Estratégias de aprendizado visando melhorias nos processos de classificação e de controle de qualidade na indústria do ramo alimentício

Abstract

Considering the great competition among industries, one of the main factors that make companies market leaders is the quality of their products. However, the techniques applied to quality control are often faulty or inefficient, due to the great dependence on the human factor, which makes the applied procedures tiring and highly susceptible to errors. Moreover, in the context of industry 4.0, the use of technologies to improve the evaluation of these products becomes increasingly essential. Therefore, this work aims to learn the most appropriate descriptors and pattern classifiers for automatic product classification and quality control in food industries, more specifically regarding cookies. For this, an extensive experimental evaluation was performed, considering different learning approaches (traditional and based on convolutional neural networks). From the obtained results, it is possible to observe that the proposed methodology can provide a more effective quality control for the company, reaching accuracies of up to 99%. It is possible to avoid offering products that do not comply with the quality standards in the market, improving the credibility of the brand with the consumer, its profitability and consequently its competitiveness.