基于深度神经网络的高位滑坡范围预测

以贵州尖山营滑坡为工程背景,通过对深度学习的总结与分析,建立多层感知器模型以对该滑坡危险区范围进行非线性预测研究。通过对深度神经网络算法的优化,构建64-128-32-1四层多层感知器模型,并以滑坡最大高差、滑坡体积、滑源区坡度、坡脚坡度、地层倾角作为输入量,以滑坡最大水平运动距离作为输出量对该模型进行训练,实现影响因素与运动距离的非线性映射。根据对贵州省尖山营滑坡调查和研究,尖山营滑坡区域面积约648 700 m2,体积约1 200万 m3,属于特大型滑坡。依据最优模型对该滑坡进行滑距预测,滑坡平面直线距离1 769 m区域内为危险区域。     

视觉测量中多光学靶点亮度自适应控制

提出了一种面向大视场高精度视觉测量的多红外目标靶点（红外LED）亮度自适应控制方法。以微控制器PIC16F873A和LED驱动芯片MAX6964作为核心器件,设计了可满足大空间范围内靶点亮度控制需求的多目标靶点控制驱动器。为了使LED在近距测量时的调节精度和远近距离测量时的发光亮度都能满足要求,根据MAX6964输出特点设计了PWM信号输出电路,并采用推挽电路驱动LED。实验证明,目标靶点在CCD相机上成像光斑稳定一致。     

基于像素差异的金属腐蚀区域分割仿真研究

研究基于图像的金属腐蚀区域分割问题,提高分割区域的准确性。针对传统基于金属本身与腐蚀颜色的区别,利用像素差异的变化进行腐蚀区域的分割,因当腐蚀颜色与金属的本体颜色相差不是很大,或者在边沿发生像素的混淆时,利用差异边缘像素相减的方法很难完整分割腐蚀区域的问题。为了解决上述问题,提出利用空间像素差异距离聚类的金属腐蚀区域图像分割算法。通过计算差异像素之间的差异距离概率,将差异像素特征进行严格划分,对划分后的结果进行排序计算,详细描述混淆边沿信息,对腐蚀区域进行完整分割。实验证明,改进方法能够对金属腐蚀造成的细微区域完整分割,取得了满意的效果,为金属腐蚀区检测提供了参考。     

基于叶片形态特征的葡萄品种自动识别

葡萄在全世界种植范围较广,葡萄品种识别对葡萄资源统计、新品种鉴定及遗传资源保护都有重要意义。以成熟叶片为基本材料,不变矩、面积、周长、主叶脉长度和叶柄长度等11维特征向量为基础数据,采用叶片特征选择、图像处理、特征值提取、建立分类模型等方法,运用国际葡萄与葡萄酒组织形态分类方法和改进的欧式距离等技术开发一个基于葡萄叶片数字图像的葡萄品种自动识别软件。共测试17种酿酒和3种野生葡萄,识别率达87%。该方法具有自动化、识别速度快、花费低、省时省力等特点,可应用于葡萄病虫害的识别、杂交新品种双亲的判定等。     

基于模糊数学的高维稀疏数据聚类统计方法设计

传统的数据聚类统计方法仅适用于低维数据聚类问题,为此,本研究设计了基于模糊数据的高维稀疏数据聚类统计方法,以期提升高维稀疏数据的聚类统计效果。以模糊C均值聚类算法为基础,通过优化初始聚类中心解决局部最优问题,缩短聚类统计时间;然后引入权重机制,令该方法适用于高维稀疏数据聚类统计。基于此,以余弦距离替换原有的欧几里德距离,提高高维稀疏数据聚类统计效果。实验证明：在数据维度不同时,该方法均有较优的聚类统计效果。当数据维度较低时,分块比例为10%时聚类统计效果最优;当数据维度较高时,分块比例为40%时聚类统计效果最优。在不同稀疏度等级时,该方法的命中率和聚类统计效率均较高。     

基于BC聚类的差分隐私保护推荐算法

为提高差分隐私保护下推荐算法的准确性,提出了一种考虑差分隐私保护的基于Bhattacharyya系数（BC）的聚类推荐算法.以BC作为项目相似性度量的标准,根据BC相似性对项目进行K-medoids聚类,并在聚类簇中进行私有项目邻居选择.最后,根据最近邻居集信息,对用户的评分进行预测和Top-n推荐.提出的方案有效地克服了已有方法中存在的相似性度量依赖于共同评分的问题,提高了相似性度量的准确性,有效避免了因隐私保护而造成的最近邻居集质量下降的问题.理论分析和实验测试的结果表明,该方法在实现隐私保护的同时还能有效保证推荐的高质量,较好地实现了隐私保护和数据效用之间的平衡,具有良好的应用潜力.     

Stiffer and super-tough poly(butylene terephthalate) based blends by modification with phenoxy and maleated poly(ethylene-octene) copolymers

New super-tough poly(butylene terephthalate) (PBT) materials were obtained by melt blending PBT with both 20 wt% phenoxy (Ph) and 0-30 wt% maleic anhydride grafted poly(ethylene-octene) (mPEO) copolymers with different grafting levels. Ph was completely miscible in the PBT matrix. The presence of mPEO did not influence either the nature of the PBT-Ph matrix or the crystallization of PBT. The overall decrease in particle size and in interfacial tension upon grafting indicated that compatibilization had taken place. Super-tough (impact strength 23-fold that of the PBT) and stiffer PBT based blends were obtained at mPEO contents equal to or higher than 15%. The dependence of the critical inter-particle distance (τ_c), on both adhesion measured by means of the interfacial tension, and on the relation between the modulus of the matrix and that of the rubbery dispersed phase (E_m/E_d), is proposed.     

基于扩散距离和MDS的非刚性模型相似性分析

在非刚性模型相似性分析方法中, 基于测地距离的等距嵌入方法对模型拓扑变化非常敏感, 对于有局部拓扑变化的完全相似的模型也无法得到正确结果。为了弥补这一不足, 提高非刚性模型相似性分析的准确性, 将扩散距离与多维尺度分析(MDS)相结合, 计算非刚性模型的等距嵌入模型, 再利用迭代最近点(ICP)算法通过计算嵌入模型的相似性来实现原始非刚性模型的相似性分析。实例证明该方法对含有拓扑变化的非刚性模型可以得到理想的相似性分析结果。     

Statistical properties of interparticle/void distance for amorphous plastics toughening

Statistical properties of interparticle/void distance (ID) for various particle/void and dispersion types are studied in relation with toughening of plastics using computer‐generated three‐dimensional models. Particle/void size groups adopted were either of constant diameter or of log‐normal distribution. Particles/voids were dispersed at uniform‐random or flocculated with multiple clusters. It was found that IDs are (a) of approximately Gaussian distribution for particles/voids of either a constant diameter or a log‐normal distribution, when they are dispersed at uniform‐random, but (b) not of Gaussian distribution for particle/void sizes of bimodal log‐normal distribution, nor for flocculated log‐normal distribution of particles/voids dispersed with multiple clusters. It was also found that the degree of ID uniformity for a single group of log‐normally sized particles/voids is not sensitive to standard deviation of particle/void size. Mixing effect on ID properties using two groups of log‐normally distributed particles/voids with similar mean particle/void diameters was simulated. It was found that, when a significant amount (36 vol %) of particles/voids of a small mean and standard deviation of ID was mixed with a group of particles/voids of a large mean and standard deviation of ID, mean and standard deviation of ID for the mixture were not substantially lower than those of the group of particles/voids of the large mean and standard deviation of ID. It was also found that the degree of ID uniformity for the mixture of the two groups was lower than those of individual groups, indicating that the mixing has deleterious effect on toughening. © 2006 Wiley Periodicals, Inc. J Appl Polym Sci 101: 4256–4262, 2006     

纹理豪斯多夫距离在人脸识别算法中的应用

很多经典的人脸识别方法难以适应姿势变化及人脸错位的情形。为了解决这一问题,提出一种基于纹理豪斯多夫距离(THD)的人脸识别算法。将人脸图像的空间量及纹理特征相结合,使其在深入的头部转动和人脸错位中都有很高的容错度。在FERET及Yale两大人脸数据库的实验表明,与其他经典的方法相比较,所提出的方法取得了更好的识别效果。