LIGHTWEIGHT FACIAL EMOTION RECOGNITION SYSTEM USING DEEP LEARNING MODELS

Facial Emotion Recognition (FER) is vital for emotionally aware systems [14]. Previous methods, such as EWDL-BFSN, used GWAF preprocessing, IBoA, SqueezeNet, and EK-ResNet50, but their complexity limits real-time deployment also heavy methods. To address this, we propose a novel lightweight FER pipeline that replaces GWAF with face alignment, CLAHE, guided filtering, and gamma correction for faster and cleaner preprocessing. Feature extraction is improved using an attention-based SqueezeNet. A compact Mini-ResNet classifier, adjusted through metaheuristics algorithms, further reduces latency and memory usage. Cross- dataset evaluation, optimization analysis, and real-time webcam testing demonstrate run-time performance and practicality compared to existing methods.