简介:以卷积神经网络为代表的深度学习算法在医学影像分析领域正引起广泛美注,并取得了令人惊叹的进步。为了进一步提高卷积神经网络在计算机辅助筛查肺结节应用的准确率,本文设计了2种改良的深度卷积神经网络,这些改进加快了神经网络的训练速度.有效地防止了算法的过拟合。相比只采用二维卷积核的其他检测模型,该模型能够有效地学习到CT影像三维重建后的图像特征。通过实验,改进的检测模型在LUNAl6数据集上的准确率明显好于其他模型,这种网络结构也可用于医学影像领域中其他三维图像的检测场景。最后,构建了一套适用于远程医疗的“计算机辅助肺癌筛查与诊断系统”,该系统能够自动检测出CT影像中肺结节,并给出结节的良恶性概率评估。通过该系统的应用,可以有效缓解放射科医生超高的劳动强度,提高阀片效率,服务更多患者;减少漏诊和误诊发生的次数,有助于提高肺结节的诊断准确率;从而促进我国肺癌早筛工作的推广。
简介:Mechanicalandphysicalproperties,suchastensilestrength,elongationatbreak,modulusofelasticity,ShoreDhardness,meltflowrate(MFR),andelectricalandthermalconductivitiesofcompositeswithhighdensitypolyethylenematrixreinforcedwithAlpowderswereinvestigatedexperimentally.Measurementsofthemechanicalandphysicalpropertieswereperformeduptoareinforcingcomponentconcentrationof30%volumeAlpowderandcomparedwithmathematicalmodelsfromtheliterature.Theobtainedresultshaveshownthatexperimentaldatawereingoodagreementwiththeoreticaldata.Theultimatetensilestrength(UTS)andelongationatbreakdecreasedwithincreasingAlpowdercontent,whichwasattributedtotheintro-ductionofdiscontinuitiesinthepolymerstructure,andmodulusofelasticityincreasedwithincreasingAlcontent.Thecompositepreparationconditionsallowedtheformationofarandomdistributionofmetallicparticlesinthepolymermatrixvolumeforsystemhighdensitypolyethylene-Al(HDPE-Al).TherewasaclusterformationofAlparticlesathigherAlcontentsinthepolymermatrix.ElectricalandthermalconductivityvaluesofHDPE-AlcompositeswerehigherthanpureHDPEvalues.
简介:TheconductivitybehaviorofAl(OH)3-acrylamidehybridpolyacrylamide(hybridPAAm)indistilledwaterwasstudied.Adiscontinuityphenomenonoftheconductivity(k)versusconcentration(c)curveofthehybridPAAminacertainconcentrationregimeisfound.ThisphenomenonisdependentonthemolecularweightofthehybridPAAmandontheparticlesizeandcontentoftheAl(OH)3colloidinthehybridPAAm.ThisphenomenonwasaccountedforassumingionizationofthehybridPAAm.
简介:Fatiguepropertiesofage-hardenedAlalloy2017-T4underultrasonicloadingfrequency(20kHz)wereinvestigatedandcomparedwiththeresultsunderconventionalloadingofrotatingbending(50Hz).Thegrowthofacrackretardedatabout500μminsurfacelengthunderultrasonicloading,whileatabout20μmunderrotatingbending.Althoughstriationsbeingatypicalfracturemechanismwereobservedunderconventionalloading,mostoffracturesurfacewascoveredwithmanyfacetsunderultrasonicloading.Thesefacetswerealsoobservedunderrotatingbendinginnitrogengas.Thedifferenceingrowthmechanismdependingontheloadingfrequencyandtheretardationofacrackgrowthunderultrasonicloadingmaybecausedbytheenvironmentatthecracktipduetohighcrackgrowthrateunderultrasonicloading.
简介:在广义系统故障诊断过程中,若系统动态模型中存在不确定性,传统的无迹卡尔曼滤波算法将失去其传感器故障估计精度。为解决该问题,提出一种改进的强跟踪卡尔曼滤波算法以实现广义连续-离散系统的传感器故障诊断及隔离。首先,提出基于多重渐消因子的强跟踪滤波算法以实现动态模型存在不确定性广义连续-离散系统的故障诊断;然后提出一种结合多模型自适应估计的强跟踪卡尔曼滤波(STUKFMMAE)算法以实现传感器故障的有效隔离。最后,针对基于广义连续-离散系统的惯性传感器故障模型提出仿真算例。仿真数据表明,传统无迹卡尔曼滤波对于传感器故障估计误差为0.002左右,而提出的基于多重渐消因子的强跟踪滤波算法对于传感器故障估计误差最大值为未超过4×10~(-4),且STUKFMMAE相较于UKFMMAE算法具有更好的隔离效果。仿真结果验证了设计方案的有效性。