Synaptic Plasticity-Inspired Neuromorphic Deep Learning for Real-Time Recognition of Micro-Expressions in Neurodivergent Individuals

Identification of micro-expressions is crucial for interpreting small signals displaying emotions, particularly in neurodiverse individuals who may struggle with ordinary social signs. This paper comes with a brand-new neuromorphic deep learning model that can recognize microexpressions in real-time and has synaptic plasticity. This model was not previously available for this particular population. Given a set of micro-expressions labeled images, we implemented a Conventional Neural Networks-Long Short Term Memory (CNN-LSTM) model, which integrates CNN and LSTM as they capture not only spatial patterns in the facial expression but also temporal sequence. Our outcomes show that our proposed approach has an accuracy of over 90% in predicting micro-expressions, which is much higher than conventional Machine Learning methodologies. Furthermore, we confirm its accuracy across various lighting conditions and subjects, demonstrating excellent generalization for practical use. We find that our model not only enhances recognition abilities but also aids in addressing affective and emotional understanding in a neurodivergent population. In conclusion, this study presents a new way to use neuromorphic deep learning to recognize microexpressions in real-time. This could be useful in psychological research, therapy, and improving social skills.