Efficient Channel and Spatial Attention Integrated Encoder Decoder Model with Deep Capsule Network for Facial Emotion Detection

A crucial component of human-computer interaction, facial expression recognition finds use in social robotics, healthcare, and security. The durability of traditional deep learning models in real-world situations is limited by their frequent difficulties with spatial hierarchies and posture fluctuations. This study presents an Efficient Channel and Spatial Attention Integrated Encoder-Decoder Model with a Deep Capsule Network for improved face emotion detection in order to overcome these difficulties. The suggested model improves feature identification across a range of facial expressions by dynamically emphasizing pertinent characteristics through the use of channel and spatial attention processes. Furthermore, the Deep Capsule Network (CapsNet) is integrated to efficiently manage affine transformations and maintain spatial links. Capsule integration improves hierarchical representation learning, while the encoder-decoder framework guarantees reliable feature extraction. In comparison to traditional CNN-based models, extensive trials on benchmark datasets such as CK+, AffectNet, and BU-3DFE show greater performance in accuracy and robustness. Because of its enhanced generalization, the suggested architecture shows promise as a real-time facial expression analysis solution for edge computing settings.