT method outperforms the other two machine learning methods, which can provide strong technical support for producing landslide susceptibility maps in TGR.

From the second gradient operator and second class of integral theorems to Gaussian or spherical mapping invariants

By combining of the second gradient operator,the second class of integral theorems,the Ganssian-curvature-based integral theorems and the Gaussian (or spheri- cal)mapping,a series of invariants or geometric conservation quantities under Gaussian (or spherical)mapping are revealed.From these mapping invariants important trans- formations between original curved surface and the spherical surface are derived.The potential applications of these invariants and transformations to geometry are discussed.

GIS-Mapping of Soil Available Plant Nutrients (Potentiality, Gradient, Anisotropy)

The research was elaborated in Kafr El-Dawar area (Egypt northern region) to study the availability of the soil plant nutrients. The research introduced three parameters to comprehensively and carefully describe the availability of the soil plant nutrients: potentiality, gradient and anisotropy. Potentiality defines the categories of soil ability to supply plant nutrients;meanwhile gradient expresses the increasing rate of the availability of the soil plant nutrients. The gradient anisotropy refers to the directions or orientation of the increasing rate of the availability of the soil plant nutrients. The introduced parameters enabled to spatially study the availability of the soil plant nutrients. Analytical data, of soil available phosphorus (P), indicated that P ranged from 0.2 ppm to 11.4 ppm to locate all studied soil samples into the low class of the soil nutritional P ability. This was not the case of available potassium (K), where the soil samples were distributed into three available K soil categories: medium, high, and very high. GIS map of soil P nutritional potentiality for plant (potato), displayed the soil studied area in one category, as low P soil nutritional potentiality to coincide with the analytical data classification. Contrary, the K map classified the soil studied area into three categories of soil K nutritional potentiality: medium, high and excessive. This obviously referred that the individual determination of soil K nutritional potentiality is misleading for interpretation of soil tests because it does care of the spatial distribution of soil available K. Nearly, all soil samples had high available micronutrients that they were located in the high category in both classification of analytical data and GIS maps. GIS gradient maps of the soil available plant nutrients referred that the soil plant nutrients, exception of K, had two gradients: non increasing-slight increasing and build up. Gradient of soil available potassium was classified into four classes: non increasing-slight increasing, build up, moderately increasing and hike. Regardless potassium case, the non increasing-slight increasing gradient class dominated the others. GIS maps of anisotropy soil availability of macronutrients (P and K) generally showed that their gradients mainly increased in two directions: north and south. The incasing directions of soil availability of micronutrients coincided with that of the macronutrients.

基于全变分加权差正则的高光谱图像去噪算法

针对现有全变分模型在高光谱图像中出现的伪影、边缘结构消失等问题,文章提出一种增强型三维全变分加权差正则模型。首先,该模型并非直接将稀疏性强加于梯度映射本身,而是对梯度映射的基矩阵添加稀疏性约束。此外,与一般稀疏约束方法不同的是,为避免由l_(1)范数自身局限性带来的去噪不良影响,利用l_(1)范数与l_(2)范数的全变分加权差(简记为l_(1-2))分别对高光谱图像的空间域与光谱域施加稀疏约束。实验结果表明,该文提出的算法有效避免了伪影的产生以及图像细节丢失的问题,具有更优的去噪效果。

利用MAPGIS制作水土流失基础底图的方法探讨

在水土流失研究中 ,可以通过 MAPGIS对研究区域的地形数据进行空间分析得到研究区域的水土流失基础底图 ,探讨了空间分析中的参数选取的方法 ,以四川盆地川东平行岭谷低山丘陵区为例 ,探索合适的制图参数 。

基于渐近均匀化的梯度板等效刚度数值计算

基于梯度板结构的渐近均匀化方法,简化梯度板微单胞的单胞方程及等效刚度列式,并推导其有限元求解列式,实现梯度加筋板微结构等效刚度的高效数值求解。相比仅能处理矩形单胞的经典数值均匀化方法,该方法可以处理一般平行四边形微结构,具有更好的普适性。数值算例通过比对梯度加筋板及其对应的等效均质板挠度,最大挠度数值相对误差均在5%以内,验证了所提数值方法的正确性及可行性。

基于深度学习的DRFM信号识别

针对数字射频存储器(DRFM)产生信号与源信号之间无法有效区分的问题,运用基于小波变换的同步压缩变换将时域的雷达信号转换为时频图,运用深度学习强大的图像识别能力,实现了基于深度学习的源信号与DRFM信号识别,从而解决了在雷达信号处理中无法有效区分回波信号和DRFM欺骗信号以及在雷达干扰识别中基于DRFM的欺骗干扰难以识别的问题。为了验证深度学习过程的可靠性,通过神经网络可解释性算法对训练结果进行了验证和分析。实验结果表明,相比于识别原始信号,识别DRFM信号神经网络需要用到更多的特征,神经网络判断准确率达到了96.33%,识别精度良好。

一种基于MAP的超分辨率图像重建的快速算法

超分辨率图像重建技术就是通过融合多幅变形、模糊、有噪、频谱混叠的低分辨率降质图像(或视频序列)来重建一幅高质量高分辨率图像.MAP估计算法是一种广泛使用的统计重建方法.针对标准MAP估计算法运算量大的问题提出了两点改进.第1点是当计算梯度时直接计算目标函数的增量,避免了函数值的冗余计算;第2点是采用非精确一维搜索确定步长,避免了运算量庞大的海塞矩阵的计算.实验结果表明,提出的改进在保持重建效果基本不变的前提下,在很大程度上提高了MAP超分辨率图像重建方法的速率,与此同时保证了算法的收敛性.

一种基于Sobel梯度的直方图均衡算法及其在红外图像上的应用

为了能在动态范围压缩的同时增强红外图像的对比度,提出了一种基于Sobel梯度直方图均衡算法(gradient histogram equalization,GHE)。与以往的直方图均衡化(histogram equalization,HE)方法不同,该方法自适应地为图像强梯度的灰阶分配高对比度,保留并增强16 bit图像中更多的细节。随后使用双Gamma映射对映射曲线进行调整,有效地抑制图像亮部的过曝现象,同时提高暗部的细节。该方法相比于传统的直方图均衡化算法在暗区细节处理、过曝抑制、对比度增强等方面都有较好的效果。

基于机器学习的流域尺度森林火灾灾害风险预测

森林是碳库,具有强大的固碳增汇功能,在应对气候变化中发挥着重要作用。然而,由于极端高温的影响,频繁发生可燃物自燃而引发森林火灾,除了影响区域水文大气循环过程以外,也给人类带来严重的人员伤亡和经济损失。现有森林火灾预测研究主要侧重可燃物研究和火灾监测等方面,较少关注大尺度地形、气象和人类活动对森林火灾的影响,但这些也是除可燃物外导致森林火灾发生的主要因素。以嘉陵江流域重庆段为研究区,区域内山地受自然火灾影响严峻。基于地理信息系统叠加地理空间因子与火灾分布点获得数据集,构建4种机器学习模型,测试模型性能,评价最优模型进行森林火灾灾害风险制图。研究结果表明,模型评估指标受试者工作曲线下面积(area under the curve,AUC)平均值为95.0%,模型性能梯度提升决策树最优,AUC值为98.3%。利用梯度提升决策树(gradient boosting decision tree,GBDT)模型预测森林火灾风险对防范大尺度森林火灾具有一定的可行性,对山城避灾规划起到借鉴作用,规划引导降低森林火灾风险,从而维护生态平衡和生态系统碳汇能力。

一种复合域台风云图增强算法分析

台风云图增强是通过增加图像的对比度和色彩,增加台风的可视化,为后续预测路径和图像分割提供准确的图像。文章提出一种有效的增强台风云图对比度的方法,首先对输入图像进行反相操作和梯度处理,获取梯度域和取反的图像信息,增加台风云图的细节信息;其次,使用一个简单有效改进直方图的模型来生成分布函数,增强图像的对比度;最后,通过对增强后的图像进行融合,得到最终的增强台风云图。与几种增强方法相比,文章提出的方法获得了更好的主观结果和客观评价。

基于ResNet的可解释性计算机视觉模型在内镜下内痔评估中的应用

目的为克服深度学习模型黑盒不可解释的缺点,本研究旨在探讨可解释性计算机视觉模型在内镜下内痔诊断及危险分级中的应用。方法收集苏州大学附属第一医院内镜中心的肛齿状线上倒镜图片,分为内痔组和正常组;根据LDRf分级标准,对内痔组进一步分级为Rf0、Rf1及Rf2三组。针对有无内痔、红色征、糜烂、血栓及活动性出血,构建基于ResNet50V2的可解释化模型,并利用江苏大学附属金坛医院内镜中心的内镜图片进行外部验证。使用准确性、敏感性、特异性以及F1值等指标对比可解释化模型与传统深度学习黑盒模型的表现,并与两位不同年资内镜医生进行比较。结果ResNet可解释化模型的准确性为0.957、敏感性为0.978、特异性为0.974,F1值为0.958,其准确性高于黑盒模型的0.938,高年资内镜医生的0.933及低年资医生的0.907。此外,模型采用Grad-CAM方法突出图像中对模型推理依据的区域。结论本研究通过收集内镜下肛齿状线上倒镜图像,构建可解释化计算机视觉模型并进行外部验证,提示该模型在内镜下内痔诊断与评级中表现优于传统深度学习黑盒模型。该模型在未来临床内镜诊疗中具有良好应用前景。

基于场景理解的双波段彩色融合图像质量评价

为了针对具体的视觉任务衡量可见光与红外彩色(夜视)融合图像的综合质量,提出了一种基于场景理解的双波段彩色融合图像综合质量客观评价模型.该模型包括融合图像特征提取器、邻域共生矩阵特征提取器和权重生成器三部分.首先,使用融合图像特征提取器从融合图像中提取像素强度信息;然后,建立邻域共生矩阵特征提取器,从邻域共生矩阵中提取像素空间关系信息;最后,建立权重生成器,使用神经网络模型从梯度图中提取结构信息,将位置信息与结构信息相结合生成权重.实验结果表明,该方法在提取丰富的图像特征基础上,考虑人眼视觉特性,提高了模型预测值与人眼主观感受的一致程度,实现了融合图像综合质量的客观评价.

深度学习在内镜下内痔诊断及危险分级中的应用

目的建立内镜下内痔诊断及危险分级的深度学习模型,探讨人工智能辅助内镜下内痔诊疗的可行性。方法收集该院内镜中心的肛齿状线上倒镜图片,分为内痔组和正常组(A任务);根据LDRf分级标准,将内痔组进一步分为Rf0组、Rf1组和Rf2组(B任务)。选取基于卷积神经网络(CNN)框架的Xception、ResNet和EfficientNet,以及基于Transformer框架的ViT和ConvMixer等5个神经网络,建立针对A、B两项计算机视觉任务的深度学习模型。模型评价指标包括准确率、召回率、精确度、F1值和读片时间。将深度学习模型的读片表现与两位不同年资内镜医生进行比较。结果5种深度学习模型在A与B任务测试集中皆展现出较好的准确性。其中,最优模型为ConvMixer,准确性最高(0.961和0.911),其次为EfficientNet(0.956和0.901),均优于高年资内镜医生(0.952和0.881)和低年资内镜医生(0.913和0.832)。同时,所有深度学习模型在验证集中读片用时均<10 s,速度快于内镜医生(均>300 s)。此外,笔者采用梯度加权分类激活映射(Grad-CAM)方法突出图像中对模型判断较重要的区域。结论建立的内痔诊断及危险分级的深度学习模型,其表现优于内镜医生。基于深度学习的计算机视觉模型可辅助内镜医师进行内痔诊断和分级,展现出潜在的临床应用前景。

基于深度学习的内镜肠道准备评分模型的建立

目的基于深度学习算法构建内镜肠道准备评分的计算机视觉模型。方法收集苏州大学附属第一医院消化内镜中心(600张)及HyperKvasir数据库(1794张)的内镜图片共2394张,根据Boston肠道准备量表完成肠道清洁度评分(0~3分,四分类),按6∶2∶2随机分为训练集(1439张)、验证集(478张)和测试集(477张)。选取3种深度学习网络(DenseNet169、DenseNet121、EfficientNet B3),利用迁移学习方式训练肠道准备分类模型,并采用测试集的混淆矩阵等指标评价模型分类能力,与高、低年资医师的分类能力进行对比。结果成功构建3个基于深度学习的肠道准备分类模型。各模型的分类准确度均较高,平均分类准确度为0.897,近似于低年资内镜医师(0.914),低于高年资内镜医师(0.941)的分类表现。其中,DenseNet169模型表现最好,分类准确度(0.914)及平均精确度(0.892)均为最高。此外,采用梯度加权分类激活映射算法,用热力图形式对模型的分类推理进行可视化呈现。结论运用深度学习算法构建的内镜肠道准备分类模型具有可行性,可通过多中心研究扩大样本来源进一步提高模型的分类及泛化能力。

基于脑电空间域表征可视化的情感识别研究

鉴于情感脑电蕴含丰富的空间模式特征,提出一种基于二维空间域表征可视化的情感识别方法。首先,提取多通道脑电Gamma频段的微分熵(Differential Entropy,DE)特征并根据导联位置映射至9×9的二维空间进行拓扑重构,使用三次插值方法进一步提高空间域特征图的分辨率;然后,针对性地设计了一种深度残差网络(Residual Network,ResNet)模型作为情感脑电解码器对情感脑电信号(Electroencephalogram,EEG)进行深层抽象特征的自动提取和端到端分类;最后,通过梯度加权类激活映射(Gradient-weighted Class Activation Mapping,Grad-CAM)方法对输入特征图进行可解释性分析,依据热力图分布定位对特定情感状态识别具有较大贡献的空间脑区。在SEED数据集上进行了相关情感识别实验,三种情感类别分类平均准确率为94.88%,达到了较先进的性能。

基于梯度加权类激活热力图的卷积神经网络故障诊断模型鲁棒性分析

深度学习近年来在故障诊断领域受到广泛应用,但基于深度学习的故障诊断模型缺乏工程上的物理解释性,难以保证其故障诊断结果的稳定性。以轴承为例,建立了以小波时频图像为故障诊断依据的卷积神经网络模型(convolutional neural network,CNN),提出了一种基于梯度加权类激活热力图(gradient-weighted class activation map,Grad-CAM)的网络模型鲁棒性分析方法,并利用美国凯斯西储大学(Case Western Reserve University,CWRU)轴承数据集进行验证。首先,将故障直径轴承数据以不同方式混合并训练大、小多个模型。其次,利用Grad-CAM方法,建立时频区域与故障模式之间的联系。最后,利用其他工况下的轴承故障数据,以及含噪数据进行测试,并根据结果结合模型最注重的时频区域进行分析。结果表明,基于深度学习的轴承故障诊断模型在参数较少时更加注重低频区域,并能使其具有更好的鲁棒性。