一种改进的最小二乘回归子空间分割方法

最小二乘回归(LSR)算法是一种常见的子空间分割方法，由于LSR具有解析解，因此它的聚类性能较高。然而LSR算法是应用谱聚类方法聚类数据，谱聚类方法初始化聚类中心是随机的，会影响后面的聚类效果。针对这一问题，提出一种基于聚类中心局部密度和距离这2个特点的改进的LSR算法（LSR-DC）。在Extended Yale B数据集上进行实验，结果表明，该算法有较高的聚类精度，具有一定的鲁棒性，优于现有LSR等子空间分割方法。     

自然光照条件下植被几何光学四分量的提取算法

目的 几何光学四分量是指在太阳光照条件下传感器所能观测的4个光学分量，即光照植被、光照土壤、阴影植被和阴影土壤。四分量是构成遥感几何光学模型的重要内容。在近地表遥感应用中，相机俯视拍照是提取四分量的一个途径。准确快速地从图像数据中提取四分量对植被冠层结构参数反演和植被长势监测具有重要意义。方法 植被与土壤二分量的识别是四分量提取的基础。目前大多数二分类算法在自然光照条件复杂时分类误差较大。本文基于卷积神经网络（CNN）和阈值法实现了多种二分类和四分量提取算法。阈值法中，使用SHAR-LABFVC （shadow-resistant algorithm：LABFVC）实现植被与土壤的二分类，并在此基础上应用二次阈值分割获取四分量，称为二次阈值法；基于CNN的方法中，采用U-Net架构，并使用RGB和RGBV数据进行训练得到U-Net和U-Net-V模型，前者完成二分类和四分量任务，后者只完成四分量提取实验。最后，对一种结合U-Net与阈值法的混合算法进行四分量提取实验。结果 本文在18幅图像（1 800个子图）数据上进行了实验，结果表明，与目视解译得到的四分量真值相比较，U-Net-V和混合法精度最高，具有相近的均方根误差（RMSE）（0.06和0.07）和相关系数（0.95和0.94）；二次阈值法与U-Net模型精度略低于上述两种算法，RMSE分别是0.08和0.09，相关系数均为0.88。在二分类实验中，U-Net的分类正确率是91%，SHAR-LABFVC为85%。结论 通过对比实验表明，在二分类问题中，U-Net可以更好地应对复杂自然光照条件下的数字图像。在四分量提取实验中，混合法和U-Net-V的结果优于U-Net与二次阈值法，可以用于提取四分量。     

Estimation of the degree of hydration of concrete through automated machine learning based microstructure analysis – A study on effect of image magnification

The scanning electron microscopy (SEM) images are commonly used to understand the microstructure of the concrete. With the advancements in the field of computer vision, many researchers have adopted the image processing technique for the microstructure analysis. Most of the previous methods are not adaptable, non-reproducible, semi-automated, and most importantly all these methods are highly influenced by image magnification. Therefore, to overcome these challenges, this paper presents a machine learning based image segmentation method for microstructure analysis and degree of hydration measurement using SEM images. In addition, the authors looked into the impact of magnification of SEM images on the model accuracy and classifier training for the degree of hydration measurement considering two scenarios. First, the image segmentation was performed using a classifier of specific magnification, and then a common classifier is trained using the image of different magnification. The results show that the Random Forest classifier algorithm is suitable for microstructure analysis using SEM images. Through the statistical analysis, it has been proved that there is no significant effect of magnification on model training and accuracy for the degree of hydration measurement. So, a single classifier can be used to process the images of different magnification of a specimen which reduces the effort of training and computational time. The proposed method can generate highly accurate and reliable results in a shorter time and lower cost. Moreover, the findings in this research can be useful for researchers to determine the optimum magnification required for the microstructure analysis.     

CO_2跨临界循环中绝热毛细管数值模拟计算与分析

前期通过搭建试验台,分析了二氧化碳跨临界循环中毛细管的节流特性,并运用无量纲分析法计算出了毛细管质量流量无量纲试验关联式,经可靠性分析发现该关联式优于diogo的关联式。若超出考虑的范围时,该关联式模型可能产生较大的误差。由此,利用微元分割法,设计VB程序对毛细管长度和流量进行数值计算,将计算结果与前期试验测试结果和文献值做比较。在流量计算方面:95%的计算值与前期试验值的偏差在10%以内,与文献试验值的偏差全部在10%以内。在长度计算方面,90%的计算值与前期试验值的偏差在15%以内,75%的长度计算值与文献试验值的偏差在15%以内;在偏差分布方面:流量计算值与前期试验值和文献试验值,均有95%以上的流量计算值与试验值的偏差在10%以内。最后利用该计算程序,在标准工况下对毛细管尺寸进行数值计算,绘制成毛细管选型图。     

Segment dependent dynamic multi-histogram equalization for image contrast enhancement

Histogram equalization (HE) method proved to be a simple and most effective technique for contrast enhancement of digital images. However it does not preserve the brightness and natural appearance of the images, which is a major drawback. To overcome this limitation, several Bi- and Multi-HE methods have been proposed. Although the Bi-HE methods significantly enhance the contrast and may preserve the brightness, the natural appearance of the images is not preserved as these methods suffer with the problem of intensity saturation. While Multi-HE methods are proposed to further maintain the brightness and natural appearance of images, but at the cost of contrast enhancement. In this paper, two novel Multi-HE methods for contrast enhancement of natural images, while preserving the brightness and natural appearance of the images, have been proposed. The technique involves decomposing the histogram of an input image into multiple segments based on mean or median values as thresholds. The narrow range segments are identified and are allocated full dynamic range before applying HE to each segment independently. Finally the combined equalized histogram is normalized to avoid the saturation of intensities and un-even distribution of bins. Simulation results show that, for the variety of test images (120 images) the proposed method enhances contrast while preserving brightness and natural appearance and outperforms contemporary methods both qualitatively and quantitatively. The statistical consistency of results has also been verified through ANOVA statistical tool.     

Learning hierarchical spatio-temporal pattern for human activity prediction

Human activity prediction has become increasingly valuable in many applications. This paper, initially from the perspective of cognition science, presents a novel approach to learning a hierarchical spatio-temporal pattern of human activities to predict ongoing activities from videos that contain only the onsets of the activities. Spatio-temporal pattern can be learned by a Hierarchical Self-Organizing Map (HSOM), which consists of two self-organizing maps (i.e., action map and actionlet map) connected via associative links trained by Hebbian learning. Ongoing activities can be predicted by Variable order Markov Model (VMM), which provides the means for capturing both large and small order Markov dependencies based on the training actionlet sequences. Experiments of the proposed method on four challenging 3D action datasets captured by commodity depth cameras show promising results.     

Nucleus and cytoplast contour detector from a cervical smear image

In this paper, a nucleus and cytoplast contour detector (NCC detector) is presented to automatically detect the cytoplast and nucleus contours of a cell in a cervical smear image. The NCC detector uses the adaptable threshold decision (ATD) method to separate the cell from the cervical smear image, and then uses the maximal gray-level-gradient-difference (MGLGD) method, proposed in this paper, to extract the nucleus from the cell. The experimental results show that the NCC detector is superior to two existing methods, the gradient vector flow-active contour model (GVF-ACM) and the edge enhancement nucleus and cytoplast contour (ENNCC) detector, in segmenting the cytoplast and nucleus of a cell.     

DiSeg 1.0: The first system for Spanish discourse segmentation

Nowadays discourse parsing is a very prominent research topic. However, there is not a discourse parser for Spanish texts. The first stage in order to develop this tool is discourse segmentation. In this work, we present DiSeg, the first discourse segmenter for Spanish, which uses the framework of Rhetorical Structure Theory and is based on lexical and syntactic rules. We describe the system and we evaluate its performance against a gold standard corpus, divided in a medical and a terminological subcorpus. We obtain promising results, which means that discourse segmentation is possible using shallow parsing.     

Unsupervised segmentation of hidden semi-Markov non-stationary chains

The Bayesian segmentation using Hidden Markov Chains (HMC) is widely used in various domains such as speech recognition, acoustics, biosciences, climatology, text recognition, automatic translation and image processing. On the one hand, hidden semi-Markov chains (HSMC), which extend HMC, have turned out to be of interest in many situations and have improved HMC-based results. On the other hand, the case of non-stationary data can pose an important problem in real-life situations, especially when the model parameters have to be estimated. The aim of this paper is to consider these two extensions simultaneously: we propose using a particular triplet Markov chain (TMC) to deal with non-stationary hidden semi-Markov chains. In addition, we consider a recent particular HSMC having the same computation complexity as the classical HMC. We propose a related parameter estimation method and the resulting unsupervised Bayesian segmentation is validated through experiments; in particular, a real radar image segmentations are provided.     

A Character Flow Framework for Multi-Oriented Scene Text Detection

Scene text detection plays a significant role in various applications,such as object recognition,document management,and visual navigation.The instance segmentation based method has been mostly used in existing research due to its advantages in dealing with multi-oriented texts.However,a large number of non-text pixels exist in the labels during the model training,leading to text mis-segmentation.In this paper,we propose a novel multi-oriented scene text detection framework,which includes two main modules:character instance segmentation (one instance corresponds to one character),and character flow construction (one character flow corresponds to one word).We use feature pyramid network(FPN) to predict character and non-character instances with arbitrary directions.A joint network of FPN and bidirectional long short-term memory (BLSTM) is developed to explore the context information among isolated characters,which are finally grouped into character flows.Extensive experiments are conducted on ICDAR2013,ICDAR2015,MSRA-TD500 and MLT datasets to demonstrate the effectiveness of our approach.The F-measures are 92.62％,88.02％,83.69％ and 77.81％,respectively.