面向深度学习的学习表情数据库的设计与实现

The Design and Implementation of Learning Expression Database Oriented for Deep Learning

学习表情数据库中学习表情图像的数量与质量将直接决定深度学习模型的效率和泛化能力,但既有表情数据库存在样本数量少、分辨率低、表情类型有限等问题,而且其建设初衷也并非出于教育教学研究的需要,这无疑限制了学习表情的识别与应用。文章从现实需求出发,建成了拥有168000幅图像的学习表情数据库,并采用主观评价法对该数据库进行了评价,以确保该数据库的实用性和有效性。评价结果显示,学习表情数据库具有较高质量,能够满足实际需要。学习表情数据库的建设,对于提高学习表情识别的效率、准确率和鲁棒性具有重要意义,并有助于实现学习者与学习环境之间的深层情感交互。

The quantity and quality of learning expression images in the learning expression database can directly determine the efficiency and generalization ability of the deep learning model.However,the existing learning expression databases had problems of small sample size,low resolution and limited expression types,and their original intention of construction were not out of the needs of education,which undoubtedly limited the recognition and application of learning expressions.Starting from the actual needs,this paper constructed a learning expression database which contained 168,000 images.In order to ensure the reliability and effectiveness of the constructed database,this paper evaluated it using a subjective evaluation method.The results showed that the learning expression database was of high quality and could meet the actual needs.This paper was of great significance for improving the efficiency,accuracy and robustness of learning expression recognition,and helped achieve harmonious emotional interaction between learning environment and learners.