Development of a hybrid system based on convolutional neural networks and support vector machines for recognition and tracking color changes in food during thermal processing. (31st August 2021)

• Record Type: Journal Article
• Authors: da Silva Cotrim, Weskley; Felix, Leonardo Bonato; Minim, Valéria Paula Rodrigues; Campos, Renata Cássia; Minim, Luis Antônio
• Is Part Of: Chemical engineering science. Volume 240(2021)
• Publication Date: 2021-08-31
• Subjects: Deep learning -- Machine learning -- Color change -- Browning; ANN Artificial Neural Networks -- AUC Area Under the Curve -- CIELab Commission Internationale de l'Éclairage Color Space -- CNN Convolutional Neural Network -- CVS Computational Vision System -- GAP Global Average Pooling -- HS Hybrid System CNN-SVM -- MAC Operations Multiplication and Accumulation Operations -- RBF Radial Basis Function -- RGB Red, Green, Blue Color Space -- ReLU Rectified Linear Unit -- SV Support Vectors -- SVM Support Vector Machine -- sRGB Standard RGB -- ΔE Color Difference; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; Chemical engineering; Periodicals; Electronic journals; 660
• DOI: 10.1016/j.ces.2021.116679
• Languages: English
• ISSNs: 0009-2509

Abstract

Highlights: The first study of CNN-SVM Hybrid System (HS) use in color change problems. The HS correctly recognize and classify the baking stages without human assistance. CNN-SVM based CVS is a noninvasive color change tracking tool. The HS overcome the performance of classical architectures. Abstract: In order to build a non-destructive tool for the recognition and classification of bread baking stages, based exclusively on the color changes of the bread crust, the present work aims to propose the development of a Hybrid System (HS) composed of a Convolutional Neural Network (CNN) and a Support Vector Machine (SVM). For training, validation and testing of the HS 374 images of the bread crust were used, distributed over seven baking periods. The results showed that the HS CNN-SVM was able to correctly recognize and classify the baking stages without human intervention, overcoming even models based solely on CNN. In addition, the HS CNN-SVM reduced convergence time and memory consumption, favoring its use in mobile or embedded systems. Finally, the HS CNN-SVM maintained the ability to extract color map characteristics present in CNN, allowing its use in the construction of process control systems for the food industry in which color changes are involved.