基于张量分解融合RGB-D图像的物体识别

为了充分利用RGB-D图像的深度图像信息，提出了基于张量分解的物体识别方法。首先将RGB-D图像构造成一个四阶张量，然后将该四阶张量分解为一个核心张量和四个因子矩阵，再利用相应的因子矩阵将原张量进行投影，获得融合后的RGB-D数据，最后输入到卷积神经网络中进行识别。RGB-D数据集中三组相似物体的识别结果表明，利用张量分解融合RGB-D图像的物体识别准确率高于未采用张量分解的物体识别准确率，并且单一错分实例的准确率最高可提升99%。     

OWL‐based acquisition and editing of computer‐interpretable guidelines with the CompGuide editor

Computer‐Interpretable Guidelines (CIGs) are the dominant medium for the delivery of clinical decision support, given the evidence‐based nature of their source material. Therefore, these machine‐readable versions have the ability to improve practitioner performance and conformance to standards, with availability at the point and time of care. The formalisation of Clinical Practice Guideline knowledge in a machine‐readable format is a crucial task to make it suitable for the integration in Clinical Decision Support Systems. However, the current tools for this purpose reveal shortcomings with respect to their ease of use and the support offered during CIG acquisition and editing. In this work, we characterise the current landscape of CIG acquisition tools based on the properties of guideline visualisation, organisation, simplicity, automation, manipulation of knowledge elements, and guideline storage and dissemination. Additionally, we describe the CompGuide Editor, a tool for the acquisition of CIGs in the CompGuide model for Clinical Practice Guidelines that also allows the editing of previously encoded guidelines. The Editor guides the users throughout the process of guideline encoding and does not require proficiency in any programming language. The features of the CIG encoding process are revealed through a comparison with already established tools for CIG acquisition.     

Computer aided diagnosis of brain abnormalities using texture analysis of MRI images

The drive of this study is to develop a robust system. A method to classify brain magnetic resonance imaging (MRI) image into brain-related disease groups and tumor types has been proposed. The proposed method employed Gabor texture, statistical features, and support vector machine. Brain MRI images have been classified into normal, cerebrovascular, degenerative, inflammatory, and neoplastic. The proposed system has been trained on a complete dataset of Brain Atlas-Harvard Medical School. Further, to achieve robustness, a dataset developed locally has been used. Extraordinary results on different orientations, sequences of both of these datasets as per accuracy (up to 99.6%), sensitivity (up to 100%), specificity (up to 100%), precision (up to 100%), and AUC value (up to 1.0) have been achieved. The tumorous slices are further classified into primary or secondary tumor as well as their further types as glioma, sarcoma, meningioma, bronchogenic carcinoma, and adenocarcinoma, which could not be possible to determine without biopsy, otherwise.     

Refurbishment or quality recovery: joint quality and pricing decisions for new product development

Refurbishment and quality recovery, as two important strategies of reverse logistics, have been widely used for consumer durables, especially consumer electronics. This trend is driven by manufacturers’ profit-seeking behaviours and the increasing environmental concerns of society. This study investigates and compares the optimal emergence of these two strategies when product quality is endogenous. Model analyses provide several notable insights. First, we find that the refurbishment strategy dominates the quality recovery strategy when the fraction of the recoverable quality-inducing components in the defective product is low enough. Moreover, when the refurbishment strategy emerges as the optimum, the quality and price of the new product as well as the total realised sales of both new and refurbished products are consistently higher than those with the quality recovery strategy. In addition, when either the refurbishment or quality recovery strategy emerges as the optimum, the overall consumer surplus is also enhanced. Finally, in the model generalisation, we show that a synthetic strategy with which some of the returned defective products are refurbished and others are remanufactured through quality recovery can certainly increase the manufacturers’ profit compared to either one of them when the fraction of the recoverable quality-inducing components in the defective product is moderate.     

基于DAE-BP神经网络的股票预测研究

股票指标数据种类多、维度高，且指标之间存在多重共线性。为了降低数据的维度、消除指标间的多重共线性和预测股票价格，首先构建了基于受限布尔兹曼机的深度自编码器，实现了高维数据向低维空间的压缩编码。然后基于BP神经网络建立了低维编码序列与股票价格之间的回归模型。实验结果表明，深度自编码器提取特征的能力优于主成分分析法和因子分析法；相比较使用降维前的数据，使用编码后的数据用预测股票价格，模型可以减少计算开销，并且获得更高的预测精度。     

基于视觉感知的智能扫地机器人的垃圾检测与分类

目的 为了提高扫地机器人的自主性和智能化程度，为扫地机器人配备视觉传感器，使其获得视觉感知能力，通过研究有效的垃圾检测分类模型与算法，实现对垃圾的定位与识别，引导扫地机器人对垃圾进行自动识别与按类处理，提高工作的目的性和效率，避免盲动和减少能耗。方法 选择检测速度较快的YOLOv2作为主网络模型，结合密集连接卷积网络，嵌入深层密集模块，对YOLOv2进行改进，提出一种YOLOv2-dense网络，该网络可以充分利用图像的高分辨率特征，实现图像浅层和深层特征的复用与融合。结果 测试结果表明，智能扫地机器人使用本文方法可以有效识别不同形态的常见垃圾类别，在真实场景中，测试识别准确率为84.98%，目标检测速度达到26帧/s。结论 实验结果表明，本文构建的YOLOv2-dense网络模型具有实时检测的速度，并且在处理具有不同背景、光照、视角与分辨率的图片时，表现出较强的适应和识别性能。在机器人移动过程中，可以保证以较高的准确率识别出垃圾的种类，整体性能优于原YOLOv2模型。     

卷积神经网络用于关节角度识别与姿势评估

为了快速准确地输出各种工作姿势风险评估结果，提出采用Kinect v2与卷积神经网络识别人体各关节角度，并输出标准姿势风险的评估得分。首先使用亚像素角点提取的棋盘标定算法标定Kinect两个摄像头，其次使用改进后的双边滤波对深度图像去噪，使用卷积神经网络识别人体关节二维位置，结合深度信息获取实际三维坐标并计算人体关节角度，最后输出姿势风险评估得分。通过两种实验分别验证了提出的Kinect角度识别与姿势评估的准确性，表明该方法关节角度识别与姿势风险评估的准确率均较高，是一种低成本、高可靠性的姿势评价方法，具有一定的科学意义和工程应用价值。     

面向智能客服的句子相似度计算方法

针对金融领域中智能客服的句子相似度计算方法进行了研究。利用基于词性的分词纠正模型减少中文歧义词、金融相关词汇的分词错误；通过词向量方法和循环神经网络分别提取词语级和句子级的语义特征，并且得到句子向量；用融合层计算出句子向量间的差异特征；对差异特征进行降维和归一化得到句子相似度计算结果。实验结果表明，该方法具有较高的准确率和[F1]值。     

时态神经网络模型及其在股票分类预测上应用

对具有时间属性的数据进行数据挖掘称为时态数据挖掘，用以发现数据在时间上的知识，当数据变化不规律时，如股票交易数据,就很难发现有价值的规律与规则。而神经网络具有并行、容错、可以硬件实现以及自我学习的优点,可作为股票分类预测应用的一种方法。通过将股票数据与时态型相结合，将股票数据转换成时态型股票数据，提出时态神经网络模型的分类方法，对收集的若干上市公司十年内的股票数据进行分析，构建了时态股票数据神经网络分类器对股票进行分类预测。经过实验验证，相比改进前的神经网络和支持向量机方法，该分类器具有更高的分类准确率。结果证明，这种时态数据神经网络模型对于多只股票的分类预测是非常有效的，可以很好地运用到股票市场的分类预测中。     

Purex流程铀钚分离工艺单元数学模型研究

在铀钚分离工艺单元单级数学模型和混合澄清槽瞬态数学模型的基础上，建立了以U(Ⅳ)-N₂H₄为还原反萃剂、混合澄清槽为萃取设备的Purex流程铀钚分离工艺单元数学模型，开发了计算机模拟程序，并使用台架实验数据对程序的可靠性进行了验证。结果表明，模拟程序的计算值和实验值符合良好。在此基础上，利用模拟软件对铀钚分离工艺单元的工艺参数进行了计算分析，结果表明：1BX₁加入位置、1BS和1BX₂酸度对钚反萃率无太大影响，但1BX₁加入位置和补萃级数对钚中去铀系数SF_U/Pu有一定影响。