基于SCAM神经网络的图像情绪识别研究

Image Emotion Recognition Research Based On Separable Convolution Attention Mechanism(SCAM)Neural Network

图像情绪识别是情绪分析领域中的一个前沿研究方向,传统图像情绪识别方法通过导入数据库文件抽取特征,对抽取到的特征直接进行分析处理;而本文从分层优化特征角度出发,结合X-ception网络模型和注意力机制,提出一种基于可分离卷积注意力神经网络的图像情绪识别方法(SCAM)。该方法对不同卷积层特征进行筛选,学习X-ception网络模型建模后的情绪识别机制,引入用于筛选特征的注意力机制模型,进而构建图像情绪识别模型。该方法在CK+数据库上达到91.52%的情绪识别效果,充分证明了该方法在图像情绪识别任务中的有效性。

Image emotion recognition is a frontier research direction in the field of emotion analysis,Traditional image emotion recognition methods extract features by importing database files,and directly analyze and process the extracted features.From the perspective of hierarchical optimization features,this paper combines the X-ception network model and attention mechanism to propose a Image emotion recognition method based on separable convolution attention mechanism neural network(SCAM).This method screens the characteristics of different convolutional layers,learns the emotion recognition mechanism after modeling the X-ception network model,introduces the attention mechanism model for screening the features,and then constructs the image emotion recognition model.The proposed method achieves a 91.52%emotion recognition effect on the CK+database,fully demonstrating the effectiveness of the proposed method in the image emotion recognition task.