Meta-Learning Applied to the Selection of the Classification Methods in Industrial Images

In industrial production, control of quality of product and analysis of state of the staff in charge are important factors, nevertheless, the collection and analysis of data imply large amounts of time and may involve a risk to health of the staff in case of quality control, to deal with these tasks, image capturing and classification tools have been used, however, there is a challenge to identify the most appropriate classification method when taking into account the type of image being studied, the challenge is greater when it is necessary for a system to process different products with different classification objectives. This paper presents a methodology based on Meta-Learning and CNN for the identification of the appropriate methods of classification of industrial images. As an object of study images of hot-rolled steel strip, shear pad of wagon train, welds x-rays, aluminum wheel x-rays and human faces were used, obtaining 96% accuracy, 99.7% AUC and 96.5% Fmeasure.