Sistema computacional baseado em reconhecimento de padrões para classificação de concretos

Abstract

The proportion of materials used in concrete composition has great importance in order to build a structure and public security, since materials have influence on concrete properties, in fresh and hardened states. The identification of concrete traces is usually carried by human experts with high levels of experience, but their results are influenced by physical and subjective factors. This paper presents a proposal based on pattern recognition to identify different concrete traces through out images and their texture features. The achieved results demonstrated the possibility of identifying concrete traces through application of pattern recognition techniques to images. The best achieved rate was 87.1% of accuracy using the SURF descriptor.