<![CDATA[Facial expression classification in digital images presents complex challenges due to variations in lighting, facial pose, and image quality, necessitating robust feature extraction methods and computationally efficient classifiers. The novelty and primary contribution of this research lie in the experimental exploration of integrating a lightweight CNN architecture—MobileNetV2—as a feature extractor (via transfer learning) with the Broad Learning System (BLS), a classification method based on horizontal network expansion that operates without backpropagation. This quantitative experimental study utilizes the Human Face Emotions Dataset, comprising 4,387 digital images categorized into five expression classes: anger, fear, happiness, sadness, and surprise. The research process includes preprocessing (resizing images to 224 x 224 pixels and normalizing pixel intensity), dataset partitioning (80% for training—3,500 images—and 20% for testing—887 images), feature extraction using MobileNetV2, feature normalization via StandardScaler, classification using the BLS (configured with 1,500 feature nodes and 800 enhancement nodes), and performance evaluation. Model performance was assessed using key metrics, including accuracy, precision, recall, F1-score, and a confusion matrix. The results demonstrate that the combination of MobileNetV2 and BLS achieves an accuracy of 93.21% on the test set, with average precision, recall, and F1-score values of 0.93 each. These findings confirm that the integration of MobileNetV2 and BLS is effective in distinguishing facial expression patterns within the tested dataset]]>
Copyrights © 2026
