基于细胞神经网络鲁棒性的彩色图像边缘检测

根据细胞神经网络(CNN)数学模型,提出一种新的彩色图像边缘检测方法。新方法继承了CNN的优点,解决了CNN现有算法不能直接检测彩色图像边缘的问题。该方法充分利用图像中的颜色信息,通过欧几里得距离度量像素之间的差异,使CNN方程可以在RGB彩色空间中进行运算。对CNN模板进行理论分析和鲁棒性研究,提出一个实现彩色图像边缘检测功能要求的CNN鲁棒性定理,为设计相应的CNN模板参数提供了解析判据。实验结果表明,该方法可以对彩色图像进行有效的边缘提取,定量评价验证了CNN边缘检测定位准确的优点。     

基于神经网络的图像边缘检测方法

提出了一种基于神经网络的图像边缘检测新方法.该方法首先基于邻域灰度极值提取边界候选图像,然后以边界候选象素及其邻域象素的二值模式作为样本集,输入边缘检测神经网络进行训练.边缘检测神经网络采用BP网络,为加快网络的训练速度,采用了滚动训练和权值随机扰动的方法.实验表明,该方法提高了神经网络的学习效率,获得的边缘图像封闭性好,边缘描述真实.     

PSO-Gabor-CNN算法在印刷品套印缺陷的检测

目的二维Gabor滤波器含有多个参数,在印刷品套印缺陷检测中,二维Gabor滤波器使用不同参数增强图像特征的效果差别较大,为了获得二维Gabor在某印刷品套印缺陷检测下的优化参数。方法在印刷品套印缺陷检测中,提出一种PSO-Gabor-CNN算法,采用Sobel算子对印刷品图像进行边缘检测,以粒子群算法(PSO)对二维Gabor滤波器的中心最大频率kmax、带宽σ、模板窗口window进行参数寻优,处理后的图像与模板图像采用加权欧式距离进行评价。然后用优化后的Gabor滤波器对图像进行滤波,最后采用卷积神经网络(CNN)对印刷品套印缺陷进行检测和分类。结果通过粒子群算法,确定了二维Gabor中心最大频率kmax为6.0476、带宽σ为0.1444、模板窗口window为27×27取得最佳效果,此时加权欧式距离为1.1927×10-33。卷积神经网络经过70次训练的均方误差为0.0035,测试样本正确率为96.93%。该方法与无数据预处理的BP神经网络(BPNN)、Sobel预处理的BP神经网络(Sobel-BPNN)、无数据预处理的卷积神经网络(CNN)、Sobel预处理的卷积神经网络(Sobel-CNN)对比,表现出了较好的识别效果。结论该方法可以获取二维Gabor滤波器的较优参数,从而获得较好的滤波效果,将其应用于套印缺陷检测,具有一定的应用价值。     

基于参数优化SDP分析的转子故障诊断方法

针对存在多种故障类型不同故障严重程度的转子故障诊断问题，提出了一种基于参数优化对称点模式(symmetrized dot pattern, SDP)分析的智能诊断方法。首先，利用SDP分析提取多个传感器信号的故障特征并将其融合为SDP图像；然后，以基于欧氏距离定义的图像区分度函数为适应度函数，基于天牛须搜索(beetle antennae search, BAS)算法获得SDP分析中角域增益因子与时间延滞系数的最佳取值；最后，利用SDP图像训练卷积神经网络(convolutional neural network, CNN)获得转子故障诊断模型。试验研究表明，该方法相较于其他故障诊断方法具有更高的诊断精度，且在强噪声环境下的诊断表现良好。基于BAS算法优化后的SDP分析放大了不同类型不同严重程度转子故障的表征差异，提高了故障诊断精度。     

一种基于蚁群聚类的多模板图像边缘融合方法

根据所采集图像的离散性特点,提出了一种基于蚁群聚类的新型的多模板图像边缘融合方法.该算法综合了图像聚类的方法和模板边缘提取的方法,运用蚁群聚类的思想来提取图像的边缘特征,并进一步融合多种边缘提取模板来设置聚类的启发式引导函数和初始聚类中心,从而避免了传统蚁群算法聚类搜索时的盲目性.系列仿真实验验证了这一方法的可行性和有效性.     

基于卷积神经网络图像分类的轴承故障模式识别

针对传统的基于数据驱动的机械故障模式识别方法中需要人工构造算法提取特征以及人工构造特征提取算法繁琐的问题,结合卷积神经网络(CNN)在图像特征自动提取与图像分类识别中的广泛应用,提出了一种基于CNN图像分类的轴承故障模式识别方法。首先,利用集合经验模态分解(EEMD)方法对轴承振动信号进行自适应分解并用相关系数对得到的本征模函数分量进行筛选。其次,对筛选得到的本征模函数分量进行伪魏格纳-威利时频分析(PWVD)计算得到信号的时频分布图,并对时频图进行预处理。最后,将轴承15种不同工况预处理后的时频图利用CNN进行特征提取与分类识别。将该方法与同类方法进行了对比,分类正确率提高了4.26%。     

基于边缘检测的邻域加窗图像去噪算法

针对目前图像去噪算法中,消除噪声的同时又破坏边缘细节信息的问题,本文提出了结合边缘检测及邻域加窗的新算法.该算法采取平稳小波基以保持相位不变性,对低频和高频子带进行边缘检测,并将检测后的边缘信息选择后融合,即可得到原图像近似的边缘信息.依据小波方向性特点和层内相关性原理,对不同的子带在非边缘信息处采用不同的模板进行加窗处理.实验结果表明,该方法在降低了图像噪声的同时又尽可能地保留了图像的细节,较好地复原了图像.     

基于数学形态学和颜色特征的车牌定位方法

为了解决单一车牌定位算法在复杂背景中定位效果不理想的现状,提出一种数学形态学和颜色特征相结合的算法对车牌进行定位.首先利用最大类间方差法（Ostu算法）找到一个最佳的阈值,根据所得阈值把得到的灰度图像二值化,然后采用一种改进的数学形态学算法对图像进行边缘检测,最终结合数学形态学和车牌颜色特征进行准确定位.实验表明该算法明显优于传统或单一的车牌定位方法,定位准确率高,对背景限制少,应用范围广.     

多区域参数估计实时图像盲复原

APEX图像盲复原技术(BIR)往往因算法速度过慢,不稳定或者输入参数过多而难以实时应用。针对这些限制,基于一般BIR过程中有一个非常重要的属性:降质图像与降质点扩散函数PSF尺度大小存在强不对称性,提出了多区域参数估计的方法,通过MAP理论证明了多次参数估计的合理性,并且通过实验说明了多次参数估计方法的可行性。通过多区域参数估计方法,提高了APEX算法估计参数的精度,进而提高其可用性。通过对仿真图像和实际的图像进行复原处理验证了算法的有效性。该算法速度快、稳定,可用于低噪声高斯类模糊图像实时图像盲复原。     

Self-adaptive image semantic classification based on tolerance granular space model

针对图像底层特征和高层语义之间存在的语义鸿沟问题,运用相容粒度空间模型对图像语义分类进行了研究,提出一种自适应的图像语义分类方法,为解决此问题探索出了一种有效途径.该方法将图像集建模为基于原始特征的相容粒度空间；在此空间中,通过引入相容参数和构造距离函数来定义相容关系,从而通过调整相容参数可有效控制对象邻域粒的大小,最终可直接处理图像的实数型特征而无需进行离散化等预处理；此外,通过引入相容度的方法实现对相容参数的自适应优化,从而自动调整邻域粒的大小,使得构造的分类器几乎不需要手工设置参数即可自动适应于各种不同类型的图像集,并获得比同类算法更好的分类准确率.实验结果验证了这种方法的有效性和可行性.     

Influence of mould design on the solidification of heavy forging ingots of low alloy steels by numerical simulation

Ingot casting of a 6-ton, heat-treatable Cr–Mo low alloy steel was simulated using finite element method in three dimensions. Effects of casting parameters including bottom pouring rate, mould slenderness ratio, mould slope, and height and shape of the hot top isolate on solidification behaviour and crack susceptibility during subsequent hot forging of the ingot were investigated. The simulation model was validated against experimental data of two different ingot mould designs. Influences of the casting parameters on the riser efficiency and possible crack formation in the intersection of hot top and ingot body during subsequent open-die forging of the cast steel ingots were discussed. Results showed that pouring the melt under a constant rate, reducing the mould slenderness ratio, and using a proper design for the hot top isolate would all improve the riser efficiency and thereby possibly reduce crack susceptibility during subsequent hot forging.     

共沉淀法合成La_(0.9)Sr_(0.1)Ga_(0.8)Mg_(0.2)O_(3-δ)及其性质的研究

采用共沉淀法合成了具有钙钛矿结构的中温固体电解质La_(0.9)Sr_(0.1)Ga_(0.8)Mg_(0.2)O_(3-δ)(LSGM),并用DTA-TGA和X射线衍射仪分析了LSGM材料中钙钛矿相的形成过程,采用SEM、交流阻抗谱等检测技术对LSGM电解质的结构及性能进行了表征.XRD分析结果表明,1200℃烧结后,粉体开始形成钙钛矿结构,随温度的升高粉体中杂相含量越来越少,于1450℃时形成了单一的钙钛矿相结构.     

Thermal design and performance of the SCUBA-2 instrument 1-K and mK systems

Various research fields require large and complex instruments containing detectors operating at millikelvin temperatures. The materials and techniques traditionally used in cryogenics are often unsuitable for the demanding requirements of such instruments. We describe the thermal design and performance of the 1-K and millikelvin systems of the SCUBA-2 instrument. This is an astronomical “camera” operating at wavelengths of 450 and 850 μm. It is the largest and most complex instrument ever built for sub-mm astronomy, and the first to use a cryogen-free dilution refrigerator. The design consists of a mix of traditional techniques (but used in demanding situations) as well as novel elements. The thermal performance has been stable and very successful, and we hope that the details described here will be useful to the designers of future large instruments.     

Study of the interface behaviour between MABS/TPU bi-layer structures obtained through over moulding

The strength of adhesion between two over-moulded polymers, methylmethacrylate–butadiene–styrene copolymer (MABS) and thermoplastic polyurethane (TPU) that constitute a bi-component laminar system has been the subject of study. Results showed that at the bi-layer interface the adhesive fracture toughness increases as temperature and surface roughness are increased. Roughness has been demonstrated to be the most important parameter defining adhesion strength and failure. With the increase of roughness the failure was observed to change from adhesive towards cohesive type. An increase in either the temperature or the pressure applied to the samples caused a rise in adhesion energy. However, pressure seemed to have a minor effect in comparison to temperature. Significant increments in adhesion were obtained after applying the corona discharge treatment (CDT) to the attaching surfaces resulting in adhesion strengths almost double those of non-treated systems.     

非晶Sm5Fe74.3Nb1.5Si11.7B4.5C2.5Cu0.5合金经预退火后的晶化动力学的DTA研究

研究了非晶Sm5Fe74.3Nb1.5Si11.7B4.5C2.5Cu0.5合金经400℃，保温10min预退火后的晶化动力学。结果表明；该合金的晶化相为α-Fe固溶体和Sm2Fe17Cx金属间化合物，两相的晶化表观激活能分别为557KJ/mol和514KJ/mol，当晶化体积分数为60%时，α-Fe相的晶化激活能达极大值；Sm2Fe17Cx相晶化激活能则随其晶化体积分数的增加而逐渐减小。     

PMMA单分子膜成膜特性研究

研究了ＰＭＭＡ单分子膜的成膜特性及其结构．结果表明，ＰＭＭＡ能够在较大的表面压范围内形成稳定的单分子膜，并且具有不可重复压缩性、表面压力的各向异性和松驰特性．ＴＥＭ照片显示，ＰＭＭＡ分子链在单分子膜中是有序平行排列的     

HDDR工艺条件及均匀化处理对Nd12.6Fe69.3Co11.6 B6.0 Al0.5合金磁性能的影响

研究了HDDR工艺条件(包括HD温度、HD时间、DR温度、DR时间)对Nd12.6Fe69．3Co11.6B6．0Al0．5合金磁性能的影响；比较了均匀化处理和未均匀化处理的合金经HDDR工艺后磁性能的差异；通过对两种舍金原始态的XRD谱线分析，找出存在差异的原因。     

Fabrication of NiCr alloy parts by selective laser melting: Columnar microstructure and anisotropic mechanical behavior

NiCr alloy, because of its wide applications in electrical elements and dental field was widely studied in the past. In this work, NiCr cubes and tensile specimens were fabricated by using a new processing technique-selective laser melting (SLM). Microstructural and mechanical behavior characterization of SLM-processed NiCr components was performed. An unusual columnar microstructural architecture composed of 〈1 0 0〉 texture (corresponding to (2 0 0) plane) oriented the building direction was observed. Moreover, it was found that the columnar grain growth across the melt pools occurred during the SLM process and the growth trend became stronger with the decrease of the laser scanning speed. Associated with the microstructural characteristic, an anisotropic mechanical behavior at different reference planes (i.e., at the horizontal and vertical surfaces) was demonstrated for the samples fabricated using different processing parameters. The results showed that with increasing the laser scanning speed, the microhardness at the horizontal surface decreased, while at the vertical surface it increased; an increase of the yield strength (YS) and the ultimate tensile strength (UTS) was observed.     

Surface Layer of Ni_(68)P_(18)Cr_(14) Amorphous Alloy

In Ni_(68)P_(18)Cr_(14) amorphous alloy, the dominant component elements Ni and P were in lower contentin the surface layer, while Cr was enriched. There was a P enrichment region just below the surfaceoxide layer, which was supposed to enhance the surface segregation of Cr. Nickel was less oxidizedthan Cr.     

Thermal Stability of Amorphous Hydride Mg_(50)Ni_(50)H_(54) and Mg_(30)Ni_(70)H_(45)

The thermal stability of amorphous ternary hydrides Mg_(50)Ni_(50)H_(54) and Mg_(30)Ni_(70)H_(45) and their corre-sponding amorphous binary alloys Mg_(50)Ni_(50) and Mg_(30)Ni_(70) were studied with X-ray diffraction(XRD) and differential scanning calorimetry(DSC). Samples of the amorphous alloys were preparedby mechanical alloying and the amorphous hydrides were obtained by charging the alloys with gas-eous hydrogen at 3.0 MPa and 423 K. It was found that the amorphous hydrides released most oftheir hydrogen before the crystallization of the essentially hydrogen depleted amorphous alloy. Thecrystallization temperature of amorphous Mg_(50)Ni_(50)H_(54) elevated and that of amorphousMg_(30)Ni_(70)H_(45) did not change in relation to the original binary amorphous alloy. This is very excep-tional for amorphous hydrides. The reason for the effects of hydrogen absorption/desorption on thecrystallization of amorphous alloys was discussed.