Papers by Keyword: Pattern Recognition

Abstract: The forming limit diagram (FLD) is at the moment the most important method for the prediction of failure within sheet metal forming operations. Key idea is the detection of the onset of necking in dependency of different sample geometry. Whereas the standardized evaluation methods provides very robust and reliable results for conventional materials like deep drawing steels, the determined forming limits for modern light materials are often too conservative due to the different failure behavior. Therefore, within this contribution a new and innovative approach for the identification of the onset of necking will be presented. By using a pattern recognition-based approach in combination with an optical strain measurement system the complete strain history during the test can be evaluated. The principal procedure as well as the first promising results are presented and discussed.

Abstract: In this paper, a novel close to real-time artificial intelligent system for enumerating Total Viable Bacteria (TVB) in drinking water was developed by using pattern recognition and machine vision technology. In order to identify the viable bacteria accurately, four shape features including circularity ratio, eccentricity, rectangularity, and compact degree, and four color features (G_Rsd, B_Rsd, H_Rsd, S_Rsd) of the stained viable bacteria image were extracted. An optimal artificial neural network was used as the bacterial recognition classifier, whose inputs were the extracted feature parameters and output was bacteria signal or non-bacteria signal. By using this intelligent system, TVB counts in each sample can be enumerated within 1 h, but the traditional Aerobic Plate Count (APC) method will take us 48 h. The comparative test also indicated that the counting results by two methods are closely correlated (R²=0.9942). This close to real-time accurate information may contribute to melioration and instauration of drinking water safety systems and risk management for TVB.

Abstract: This article describes the classification of China bank check, green fluorescent reaction under ultraviolet light source illumination of the featured regions—— invisible flower mission and motif pattern, and recognize the accuracy of detecting bank check feature area by signal processing and pattern recognition of UV green fluorescent response signal receiver at particular locations. It described the light filter for bank check, which is different from RMB Banknote

Abstract: In order to study different types of partial discharge inspired by defects in GIS and increase the rate of correct identification on defects, four kinds of typical insulation defects physical model are designed based on the insulation defects of 110 kV GIS and its partial discharge characteristics. Ten feature parameters including the signal peak and kurtosis are acquired from 222 groups of partial discharge signal data, and recognized by BP neural network which is optimized by input genetic algorithm. Recognition results show that this method works well, owning a higher recognition rate than adaptive momentum BP neural network

Abstract: Aim at the real-time problem of industrial robot vision system, design a embedded robot vision system based on DSP microprocessor. This system can use CCD camera and the ultrasonic sensor to collect the target environment information. It also can use the processor DSP to process the images and recognize target. And then through the communication module, send results in the form of wireless to the upper computer, providing target object information for robot control layer. This system completes the software and hardware system design, image collection & processing and robot control, as well as meet the real-time requirements of machine vision system.

Abstract: By analysis the difference of applying the rough set method and the neural network method to pattern recognition, a improved recognition method that the rough set method is the front system of neural network was produced. the advantages of this method is that the knowledge representation system is reduced without affecting the recognition precision, so the complexity of neural network system and the time of calculating the attribute value is declined ; at the same time ,the neural network as the postpositional system has the tolerance and anti-jamming capability, but it is difficult to do this with rough set method. The example about how to combine these two methods and conclusions from this combination was given.

Abstract: To have a comprehensive summary of existing methods for damage identification of transmission tower so as to provide systematic and effective information, from aspeces of data sources and analytical methods, the existing methods of damage identification were summarized to sort out the technical evolution of the various damage identification methods including the static and dynamic damage identification and inversion and pattern recognition methods.Then, the application status, advantages and disadvantages of various methods were analzed, and the improvement ideas and methods were summarized. Finally, establishment of four-in-one multi-system damage identification method was proposed based on construction monitoring, loads tsets, health monitoring and artificial patrol, which is the development trend of the transmission tower structure damage identification.

Abstract: Quality in education has been the subject of many debates, be it among managers in schools, in the media or in literature. However, the literature appears not to include methods or techniques for exploring databases to obtain classifications for this quality; nor is there a consensus as to the definition of “education quality”. To address these issues, this article proposes a methodology similar to the KDD (Knowledge Discovery in Databases) to classify Education Quality in schools comparatively, based on grades scored in school performance tests. For the purposes of this study, the test used was the Prova Brasil examination, which is part of the Basic Education Development Index (IDEB) used in Brazil. The methodology was applied to public municipal schools in the town of Araucária in the metropolitan district of Curitiba in Paraná State. Seventeen schools that offer elementary and junior high school education were included. All the grades of every student were considered from the early and later years at the schools. During the Data Mining stage, the main stage of the KDD process, three comparative methods were used for Pattern Recognition: Artificial Neural Networks, Support Vector Machines and Genetic Algorithms. These methods supplied satisfactory results in the classification of schools represented by way of a “Quality Label”, with Artificial Neural Networks having the best performance for the problem in question. With this quality label, educational managers can decide on which measures to adopt at all the schools to help them achieve their goals.

Abstract: A material on-line identification system based on THz time-domain spectroscopy using VC was developed in this paper. The main functions including data reading, linear motor driving, the calculation of optical parameters, the display and management of curve, the management of interface, the analysis of historical data , the maintenance and updating of database, and on-line identification based on principal component analysis were realized. As the prototype of material detection application, the material whose corresponding type characteristic spectrum has already existed in the database could be well identified.

Abstract: An electronic tongue was employed to detect different brands of Chinese rice wine. The results showed that all of the seven classes of Chinese rice wine can be discriminated by Discriminant Factor Analysis (DFA) and Principal Component Analysis (PCA). Based on Artificial Neural Network (ANN), the electronic tongue can predict the marked age of Chinese rice wine, and the accuracy of prediction was above 90% averagely.