Análise do desempenho de perceptrons multicamada modernos na classificação de objetos do cotidiano.

Abstract

Electronic devices that assist with household tasks have been gaining popularity in recent years and help people save time in their routines. Individuals with physical limitations need even more support in their daily activities. However, there are not many solutions currently available that can perform tasks normally performed by humans, due to hardware limitations, for example. One important skill is recognizing objects in a visual scene. Thus, this research aims to evaluate the performance of some multilayer perceptron (MLP)-based models in classifying images of objects commonly found in domestic environments, to verify the effectiveness of these solutions in this application context. Classification experiments were conducted with the models, observing the metrics obtained, such as accuracy, training and testing times, to qualify the performance. The analysis of the models confirmed the ability to classify objects with a good accuracy rate. The results obtained indicate that it is possible to apply MLPs in solutions to assist with household activities, reducing the computational cost of implementation in relation to more complex models.