采用主监视窗扩展技术的增强现实三维注册方法

在增强现实三维注册中,采用重新检测策略时顿卡现象严重。提出一种采用主监视窗扩展技术的增强现实三维注册方法。在检测阶段,引入特征点可信度和主监视窗,并设计主监视窗的分割、扩展和调控策略,大幅提升了计算速度;在跟踪阶段,采用基于图像金字塔的光流算法进行特征点跟踪。实验证明,该方法不仅有效缓解了顿卡现象,也提升了系统整体的实时性。     

基于直方图的水墨画艺术风格研究

基于内容的照片或自然图像分析和检索已经得到广泛的研究,但多数都基于颜色特征,不能直接用于水墨画艺术作品的分析。对图像进行预处理,提出基于灰度直方图描述水墨画笔道力度分布信息及构图风格,并利用神经网络学习水墨画整体技法风格特征,依据此特征进行不同艺术家间的分类。实验结果表明,通过对整体风格特征的提取,直方图模型能够有效表征画家不同的艺术风格,实现了水墨画的自动分类。     

Failure load prediction of structural adhesive joints : Part 2: Experimental study

The objective of this study was to determine how the fracture of adhesive joints depends on elastic beam parameters describing the adherends and the applied loads. The basic specimen geometry was the cracked lap shear joint constructed of aluminium alloy with various adherend and bondline thicknesses. Loads were applied in different combinations of bending, tension and shear to generate a failure envelope for each adhesive and specimen geometry. It was found that crack propagation for precracked specimens occured at a critical strain energy release rate but was also a function of the G_I/G_II ratio and the bondline thickness. The experiments also showed that the loads required to propagate a crack in a precracked specimen were always lower than the loads required to break the fillet. Hence, by treating uncracked joints as being cracked, where the fictitious crack tip is assumed to coincide with the location of the fillet, a conservative estimate of the failure load is obtained.     

基于改进卡方统计的微博特征提取方法

通过对微博文本特征信息的分析与研究,提出一种基于改进卡方统计的微博特征提取方法。扩充微博信息分类特征,在传统的卡方统计量的基础上,引入了频度等因素,改进特征选择方法;在传统的特征项权值计算的基础上,提出了新的改进卡方统计量的方法,改进权重计算效果。对上述方法利用经典KNN和SVM算法进行了测试,实验结果表明该方法提高了微博信息分类的准确率。     

多尺度空间特征提取的脊柱图像拼接算法

针对脊柱图像视野有限,提出一种基于优化MSFE的脊柱图像自动拼接算法。设计出一种基于尺度因子变化的高斯卷积模板尺寸自适应调整以及双向配对办法;相似性度量采用城市距离;利用RANSAC算法去除错配,从而确定待拼接图像之间的变换参数,最后利用加权平均对图像融合。对实际取得的多幅脊柱图像拼接结果表明该算法具有很好的实时性和鲁棒性。     

基于退化四元小波变换的纸币识别

针对如何提取纸币图像特征和提高识别率的问题,综合利用退化四元小波变换具有的相位特性,提出一种基于退化四元小波变换的纸币识别方法.该方法首先对采集的纸币图像进行倾斜校正和边缘检测,然后运用退化四元小波对纸币图像进行分解操作,并对分解系数进行统计分析,将每个分解子带系数的能量和标准差作为该纸币图像的特征向量,最后将支持向量机作为分类器对纸币图像进行识别.本文方法在资源约束的嵌入式清分系统上实现,实验结果表明采用本文提出的算法突破了传统纸币识别系统识别率很难再提高的瓶颈,同时能够满足清分系统的实时性要求.     

基于数据分布的聚类联邦学习

联邦学习(federated learning)可以解决分布式机器学习中基于隐私保护的数据碎片化和数据隔离问题。在联邦学习系统中,各参与者节点合作训练模型,利用本地数据训练局部模型,并将训练好的局部模型上传到服务器节点进行聚合。在真实的应用环境中,各节点之间的数据分布往往具有很大差异,导致联邦学习模型精确度较低。为了解决非独立同分布数据对模型精确度的影响,利用不同节点之间数据分布的相似性,提出了一个聚类联邦学习框架。在Synthetic、CIFAR-10和FEMNIST标准数据集上进行了广泛实验。与其他联邦学习方法相比,基于数据分布的聚类联邦学习对模型的准确率有较大提升,且所需的计算量也更少。     

Order-k α-hulls and α-shapes

We introduce order-k α-hulls and α-shapes – generalizations of α-hulls and α-shapes. Being also a generalization of k-hull (known in statistics as “k-depth contour”), order-k α-hull provides a link between shape reconstruction and statistical depth. As a generalization of α-hull, order-k α-hull gives a robust shape estimation by ignoring locally up to k outliers in a point set. Order-k α-shape produces an “inner” shape of the set, with the amount of “digging” into the points controlled by k. As a generalization of k-hull, order-k α-hull is capable of determining “deep” points amidst samples from a multimodal distribution: it correctly identifies points which lie outside clusters of samples.The order-k α-hulls and α-shapes are related to order-k Voronoi diagrams in the same way in which α-hulls and α-shapes are related to Voronoi diagrams. This implies that order-k α-hull and α-shape can be readily built from order-k Voronoi diagram, and that the number of different order-k α-shapes for all possible values of α is proportional to the complexity of order-k Voronoi diagram.     

Discriminative multi-task objects tracking with active feature selection and drift correction

In this paper, we propose a discriminative multi-task objects tracking method with active feature selection and drift correction. The developed method formulates object tracking in a particle filter framework as multi-Task discriminative tracking. As opposed to generative methods that handle particles separately, the proposed method learns the representation of all the particles jointly and the corresponding coefficients are similar. The tracking algorithm starts from the active feature selection scheme, which adaptively chooses suitable number of discriminative features from the tracked target and background in the dynamic environment. Based on the selected feature space, the discriminative dictionary is constructed and updated dynamically. Only a few of them are used to represent all the particles at each frame. In other words, all the particles share the same dictionary templates and their representations are obtained jointly by discriminative multi-task learning. The particle that has the highest similarity with the dictionary templates is selected as the next tracked target state. This jointly sparsity and discriminative learning can exploit the relationship between particles and improve tracking performance. To alleviate the visual drift problem encountered in object tracking, a two-stage particle filtering algorithm is proposed to complete drift correction and exploit both the ground truth information of the first frame and observations obtained online from the current frame. Experimental evaluations on challenging sequences demonstrate the effectiveness, accuracy and robustness of the proposed tracker in comparison with state-of-the-art algorithms.     

Effect of granule size and the placement geometry on the efficiency of urea supergranules for wetland rice grown on a permeable soil

In a laboratory experiment 5 cm depth of water was allowed to percolate daily down through a 15 cm thick soil (Typic Ustipsamment) layer. It was observed that leaching losses of urea supergranules (USG)-N could be decreased by about 20% by the placement of four 0.25 g granules at four points instead of one 1 g granule at one point. In field microplots, the placement of approximately 30 granules of 0.30 g size instead of 9 granules of 1.00 g size resulted in reduced leaching of USG-N and, in turn, increased rice yield. In a follow-up field study, the advantage of more frequently placed USG was confirmed. As compared with 1 g USG placed in the usual manner in the center of four rice hills, increasing the density of placement in soil produced 15% more rice grain. Further increase in rice yield could be obtained by increasing the number of USG placed in the soil and decreasing the size of the granule from 1.00 g to 0.70 or 0.35 g. With USG of 0.35 and 0.70 g yields were equal or sometimes even slightly higher than with split application of prilled urea on a heavily percolating, low-CEC, light-textured soil.