Applied Mechanics and Materials Vols. 241-244

Abstract: Discussed the feasibility and implementation methods of path based hierarchy aggregation in the DS-TE environment, established the mathematical model for the path based hierarchy aggregation by using analog circuit switching method. The results of experiments show that the path based hierarchy aggregation method can normally effectively improve the delay performance and the extensibility of the network systems.

Abstract: Studies propose to combine standard SVM classification with the information entropy to increase SVM classification rate as well as reduce computational load of SVM testing. The algorithm uses the information entropy theory to per-treat samples’ attributes, and can eliminate some attributes which put small impacts on the date classification by introducing the reduction coefficient, and then reduce the amount of support vectors. The results show that this algorithm can reduce the amount of support vectors in the process of the classification with support vector machine, and heighten the recognition rate when the amount of the samples is larger compared to standard SVM and DAGSVM.

Abstract: Most algorithms for the high-dimensional data clustering are not intuitive and the clustering results are difficult to explain. To solve these problems, a new method based on the interactive visualization technology was proposed in this paper. First, the entropy-weight was adopted to determine the main attributes and how to arrange them. Every data was described in an improved radar chart in which polar radius stood by attribute values and polar angles stood by the attribute weights. Then the points in the radar chart were clustered through applying an improved k-means algorithm. The number of clusters was not given before. And initial centers were optimized according to the point density and their distance. Finally, the experiment showed that the improved radar chart reflected the distribution of the data better and that the improved k-means algorithm was more efficient and accuracy.

Abstract: Speech under G-force which produced when speaker was under different acceleration of gravity was analyzed and researched, considered as principal part and stressed part to research. An isolated word recognition approach was proposed which combined difference subspace means with dynamic time warping technique. The method recognized speech under G-force by constructing a difference subspace to remove the stressed part. Dynamic time warping technique was adopted to make all feature vectors of one word in the training set have equal length, and a corresponding decision criterion was suggested. For a small vocabulary including 15 words, the method obtained the average recognition rate of 98.3%, which almost equal to the rate in normal environment. The method not only worked well in normal conditions but also had good performance for speech under G-force.

Abstract: In this paper, a novel technology, which detects and forecasts human's abnormal behavior, is proposed. Taking an abnormal behavior, the robbery, as an example, the accuracy of the technology is demonstrated successfully. There are 3 innovations in this paper. A) A robbery incident is divided into 5 phases occurred orderly. By its order, a robbery is detected. B) By human body's motion character, the human body's speeds are measured in different areas and at different times. It improves the accuracy rate of a classifier. C) The technology can forecast an abnormal human behavior in a short time advance. Experiment shows that the technology detecting and forecasting accuracy are high enough to be used in practice.

Abstract: This paper presents a new illumination normalization method for robust face recognition under varying lighting conditions. In the proposed method, the illumination component is estimated by applying nonlocal total variation model in the logarithmic domain, and then the reflectance component is obtained based on reflectance model. The proposed method restrains the halo effect effectively while preserves the adequate texture information on the reflectance images. As an illumination invariant facial features, the reflectance images are directly utilized for face recognition. Experimental results on Yale face database B and CMU PIE database show that the performance of proposed method is robust and reliable in illumination invariant face recognition.

Abstract: With the development of the Semantic Web, ontology has become the primary means of expression of many fields of knowledge. Introducing the Semantic Web technology into the field of search engine is a valuable research topic. In order to meet the complex semantic retrieval demands, the paper proposes a search engine model based on multi-domain ontology, the model using ontology mapping rewrite the user query to achieve multiple ontology query, and provide a richer and accurate semantic information for the retrieval of cross-domain knowledge; And the paper proposes a method of cross-domain ontology annotation, providing a basis for the user semantic retrieval. The experimental results show that the search results improve the precision and recall rate.

Abstract: This paper describes the study of techniques used in hand gesture recognition in sign language interpretation. The study is discussed from two aspects: the process of hand gesture recognition and the techniques of feature extraction in hand gesture recognition. The techniques of feature extraction in hand gesture recognition are grouped into five categories: Hidden Markov Model (HMM), Artificial Neural Networks (ANN), Support Vector Machines (SVM), Dynamic Bayesian Network (DBN), and Dynamic Time Warping (DTW). The main ideas and the application of each technique are described in detail.

Abstract: In this paper, an embedded design method is presented for the access control system in the intelligent community, combining its characteristics of the application. Using speech recognition technology, intelligent community can control access more timely, conveniently, effectively, and save a lot of manpower. The experimental results show that the correct recognition rate of voice command is up to 92% or more. The access control system can be widely used in various kinds of intelligent buildings, with a good prospect.

Abstract: By multi-rate networked control systems (NCS), we mean the sampling periods of the sensor, the controller and the actuator in networked control systems are not the same, that is to say there are more than one sampling rate in networked control systems. For the long time delay multi-rate NCS with event-driven controller and actuator, a stochastic discrete model is established under operator. The state feedback control laws for the multi-rate NCS in operator domain are designed by using a dynamic programming approach. The derived optimal LQG controller can be used as a delay-compensator for multi-rate NCS with long time delays. An example is given to verify the theory results of this paper.