自适应卷积神经网络在面部表情识别中的应用

针对无法对面部表情进行精确识别的问题，提出了基于ResNet50网络融合双线性混合注意力机制的网络模型。针对传统池化算法造成图像特征提取残缺、模糊等问题，提出了一种基于Average-Pooling算法的自适应池化权重算法，同时基于粒子群算法对卷积神经网络模型超参数进行自适应调节，从而进一步提升模型识别精度。基于改进的网络模型，设计了一款实时面部表情识别系统。经验证，在Fer2013数据集和CK+数据集上，改进的模型在测试集中的识别精度分别为73.51%和99.86%。

Application of adaptive convolutional neural network in facial expression recognition

For the problem that facial expressions cannot be accurately recognized,a network model based on ResNet50 network integrating bilinear mixed attention mechanism is proposed.For the problem of incomplete and fuzzy image feature extraction caused by traditional pooling algorithm,an adaptive pooling weight algorithm based on average-pooling algorithm is proposed.At the same time,the hyperparameters of the convolutional neural network model are adjusted adaptively based on particle swarm optimization algorithm to further improve the model recognition accuracy.Finally,a real-time facial expression recognition system is designed based on the improved network model.It is verified that on Fer2013 data set and CK+ data set,the recognition accuracy of the improved model in the test set is 73.51% and 99.86% respectively.