随着时代和科技的进步,语音处理技术已经广泛应用于日常生活的多个领域,实现人机交互的自然语言处理是当下科学技术的发展前沿.针对高校学生思政课程存在的问题及相关知识问答,通过采用语音识别及合成技术,借助科大讯飞人工智能平台的语音交互技术,设计符合高校师生思政课程知识问答的数据库,在小雅智能助教平台的基础上,实现了高校思政课虚拟教师语音答疑系统,积极探索互联网+教育新模式,让高校的思政教育从讲台走向"指尖",从耳边走入"心间".

"一带一路"沿线国家在长期的商贸往来和人文交流中留下了丰富的文化遗产.在新时代背景下,保护和利用好丝路文化遗产,是讲好丝路故事、推动各国文化交流的关键.该文章通过分析"一带一路"沿线文物保护现状以及相关研究成果,结合"一带一路"文化传播需求和日前兴起的增强现实(Augmented Reality,简称AR)技术的特点,介绍一种AR+"一带一路"的文物展示模式:面向"一带一路",构建一个沿线文物介绍数字媒体库,应用于一个基于AR技术的文化传播交互系统,打破丝路文化遗产展示在时间、空间上的限制,增强文物展示中的交互性,实现"一带一路"文化遗产的创造性转化.

三维建模是通过三维制作软件在虚拟三维空间构建出具有三维数据的模型,如今越来越广泛地应用到各个领域.通过对丝绸之路文物建模的实例,介绍了一套与时俱进的完整制作丝路文物三维模型的基本流程及方法,并通过结合AR等技术对模型进行展示与研究.文物三维模型的建立,为文物的数字化展示和鉴赏、虚拟数字博物馆的建立提供了帮助.

To improve the efficiency of Web service QoS prediction,providing users with high-quality Web services,a collaborative filtering method fused matrix factorization and nearest neighbor rule is proposed.Firstly,this method uses the location information of users and services to select neighborhoods of user and service,to overcome the inaccurate or insufficient use of location information in traditional QoS prediction algorithms.Then the neighborhood information is combined into the matrix factorization framework to improve the lack use of local information in matrix factorization technique,and gradient descent algorithm is used for QoS prediction.Finally,based on real Web service dataset WSRec,we conducted experiments compared with other algorithms and the experimental results validate the approach.

Recently the method of sparse representation based on prototype plus variation model is applied in face recog-nition effectively. This algorithm only considers about the holistic face, but ignores the effect of local feature in the entire process. To solve this problem, this paper adds the idea of block processing. The Borda count method is used to assign dif-ferent votes to various classes of the sub-modular. Then it classifies the faces according to the final votes. The experimen-tal results on AR datasets validate that this method performs better than other methods when the faces are partially obscured or under different light conditions.

Gist特征和PHOG特征分别作为描述场景图像全局性质和局部性质的特征,两者各自有不足之处。若能吸取两者优势互补,则场景图像分类准确率将得到提升。论文提出了一种基于D-S证据理论的融合Gist特征和PHOG特征的场景图像描述方法。该方法首先提取场景图像的Gist特征和PHOG特征,然后基于D-S证据理论得到融合的特征向量。使用支持向量机作为分类器,在OT场景图像库下,分别建立单一的Gist特征、单一的PHOG特征、传统串联融合特征以及证据理论融合特征的分类模型,采用正确率和混淆矩阵作为评价指标,分别进行四组实验。实验结果表明,论文提出的方法有效提高了场景图像分类的准确度。

Aiming at the problems of high cost and low efficiency in existing semantic Web service composition method based on graph search,this paper proposes an automatic composition optimization method of semantic Web service based on user request input closure and Complete Backward Tree(CBT).This method uses Request Input Closure Constructor(RICC) algorithm to construct user request input closure and determines whether the user requests can be met in the semantic Web service rule repository based on user request input closure in Closure_Complete Backward Tree_Optimal Valid Generation Path(C_CBT_OVGP) algorithm.When user request is not satisfied,it illustrates the request without a solution and then ends the algorithm.Otherwise,it prunes the backwark tree by user request input closure and setting object sets of new nodes so as to avoid unnecessary repetition to build the same node on a number of different branches.Example analysis result shows that the optimization algorithm improves the efficiency of semantic Web service composition,especially for dissatisfied user request.

A sequential minimal optimization(SMO)algorithm is a very effective method which can improve the training speed of support vector machine(SVM). However,the SMO algorithm is still quite slow in the large?scale datasets. In order to increase the training speed,an optimization strategy which can maintain the training accuracy is proposed. The strategy is to skip the part of the vector irrelevant to accuracy,prematurely finish cycle and relaxe KKT conditions so that it can shrink the working set. The results show that this strategy can significantly reduce the training time and the accuracy is still high in several datasets.

Aiming at the features of component choosing issue in Internetware,in the paper we map the component selection problem in Internetware(CSPI) of the component choosing in Internetware as the MMKP(multiple choices multi-dimensional knapsack problem) and propose a component selection approach based on GA.According to the function facet of the component,this method classifies and retrieves components from component library first,and obtains a component retrieval set,then introduces genetic algorithm,and selects from the retrieval set a couple of appropriate components to construct Internetware;And the idea of positive feedback in ACO is introduced to improve mutation operation in GA.At last,the experimental results show that the GA can select proper components to construct Internetware and the improved GA can do this work even better.

Aiming at the features of Web service environment such as open,deceptive and uncertain,a Web service trust evaluation model based on fuzzy theory is proposed.This model uses the rough set and information entropy theory method to evaluate the Web service trust,and can provide a more reasonable decision to select the trusted Web service.Finally,it shows that the model on the evaluation of Web service trust is more accurate and reliable through the processing of simulation experiment.

To meet the requirements of multi-directional choice,a new approach to the invariant feature extraction of handwritten amount Chinese characters was raised,with Ridgelet transform as its foundation.As far as this approach is concerned,first of all,the original images would be rotated to the Radon circular shift by means of Radon transform.On the basis of the characteristic that Fourier transform is row shift invariant,then,the one-dimensional Fourier transform would be adopted in Radon field to gain the conclusion that magnitude matrixes bear the rotation-invariance as a typical feature,which was pretty beneficial to the invariant feature extraction of rotation.When this was done,one-dimensional wavelet transform would be carried out in the direction of rows,thus achieving perfect choice of frequency,which made it possible to extract the features of sub-line in the appropriate frequencies.Finally,the average values,standard deviations and the energy values would form the feature vector which was extracted from the Ridgelet sub-bands.The approaches mentioned in the paper could satisfy the requirements from the form automatic processing on the recognition of handwritten amount Chinese characters.

The method of component description and retrieval is a hot subject of research in component-based software development. In order to describe the component more completely and accurately, an approach based facet and ontology is brought forward. The method proposed in the paper firstly choose the exactly type and application components, then choose components further through component's functional match, at last users make final choose after looking over all the candidate components. The description method of components is the base for users to find the exact components. An implementation of the proposed approach shows that the users can select components quickly and precisely.

A handwritten amount Chinese characters recognition algorithm based on wavelet packet transform is proposed.Firstly,wavelet packet transformation is used to decompose the character images whose proper partial decomposition tree can be chosen based on the variance characterization of the energy function.Secondly,each sub-image is divided into several local windows whose energy values are calculated to combine the feature vectors.Thirdly,the PCA transform is applied to all the feature vectors in order to determine a few significant features to reduce the samples of SVM.Finally,multi-class SVM is used for classification.The efficiency of this method is proved by the experiments which effectively improves the recognition rate of the amount Chinese characters.

本文结合多年的教学经验,从操作系统的教学方法、教学手段、实验设计等方面探讨了操作系统课程的理论与实践教学。经过多届学生的实践推广,证明这些措施能够有效促进学生对"操作系统原理"的基础理论学习和实践动手能力提高,进而增强其分析问题及编程能力。

Topic map is one of the hottest areas in information retrieval field in that it enables a user to navigate and access the documents he needs in an organized manner, rather than browsing through hyperlinks that are generally unstructured and often misleading. This paper proposes a new framework for information retrieval. Through actions of registry, discovery and access in grid information service discovery process, a grid space manager will find out the suitable services. And grid agent should remember the success and quality of the former cases by knowledge element mining and topic map building and reasoning, therefore the agent can gain experience of user's preference. We also detail the methods through which we can mine knowledge elements from web page resources' and to build extended topic maps including a new knowledge elements level.

Forms are different from common documents and more difficult to process. It is not very well done if processed by traditional document analyzing technology. The purpose of this thesis is to research on the key technologies of form analysis and understanding aiming at forms which are multi-type, large quantity, noisy, and variety of paper quality used by bank and revenue in our country. Based on analyzing knowledge of forms, a model of form based on object-orient sorting tree knowledge base and triple node representation is put forward. As for form feature extraction, hierarchical regulated hit or miss transform (HRHMT) is proposed. Feature extraction algorithm is also given. The algorithm is proved to be feasibility theoretically. Analysis on theoretical complexity and experiments shows their efficiency and superiority.

The identification of language and character type has been an active area of research after recognition of machine printed text.Research on identification of handwritten text and printed text is seldom conducted.But it is common used in recognition of form.For character type identification,manifold learning algorithm Locally Linear Embedding (LLE) is imported.A generalizing method and a parameters estimation method are proposed.Experiments in identification printed/handwritten Chinese characters and digits show that its performance is higher than Support Vector Machine (SVM) classification.The combination of dimensionality reduction of LLE and Linear Discriminant Analysis(LDA) classification achieves a similar accurate rate as or higher than the combination of LLE and SVM classification but runs much faster than it.

According to the traits of wavelet packet transformation,a handwritten amount Chinese characters recognition based on wavelet packet transformation is proposed.Firstly,wavelet packet transformation is used to decompose the character images whose proper partial decomposition tree could be chosen based on the variance characterization of energy function.Secondly,each sub-image is divided into several local windows whose energy values are calculated to combine the feature vectors.Thirdly,the PCA is applied to all the feature vectors in order to determine a few significant features to reduce the samples of SVM.Finally,SVM is used for classification. The efficiency of this method is proved by the experiments which effectually improves the recognition rate of the amount Chinese characters.

The identification of language and character type has been an active area of research after recognition of machine printed text. Research on identification of handwritten text and printed text is seldom conducted. But it is common used in recognition of form. For character type identification, manifold learning algorithm locally linear embedding (LLE) is imported. A generalizing method and a parameters estimation method are proposed. Experiments in identification printed/handwritten Chinese characters and digits show that its performance is higher than support vector machine (SVM) classification. The combination of dimensionality reduction of LLE and linear discriminant analysis (LDA) classification achieves a similar accurate rate as combination of LLE and SVM classification but runs much faster than it.

We introduce a system that incorporates a global optimization technique based on a genetic algorithm for dynamically selecting the set of classifiers and combination rules to use in the multiple classifiers combination method so as to integrate different classifiers' superiority and complementarities and improve the classification performance.To this end we test the proposed system on amount Chinese character recognition and investigate the performance of our system in comparison to a number of alternative combination strategies,the results indicate that significant gains can be obtained by integrating genetic algorithm into the multiple classifier systems.