Aplicação de modelos de Markov ocultos na obtenção de taxas de mortalidade das larvas do mosquito da Dengue

AUTOR(ES)

Kleber Padovani de Souza

FONTE

IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia

DATA DE PUBLICAÇÃO

26/02/2010

RESUMO

In order to prevent the proliferation of Dengue transmitter - scientifically named as Aedes ae- gypti - and therefore decrease human contamination by such insect, many larvaecides have been developed recently. Researchers from Dom Bosco Catholic University evaluate the efectiveness of vegetal-derived substances capable to combat such animal larvae. Death rate is among many evatuation criteria of these substances. Such rate may be obtained by periodic visual observation of larvae subjected to some substance. However, human limitations, such as relative conclusions and exhaustion susceptibility, can damage a larvaecide investigation. Thus, a computer system which can automatically observe larvae while experiments are performed is a great support for science. For this type of systems, computer vision (CV) algorithms may be appropriated once larva observation is a visual-based task. In addition, such systems can be substantially improved when using techniques from other areas, specially pattern recognition and machine learning. In this context, the research group of CV from the aforementioned university created a project, named LARVIC, so as to develop an automatic system based on CV. In order to collaborate with this project, the present work investigate the applicability of Hidden Markov Models in larvae behaviour classification. For this purpose, image sequences of dead and alive larvae inside a specific recipient were captured and then processed by algorithms of CV and related areas, producing relevant data which allowed the analysis of the proposed application. Three aspects were considered during the examination, namely: 1) initial probabilities of models, which are base for parameters reestimation, 2) stop criterion of reestimation stage, and 3) how to use the models as a classifier. Partly random and previously computed initial probabilities were used. Stronger criteria were examinated to analyse standard criterion. A combination with models and machine learning techniques was built to increase the perfomance of classifiers based on the major technique of this work. External libraries and additional implementations made experi- ments possible. Based on error rate and performance graphs, we conclude with this work that the performance of classifiers based on Hidden Markov Models with no combination, though inferior to some classifiers, was substantial for the approached application.

ASSUNTO(S)

modelos de markov ocultos comportamento animal reconhecimento de padrões computabilidade e modelos de computacao hidden markov models animal behavior pattern recognition

Documentos Relacionados

• Reconhecimento de fala continua usando modelos ocultos de Markov
• Estimação paramétrica e não-paramétrica em modelos de markov ocultos
• Decomposição de sinais eletromiográficos de superfície utilizando Modelos ocultos de Markov
• PrediÃÃo de palavras baseada em modelos ocultos de Markov
• Desenvolvimento de arquitetura para sistemas de reconhecimento automático de voz baseados em modelos ocultos de Markov