Title: Children's expression recognition based on a multiscale mixed attention mechanism

Authors: Wenming Wang; Mideth Abisado

Addresses: College of Computing and Information Technologies, National University, Manila, 1008, Philippines; Experimental Training Teaching Management Department, West Anhui University, Lu'an, 237012, China ' College of Computing and Information Technologies, National University, Manila, 1008, Philippines

Abstract: A multiscale mixed attention mechanism network is proposed to aim at the problems that ordinary convolutional neural networks are difficult to extract effective features and have insufficient generalisation ability in children's facial expression recognition. Firstly, a multiscale convolution unit is introduced into the network to enhance both the depth and width of the network and extract feature information of different scales of children's expressions. Secondly, a mixed attention mechanism is proposed, which combines the features extracted from multiscale convolution into a mixed attention mechanism to solve the problems of insufficient generalisation ability, recognition accuracy, and efficiency of facial expression recognition models in complex environments. The experimental results show that this method improves the accuracy of children's facial expression recognition, has a better recognition effect on a large-scale facial expression dataset, and has a strong generalisation ability, which has certain application scenarios.

Keywords: children's expression recognition; multiscale convolution; mixed attention mechanism; channel attention; spatial attention; convolutional neural networks; CNNs.

DOI: 10.1504/IJSNET.2023.134288

International Journal of Sensor Networks, 2023 Vol.43 No.2, pp.116 - 127

Received: 05 May 2023
Accepted: 07 May 2023
Published online: 17 Oct 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article