A CNN-Based Facial Emotion Recognition Model with Data Augmentation and Transfer Learning for Real-Time Edge Device Applications
Qiuying Han, Shaohui Zhang
Zhoukou Normal University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
The field of facial expression and recognition, based on CNN, is rapidly expanding within the scope of computer vision, while applications are quickly developing for human-computer interface, security, and health. This work puts together a novel framework for FER that solves issues such as the need for real-time processing on device with limited resources, the requirement of large amounts of labelled data, and changes in lighting and face orientation. The task posed a need to secure gestures to be generically performed in different real-world settings. To this effect, data augmentation and transfer learning are two mechanisms with which systems elevated such goals. In addition, various techniques, such as greyscaling, scaling the image, and normalization, have been made to effectively pre-process the input data. Evaluation of the model performance is done with accuracy (98.75%), precision (98.76%), recall (98.75%), and F1-score (98.75%), clearly indicating it as an excellent classifier of facial emotions. With Disgust (0.997261), Conempt (0.999279), and Fear (0.999593) showing near-perfect precision and recall, the model depicts a set of great AUC values. Yet, in terms of accuracy and computational effort, the proposed method is an excellent one for a real-time system on edge devices. Future research will seek to increase the generalization of this system and reduce the overfitting issues attached to it, deciding the diversity of the dataset so that it becomes heterogeneous in nature, plus implementing this system on low-resource devices.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIEmotion and Mood Recognition
Face recognition and analysis · Face and Expression Recognition