Driven by advancements in deep learning technologies, substantial progress has been achieved in the field of facial expression recognition over the past decade, while challenges remain brought about by occlusions, pose variations and subtle expression differences in unconstrained (wild) scenarios. Therefore, a novel multiscale feature extraction method is proposed in this paper, that leverages convolutional neural networks to simultaneously extract deep semantic features and shallow geometric features. Through the mechanism of channel-wise self-attention, prominent features are further extracted and compressed, preserving advantageous features for distinction and thereby reducing the impact of occlusions and pose variations on expression recognition. Meanwhile, inspired by the large cosine margin concept used in face recognition, a center cosine loss function is proposed to avoid the misclassification caused by the underlying interclass similarity and substantial intra-class feature variations in the task of expression recognition. This function is designed to enhance the classification performance of the network through making the distribution of samples within the same class more compact and that between different classes sparser. The proposed method is benchmarked against several advanced baseline models on three mainstream wild datasets and two datasets that present realistic occlusion and pose variation challenges. Accuracies of 89.63%, 61.82%, and 91.15% are achieved on RAF-DB, AffectNet and FERPlus, respectively, demonstrating the greater robustness and reliability of this method compared to the state-of-the-art alternatives in the real world.

The facial expressions of students in the class are one of the most direct and effective ways to reflect the emotional investment of learning. Driven by a new generation of artificial intelligence technology, facial expression recognition in classroom and learning emotion analysis based on video image comprehension has become a hot topic in the field of Wisdom Education. In the wisdom classroom environment, the emotional state of the students in classroom can be effectively recognized and perceived through facial expression recognition technology, so as to capture the emotional input of the students in the classroom in real-time. This provides an effective means to achieve differentiating instruction, personalized learning, and accurate teaching decision-making. Based on a large number of frontier research and recent achievements on domestic and overseas, this paper makes a comprehensive review of the technology of facial recognition and the study of emotional analysis in classroom. Firstly, the classification of students' expressions and the construction of the database are summarized and analyzed. Secondly, this paper summarizes the research status, key points, and development trend of class facial recognition technology based on deep learning in ideal and non-ideal scenarios. Thirdly, based on the classical emotion analysis model, the typical application of emotion analysis based on students' facial recognition is discussed in detail. Finally, this paper summarizes the current challenges and future research directions of emotion recognition technology and learning emotion analysis in classroom. It provides references and guidance for the research in this field.

Nowadays, with the development of deep learning technology, intelligent analysis of classroom teaching action has become a focus in the field of smart education. Analyzing teacher action might reveal their teaching philosophy, which is important for carrying out intelligent teaching evaluations. The frameworks used in the current studies for teacher action recognition based on deep learning are still largely conventional, and their accuracy is not very good, especially for small-scale datasets. In this paper we propose a fusion framework that combines pre-trained language and video models with prompts, to transfer these models to the teacher action recognition task. Through experiments, our model outperforms the conventional methods on the teacher action recognition task, and achieve higher performance than baseline models on public small-scale action recognition datasets.