Extração de conhecimento em bases de dados espaciais: algoritmo CHSMST+

Pós-graduação em Ciência da Computação - IBILCE

Machine vision to identify broiler breeder behavior

Animal behavioral parameters can be used to assess welfare status in commercial broiler breeders. Behavioral parameters can be monitored with a variety of sensing devices, for instance, the use of video cameras allows comprehensive assessment of animal behavioral expressions. Nevertheless, the development of efficient methods and algorithms to continuously identify and differentiate animal behavior patterns is needed. The objective this study was to provide a methodology to identify hen white broiler breeder behavior using combined techniques of image processing and computer vision. These techniques were applied to differentiate body shapes from a sequence of frames as the birds expressed their behaviors. The method was comprised of four stages: (1) identification of body positions and their relationship with typical behaviors. For this stage, the number of frames required to identify each behavior was determined; (2) collection of image samples, with the isolation of the birds that expressed a behavior of interest; (3) image processing and analysis using a filter developed to separate white birds from the dark background; and finally (4) construction and validation of a behavioral classification tree, using the software tool Weka (model 148). The constructed tree was structured in 8 levels and 27 leaves, and it was validated using two modes: the set training mode with an overall rate of success of 96.7%, and the cross validation mode with an overall rate of success of 70.3%. The results presented here confirmed the feasibility of the method developed to identify white broiler breeder behavior for a particular group of study. Nevertheless, more improvements in the method can be made in order to increase the validation overall rate of success. (C) 2013 Elsevier B.V. All rights reserved.

Deep learning architecture for gesture recognition

Pós-graduação em Engenharia Mecânica - FEG

Estratégia de Investimento baseada em padrões descobertos por mineração de dados em séries históricas de preços diários do Indice IBOVESPA e das ações da PETROBRAS-PN (PETR4.SA)

With the increase of stakeholders and consequently increase of amount of nancial transaction the study of news investment strategies in the stock market with data mining techniques has been the target of important researches. It allows that great historical data base to be processed and analysed looking for pattern that can be used to take a decision in investments. With the idea of getting pro t more than the real indexs' gain, we propose a strategy method of transactions using rules built by algorithm classi cation. For that, diary historical data of Ibovespa index and Petrobras stocks are organized and processed to nding the most important attribute that act decisively when taking a investment decision.To test the accuracy of proposed rules, a non real portfolio management is created, showing the decisions' performance over the real index and stocks' performance. Following the proposed rules, the results show that the strategy of investment give me back a high return that Stock market's return. The exclusive characteristics of algorithms maximize the gain inside the analysed time allowing to determine the techniques' return and the number of the days necessary to double the initial investment. The best classi er applied on the time series and its use on the propose investments strategy will demand 104 days to double the initial capital

Análise da cintilação ionosférica no Brasil empregando GNSS e técnicas de mineração e visualização de dados

Pós-graduação em Ciências Cartográficas - FCT

O campo da Ciência da Informação: contribuições, desafios e perspectivas da mineração de dados para o conhecimento pós-moderno

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)