基于信息熵的软件维护风险度量

 针对软件维护过程中不确定信息难以量化的问题,使用信息熵定量度量软件的维护风险.基于信息熵,引入信息熵定量分析算法,提出了软件维护风险模型,使用信息熵算法定量计算软件维护过程中的不确定程度和损失度.仿真结果表明,基于软件维护风险模型,使用信息熵算法能够定量度量软件的维护风险.     

软件维护性与软件维护质量的度量

分析了软件维护性、软件维护质量度量的若干定性定量方法,对于提高软件维护性、软件维护的质量具有一定的理论和工程意义。     

基于特征类内紧凑性的不平衡医学图像分类方法

近些年,基于深度学习的算法和模型在各种图像分析任务中都取得了显著的成功,与常见的自然图像相比,医学图像数据集依然面临高度不平衡的问题,不平衡数据会导致特征空间里的决策边缘倾向样本多的类别,导致分类效果的下降.为了解决该问题,提出一种基于卷积神经网络考虑特征类内紧凑性的不平衡医学图像分类方法（Z-Score Compactness-based Convolutional Neural Network,ZC3NC）.首先,从一个卷积神经网络的最后一层卷积层提取训练集样本与测试集样本的特征图,随后引入一个新的Z分数来度量测试集数据的特征图相对训练集每个类在特征空间上的偏离度,偏离度的度量基于类内的紧凑度,其主要关注样本的分布特性,对各类样本数量的不平衡性不敏感.最终,根据计算的偏离度,对测试集的数据进行分类.在DermaMNIST数据集上的实验表明,在不对数据和神经网络模型做任何额外增强的情况下,该方法的平衡准确率比原卷积神经网络模型平均提高11.15%,最多提高14.08%,证明提出的分类方法能有效地提高多种卷积神经网络对不平衡医学图像数据的分类性能.此外,和最先进的不平衡分类方法 Und...     

基于FisherVector的图像精细分类方法

该发明提出一种基于Fisher Vector的图像精细分类方法,用以解决现有图像精细分类方法中存在分类准确率低的技术问题,包括如下步骤:读取图像库数据,得到包含各个类别的训练图像集和测试图像集;提取训练图像集和测试图像集中每幅图像的RGB特征;求取关于训练图像集RGB特征描述的混合高斯参数;求取匹配图像块集的Fisher Vector特征矢量;求取训练图像集的最终特征描述和测试图像集的最终特征描述;利用SVM对训练图像集的最终特征描述进行训练,得到分类模型;利用分类模型对测试图像集的最终特征描述进行分类。该发明具有分类准确率较高的优点,可应用于互联网通信、交通和公共安全领域。     

基于深度卷积神经网络与中心损失的人脸识别

传统人脸识别方法手工设计特征过程复杂、识别率较低,对于开集人脸识别通用深度学习分类模型特征判别能力较弱。针对这两方面的不足,提出了一种以分类损失与中心损失相结合作为模型训练监督信号的深度卷积神经网络。首先,利用构建的应用场景数据集优调从公共数据集获得初始化参数的深度人脸识别模型,解决训练数据过小和数据分布差异问题,同时提高模型训练速度;然后,以传统损失函数和新的中心损失作为迁移学习过程中的监督信号,使得类内聚合、类间分散,提高模型输出人脸特征的判别能力;最后,对人脸特征进行主成分分析,进一步去除冗余特征,降低特征复杂度,提高人脸识别准确率。实验结果表明,与传统人脸识别算法相比该算法可以自动进行特征提取,并且相对于通用深度学习分类模型该算法通过度量学习使特征表示更具判别力。在自建测试集和LFW、YouTube Faces标准测试集上都取得了较高的识别率。     

符号序列的预训练HMM分类方法

隐马尔可夫模型(Hidden Markov Model,HMM)是一种双重随机概率模型,已广泛应用于序列数据建模.针对符号序列分类中距离度量定义的困难,提出一种符号序列的预训练HMM分类新方法 .首先,定义一种基于HMM状态转移矩阵的序列距离新度量;其次,为得到不同序列在HMM隐状态共享条件下的状态转移矩阵,提出一种两阶段的预训练方法,先在所有序列上进行HMM预训练以学习所有序列共享的隐状态,再使用共享状态为每条序列进行训练得到各自的状态转移矩阵;最后用近邻分类器对符号序列进行基于距离的分类.在三个应用领域的真实序列上进行了实验,并与基于子序列、HMM变体模型等现有分类方法进行对比,结果表明,所提出的方法能使用较低的特征维度取得较理想的分类精度.     

基于聚类和SVM的数据分类方法与实验研究简

提出一种新的数据分类方法,K-均值聚类分组支持向量机(CG-SVM)方法,在不删减任何样本信息的情况下,该方法使用轮廓系数作为类分离度量参数对数据的正、负类样本进行聚类、分组,对小组进行交叉结合训练,得到不同的SVM分类器,根据测试样本到聚类中心欧氏距离的加权来选择合适的SVM分类器进行类别判断,该方法实现了一个二分类问题转化为多个细分的二分类过程,效果由差变好.对所提方法分别进行了仿真数据和实际数据分类实验.实验结果表明,与传统的SVM方法相比,该算法可以有效地提高分类的准确率.     

基于聚类和SVM的数据分类方法与实验研究

提出一种新的数据分类方法,K-均值聚类分组支持向量机(CG-SVM)方法,在不删减任何样本信息的情况下,该方法使用轮廓系数作为类分离度量参数对数据的正、负类样本进行聚类、分组,对小组进行交叉结合训练,得到不同的SVM分类器,根据测试样本到聚类中心欧氏距离的加权来选择合适的SVM分类器进行类别判断,该方法实现了一个二分类问题转化为多个细分的二分类过程,效果由差变好.对所提方法分别进行了仿真数据和实际数据分类实验.实验结果表明,与传统的SVM方法相比,该算法可以有效地提高分类的准确率.     

基于AlexNet网络的动物图片分类

Caffe是目前广泛应用于计算机视觉处理的深度学习框架之一,支持卷积神经网络的模型训练与预测。本文利用caffe支持的AlexNet卷积神经网络分别基于加载与不加载基础模型两种模式对五类动物图片进行分类学习与训练,发现加载基础模型的网络模型收敛耗时仅2.77 s,测试集准确率接近100%,实用测试准确率达到99%,且训练与测试损失曲线高度重合,但另一模式的网络模型收敛耗时多达68.89 s,测试集准确率仅为95%,实用测试准确率仅94%,且训练与测试损失曲线存在严重分化。图像分类不仅可以对不同物类的图像进行准确分类,同样可以对不同属性、状态或特性的图像进行准确分类。     

基于注意力机制的细粒度图像分类

针对细粒度图像分类中数据分布具有小型、非均匀和不易察觉类间差异的特征,提出一种基于注意力机制的细粒度图像分类模型.首先通过引入双路通道注意力与残差网络融合对图像进行初步特征提取,然后应用多头自注意力机制,达到提取深度特征数据之间细粒度关系的目的,再结合交叉熵损失和中心损失设计损失函数度量模型的训练.实验结果表明,该模型在两个标准数据集102 Category Flower和CUB200-2011上的测试准确率分别达94.42%和89.43%,与其他主流分类模型相比分类效果更好.     

Softw are Maintainability Evaluation Based on Software Configuration Management Data

As one of the most important attributes of software quality, software maintainability has been widely recognized.However,the existing maintainability evaluation methods are mostly based on subjectively judgment. Thus it is inapplicable or unbelievable. To evaluate software maintainability objectively,the software configuration management( SCM) data are collected to establish a maintainability model. Based on the hidden Markov chain( HMC), a three-state maintainability estimation model is constructed. To validate the feasibility of the model,a real software example of software maintenance activity is given and the result from the example shows the effectiveness of the proposed method.     

Software Maintainability Prediction with UML Class Diagram

Software system can be classified into many function modules from the perspective of user. Unified modeling language( UML) class diagram of each function module was extracted,and design characteristic metrics which influenced software maintainability were selected based on UML class diagram.Choosing metrics of UML class diagram as predictors,and mean maintenance time of function module was regarded as software maintainability parameter. Software maintainability models were built by using back propagation( BP) neural network and radial basis function( RBF) neural network, respectively and were simulated by MATLAB. In order to evaluate the performance of models,the training results were analyzed and compared with leaveone-out cross-validation and model performance evaluation criterion. The result indicated that RBF arithmetic was superior to BP arithmetic in predicting software maintainability.     

Software Maintainability Evaluation Based on Quantitive Model

Software maintainability is one of the most important factors of software quality,but it is seriously difficult to evaluate the maintainability. Without evaluation,it is impossible to control. To estimate software maintainability state,parameter system of software was built up and maintainability state was defined into three states.Thought of application on maintainability evaluation based on hidden Markov chain( HMC) and fuzzy inference was presented.Three-state maintainability estimation model was constructed. To testify the feasibility of the model, a real example of software maintenance activity was carried out and the result from the example validated that the results of this study were applicable.     

基于程序信息的软件可维护性度量

本文提出一种以量化程序信息为基础,利用灰色评估法对特性的质量进行模糊处理的软件可维护性度量模型,并通过实例得出软件可维护性的定量评价结果,证实该模型和评估方法科学合理。     

面向对象的高校科研业绩评估系统的分析研究

介绍了高校科研业绩评估系统研究的背景及意义,以科研业绩评估系统的实际需求为主线,详细的分析和介绍了高校科研业绩评估系统的实际业务流程,确定系统应该具有科研材料收集、科研材料评估、科研业绩上报、分类汇总和教师信息管理等基本功能。在此基础上,通过面向对象的系统分析方法和设计方法,逐步建立了评估系统的需求分析模型。合理设计系统存在的类、对象及其之间的关系,并以此为基础建立系统的软件结构,形成具体的类和对象,通过建立顺序图、协作图等动态模型详细解析业务流程及部分模块的具体实现算法。系统能够很好的完成科研业绩材料的评估,系统可靠性高,可维护性强、安全性好。     

Research on software-cell-based software system

The aim of research on software architecture is to improve the quality attributes of software systems, such as security, reliability, maintainability, testability, reassembility, evolvability. However, a single running system is hard to achieve all these goals. In this paper, software-cell is introduced as the basic unit throughout developing process. Then it is further advanced that a robust, safe and high-quality software system is composed of a running system and four supportive systems. This paper especially discusses the structure of software-cell, the construction of the five systems, and the relations between them. Foundation item: Supported by the National Natural Science Foundation of China (No. 70073022) Biography: Tang Sheng-qun(1946-), male, Professor, research interest: software engineering, architecture     

Refinement and Transformation from Z Specification to C + +

As a kind of formal specification language, Z has gained a position in the field of software development, but there is still no standard way of transforming Z specification into executable code that is promising in increasing the quality, reusability and maintainability of software. With the automatic programming model of software engineering, through the analysis for Z specification language, a feasible semi-automatic way of refinement and transformation is proposed, and the correctness of the procedure is also discussed.     

Software defect prediction using fuzzy integral fusion based on GA-FM

The fuzzy measure and fuzzy integral are applied to the classification of software defects in this paper. The fuzzy measure of software attributes and attributes＇ sets are treated by genetic algorithm, and then software attributes are fused by the Choquet fuzzy integral algorithm. Finally, the class labels of soft- ware modules can be output. Experimental results have shown that there are interactions between characteristic attributes of software modules, and also proved that the fuzzy integral fusing method using Fuzzy Measure based on Genetic Algorithm （GA-FM） can significantly improve the accuracy for software defect prediction.     

一种软件过程模型的阶梯式重用机制

针对每一个软件项目都建立软件过程模型是比较繁琐和费时的工作,重用已有的软件过程模型不仅能缩短开发周期,也能提高软件的可维护性和可靠性.基于上述思想,提出了一种阶梯式重用方法,给出了形式化描述,以及重用实现算法,最后用一个实例来说明重用方法的执行过程及结果.     

Predicting the maintainability of open source software using design metrics

This paper empirically investigates the relationships between 15 design metrics and maintainability of 148 Java open source software. The results show that size and complexity metrics are strongly related to the maintainability of open source software. However, cohesion and coupling, as currently captured by existing metrics, do not seem to have a significant impact on maintainability. When used together, these metrics can predict system maintainability fairly accurately （mean MREs below 30%）.