摘要
目的:搭建了一种基于EfficientNet的癌痛面部表情识别与强度分类的智能评估模型并评价模型准确性。方法:采集72位志愿者在6种不同疼痛强度状态下的面部表情图片,设计深度卷积神经网络并进行训练,使用特征级差异进行面部疼痛表情分类。在原网络基础上进行了以下两点优化:1) 将激活函数swish改为mish。2) 用inception v4代替网络中的mobile net模块。结果:优化后的网络与原网络相比,模型训练损失衰减更快,模型对于测试集的分类准确率有了显著的提升。结论:该方法在疼痛表情识别中准确有效,评分结果具有显著性。
Objective: To establish an intelligent evaluation model for cancer pain facial expression recognition and intensity classification based on EfficientNet and evaluate the accuracy of the model. Methods: The facial expressions of 72 volunteers in 6 different pain intensity states were collected, deep con-volutional neural network was designed and trained, and facial pain expressions were classified using feature level difference. On the basis of the original network, the following two optimizations were performed: 1) The activation function swish was changed to mish. 2) Replace the mobile net module in the network with inception v4. Results: Compared with the original network, the training loss of the optimized network decreases faster, and the classification accuracy of the model for the test set is significantly improved. Conclusion: This method is accurate and effective in pain expres-sion recognition, and the score is significant.
出处
《临床医学进展》
2023年第7期11934-11942,共9页
Advances in Clinical Medicine
