Advanced Materials Research Vols. 998-999

Abstract: LPCC and MFCC are methods of extracting voice characteristic, and they are based on pronunciation models and human auditory characteristics. In this paper, both of the two characteristics are used, LPCC and the First Order Differential are used to describe the dynamic changes of speaker channels; MFCC and the First Order Differential are used to describe the audible frequency characteristics of human ears, and the characteristics of input voice are extracted by using Speech Processing Toolbox in MATLAB, and VQ and HMM are combined to applying to speaker recognition, and the experiment result showed that the performance of the speak recognition is obviously improved.

Abstract: At present, in the most of the digital image system, the input image is used to freeze the multi-dimensional image scanning way again into a one dimensional signal, then carries on the processing, storage, transmission and processing. Finally tend to form multi-dimensional image signal, and the image noise will also be decomposed and compounded. Electrical systems and outside influences in these procedures will enable precise analysis of complexity of image noise. According to off-line learning method of neural networks, this paper focus on the noise filter in the 3Dreconstruction process in order to make the image clearer.

Abstract: Spatial topological relation is an important and typical multilayer spatial relation, when Apriori is used to mining spatial constraint topology association rules, it will has some repeated computing. And so this paper proposes an algorithm of spatial constraint topology association rules mining based on complement set, which is used to mining spatial multilayer transverse association rules with constraint condition from spatial database. This algorithm generates candidate frequent topological itemsets with constraint condition not only by down-top search strategy as Apriori, but also by computing complement set of candidate from down-top search strategy, which is suitable for mining any spatial topological frequent itemsets with constraint condition. This algorithm compresses a kind of spatial topological relation to form an integer. By the way, firstly, the algorithm may efficiently reduce some storage space when creating mining database. Secondly, the algorithm is fast to obtain topological relation between two spatial objects, namely, it may easily compute support of candidate frequent itemsets. Finally, the algorithm may fast generate candidate via double search strategy, i.e. one is that it connects (k+1)-candidate frequent itemsets with constraint condition of k-frequent itemsets as down-top search strategy, the other is that it computes complement set of (k+1)-candidate frequent itemsets with constraint condition. The result of experiment indicates that the algorithm is able to extract spatial multilayer transverse association rules with constraint condition from spatial database via efficient data store, and it is very efficient to extract any frequent topology association rules with constraint condition.

Abstract: In order to improve the accuracy of some items in vehicle inspection, the extraction method of automobile center axis based on image processing technologies was studied. The commonly used methods of detecting the center axis of automobile were introduced, including approximation method and linear fitting method, which advantages and shortcomings were analyzed. The center axis detection method based on centroid method was proposed. The mass center of automobile outline was gotten based on the binary image. By detecting the nearest two points in the outline fitting line from the center, the center axis of automobile was gotten. This method can greatly reduce the calculation amount and have certain application value.

Abstract: Threshold segmentation method was widely applied in image process and the selection of threshold affected the final results of image segmentation to a large extent. In order to improve the accuracy and the calculation speed of image segmentation, an Otsu threshold segmentation method based on genetic algorithm was offered. According to the threshold and the gray scale values of pixels, the pixels were divided into two categories, and then the genetic algorithm was used to find the maximum variance between clusters and obtain the optimal threshold of segmentation image. The experimental results show that this method can be used to segment the image effectively, which make the basis for image processing and analysis in the next step.

Abstract: In the segmentation algorithms of the depth image, because the object and its support surface are continuous in the depth data ,the traditional method of edge detection methods can’t detect the edge between the object and its support surface. To solve this problem, the segmentation algorithm of the depth image is studied in this paper. Firstly, we use canny operator to detect the edge the of depth image of the scene. Then the depth image of the scene is transformed into points of a 3-D space coordinate and normal vector is calculated for each point. The method of calculation the direction of the normal vector is used to determine the point of which belongs to the support surface area, which determine the support surface area of the scene. Finally, we detect image edge of the image that the support surface area is extracted, and fuse the result of canny operator edge detection and edge of the image that the support surface area is extracted. Experiments show that the segmentation algorithm works well, which the problem of detection the edge between the support surface area and the object and can also achieve a good depth image segmentation.

Abstract: IE is not only the important content in (PRA）, but also is the foundation of PRA. Analyzing the IE includes: identification, screening and grouping, however, there is no scientific method for screening IE. Therefore, a method for screening IE based on the improved Borda ranking method has been proposed. Firstly, we apply the Borda ranking method to make up the deficiency of the risk evaluation index. Also, combining with the characteristics of IE, which is the beginning of the event chains, we improved the Borda ranking method.Secondly, taking the Filling system as the research object, we analyze the portion IE causing the propellant leak, and apply the improved method to rank and screen IE. Lastly, comparing the results of improved method with index and traditional Borda method, also combining the actual engineering situation in filling system, we validate the correctness and effectiveness of the improved method.

Abstract: With the vigorous development of the Internet information age, work efficiency can be improved by finding the information needed accurately and quickly. Therefore, it is of vital importance to make an ordering for the relevant information web pages that are provided by the Internet. This paper proposes a kind of PageRank algorithm based on matrix partitioning to complete the ordering of relevant information web pages and applies this algorithm in the calculation cases whose experimental results on the aspect of improving PageRank computational efficiency show matrix partitioning can reduce iterations and improve computational efficiency.

Abstract: As the earliest practical controller, PID controller has more than 50 years of history, and it is still the most widely used and most common industrial controllers. PID controller is simple to understand and use, without a prerequisite for an accurate model of the physical system, thus become the most popular, the most common controller. The reason why PID controller is the first developed one is that its simple algorithm, robustness and high reliability. It is widely used in process control and motion control, especially for accurate mathematical model that can be established deterministic control system. But the conventional PID controller tuning parameters are often poor performance, poor adaptability to the operating environment. The neural network has a strong nonlinear mapping ability, competence, self-learning ability of associative memory, and has a viable quantities of information processing methods and good fault tolerance.

Abstract: This paper analyzes two classic indoor location methods based on Radio Frequency Identification (RFID),LANDMARC and VIRE,and presents a location method based on the lagrange interpolation.Make up for the inadequacy of VIRE algorithm uses linear interpolation to get the virtual reference tags signal strength value lead to inaccurate positioning.Then making improvement to weight definition to make it more accurately reflect the weight of each selected nearest label to get more accurate positioning results. The experimental results show that the improved algorithm is better than the original algorithm has better positioning results and stability.