Work

Hybrid Attention-Based Optimized ResNet for Facial Expression Recognition and Its Application to Intelligent Assistant Systems

Year: 2026
Type: article
Abstract: With the development of computer vision and deep learning, facial expression recognition (FER) has become indispensable for parsing underlying emotional changes in practical machine-based applications; yet subtle expression discrimination remains challenging, with typical accuracies around 94.00–96.00% on CK+ and 70.00–73.00% on FER2013. This paper mainly investigates hybrid attention-based optimized ResNet for FER and its application to intelligent assistant systems. To build a more effective visual recognition system, an optimized ResNet expression recognition method combined with a Convolutional Block Attention Module (CBAM) is proposed to accurately identify seven distinct facial expressions. Furthermore, we design an improved residual block by adding a max-pooling layer to boost FER accuracy. Evaluated on public CK+ and FER2013 datasets, the proposed method achieves recognition accuracies of 98.00% and 74.29%, respectively, outperforming classical CNN baselines (AlexNet, VGG-19, vanilla ResNet) and single-attention enhanced variants (ResNet-SE, ResNet-ST). The proposed hybrid-attention-based optimized ResNet is deployed on a Raspberry-Pi-based intelligent assistant prototype to demonstrate its practical deployability for facial expression recognition. Experimental results demonstrate that the proposed Optimized-ResNet with hybrid attention and improved residual blocks outperforms the classical ResNet baseline and ResNet variants with single attention modules on CK+ and FER2013 datasets.
Source: Electronics
Language: English
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7236690265
FWCI:
Cites: 21
Cited by:
Related to: 10