Revista ELECTRO
Vol. 46 – Año 2024
Artículo
TÍTULO
Costo Computacional de Redes Neuronales Profundas para Reconocimiento de Emociones Mediante Expresiones Faciales
AUTORES
Amparán-Ortega, N.C.; Corral-Sáenz, A.D.; Ramírez-Quintana, J.A.
RESUMEN
El reconocimiento de emociones es un área destacada en el contexto de la Inteligencia Artificial (IA) ya que tiene una amplia aplicación en computación afectiva. El enfoque que ha generado mejores resultados en el reconocimiento de emociones es la aplicación de las redes n euronales profundas en análisis de expresiones faciales; sin embargo, está asociado con una carga computacional considerable. Este estudio compara el costo computacional de las arquitecturas VGG16, ResNet50 e InceptionV3 usando la base de datos JAFFE. Se evalúa el rendimiento en dos entornos de hardware: Jetson TX2 y una computadora con procesador Intel Xeon E5-1603 v3, centrado en la velocidad de procesamiento (FPS) y consumo de memoria. Los resultados muestran una precisión del 81.25%, 71.88% y 43.75% par a VGG16, InceptionV3 y ResNet50, respectivamente. La computadora con procesador Intel Xeon presentó un rendimiento superior a la tarjeta Jetson TX2, que no utilizó la GPU para la inferencia.
Palabras Clave: VGG16, ResNet50, InceptionV3, FPS, redes neuronal es profundas.
ABSTRACT
Emotion recognition is a prominent area in the context of Artificial Intelligence (AI) due to its wide application in affective computing. The approach that has yielded the best results in emotion recognition is the application of deep neural networks in facial expression analysis; however, it is associated with a considerable computational load. This study compares the computational cost of the VGG16, ResNet50, and InceptionV3 architectures using the JAFFE database. The performance is evaluated in two hardware environments: Jetson TX2 and a computer with an Intel Xeon E5-1603 v3 processor, focusing on processing speed (FPS) and memory consumption. The results show an accuracy of 81.25%, 71.88%, and 43.75% for VGG16, InceptionV3 and ResNet50, respectively. The computer with the Intel Xeon processor demonstrated superior performance compared to the Jetson TX2, which did not utilize the GPU for inference.
Keywords: VGG16, ResNet50, InceptionV3, FPS, deep neural networks.
REFERENCIAS
[1] S. K. Khare, V. Blanes-Vidal, E. S. Nadimi, and U. R. Acharya, “Emotion recognition and artificial intelligence: A systematic review (2014 –2023) and research recommendations,” Information Fusion, vol. 102, p. 102019, Feb. 2024, doi: https://doi.org/10.1016/j.inffus.2023.102019.
[2] P. Ekman, R. W. Levenson, and W.V. Friesen, “Automatic nervous system activity distinguishes among emotions”, Science, vol. 221, no. 4616, pp. 1208-1210, Sep. 1983.
[3] P. Naga, S. D. Marri, and R. Borreo, “Facial emot ion recognition methods, datasets and technologies: A literature survey,” Materials Today: Proceedings, Jul. 2021, doi: https://doi.org/10.1016/j.matpr.2021.07.046.
[4] Y. S. Can, B. Mahesh, E. André, “Approaches, applications, and challenges in physiological emotion recognition —a tutorial overview,” Proceedings of the IEEE, pp. 1-27, 2023
[5] N. Bukhari, S. Hussain, M. Ayoub, Y. yu, and A. Khan, “A Deep Learning-based Framework for Emotion Recognition usi ng Facial Expression”, Pakistan Journal of Engineering and Technology, vol. 5, no. 3, pp. 51-57, 2022
[6] M. Lyons, S. Akamatsu, M. Kamachi, J. Gyoba, “Coding facial expressions with Gabor wavelets”, Proceedings Third IEEE International Conference on Automatic Face and Gesture Recognition, pp. 200-205, 1998. Revista ELECTRO, Vol. 46, 2024, pp. 333-338 https://itchihuahua.mx/revista_electro ISSN Electrónico: 3061-774X 338
[7] K. Simonyan and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” Computer Science, 2014, doi: https://doi.org/10.4 8550/arXiv.1409.1556
[8] S. Bangar, “VGG-Net Architecture Explained,” Medium, Jun. 28, 2022. [Online]. Available: https://medium.com/@siddheshb008/vgg-net-architecture-explained-71179310050f
[9] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Ima ge Recognition,” arXiv (Cornell University), Dec. 2015, doi: https://doi.org/10.48550/arxiv.1512.03385
[10] Nayan Chaure, “Variants of ResNet: A Comparative Analysis-Nayan Chaure-Medium,” Medium, Apr. 27, 2024. https://medium.com/@nayanchaure601/variants-of-resnet-a-comparative-analysis-63fdc1573b34
[11] C. Szegedy et al., “Going Deeper with Convol utions,” arXiv (Cornell University), Sep. 2014, doi: https://doi.org/10.48550/arxiv.1409.4842
[12] J.J. Muñoz, “Manual de usuario y herramientas para introducción al uso y desarrollo de redes neuronales en el sistema embebido NVIDIA Jetson TX2”, Instituto Tecnológico de Chihuahua, pp. 1-28, 2024
CITAR COMO:
Amparán-Ortega, N.C.; Corral-Sáenz, A.D.; Ramírez-Quintana, J.A., "Costo Computacional de Redes Neuronales Profundas para Reconocimiento de Emociones Mediante Expresiones Faciales", Revista ELECTRO, Vol. 46, 2024, pp. 333-338.
VERSIÓN PDF