基于改进YOLOv5s的人脸疼痛表情识别

Pain Expression Recognition Based on Improved YOLOv5s

针对目前疼痛表情识别模型结构复杂、计算量大、检测速度慢、不易移植等问题,提出一种针对移动端设备的轻量化人脸疼痛表情识别算法。首先引入GhostNet网络结构中的Ghost模块卷积,压缩模型的参数量,减小计算开销;之后用改进的FReLu激活函数替换SiLu激活函数,提升识别精度与检测效率;最后引入CA注意力机制,对人脸疼痛表情特征区域增加关注度,提升算法对疼痛表情模型的识别精度。实验结果表明,改进后的模型对疼痛表情识别精度达到96.9%;每张图片检测时间为53 ms,相比YOLOv5s模型用时缩短18%;模型大小相比YOLOv5s下降41.3%。适用于移动端设备的实时疼痛表情识别。

This paper proposes a lightweight facial pain expression recognition algorithm for mobile terminal devices.Firstly,the convolution of Ghost modules in GhostNet network structure is introduced to compress the number of parameters in the model and reduce the calculation cost.Then the SiLu activation function is replaced by the improved FReLu activation function to improve the identification accuracy and detection efficiency.Finally,CA attention mechanism is introduced into the output end of the backbone network to increase attention to the facial pain expression feature area and improve the recognition accuracy of pain expression model.The experimental results show that the accuracy of the improved model can reach 96.9%.The detection time of each image is 53 ms,which is 18%shorter than that of YOLOv5s model.