Key points are not available for this paper at this time.
Aiming at key challenges in emotional recognition for college students in online learning, including the lack of experimental data sets, the imperfect emotion classification system, and the poor robustness of emotion recognition algorithms, this paper constructs an emotion recognition model for college students' online learning based on feature fusion and attention mechanism.First, based on face recognition technology, the Convolutional Neural Network (CNN) deep features, Histogram of Oriented Gradients (HOG) texture features, and Scale-Invariant Feature Transform (SIFT) features of images are extracted and fused, because three feature extraction algorithms have different feature extraction capabilities and have certain robustness to changes in image lighting, rotation, scale, and other factors, fusion can make feature vectors more comprehensive and rich in information, thereby improving the accuracy and robustness of detection and recognition; second, a ResNet network is constructed to complete the basic classification of learning expressions and verify the experimental results; In order to enhance the ability of deep learning methods to learn discriminative features from noisy signals and further improve classification accuracy, then, by combining the channel attention mechanism and soft thresholding to improve the ResNet network, a Deep Residual Shrinkage Network (DRSN) is constructed to achieve emotion recognition of college students' online learning; finally, through a horizontal comparison experiment of multiple different network structures, the effectiveness of the soft threshold attention mechanism is verified.This method obtains more complete facial expression features for emotion classification through feature fusion, and combines the channel attention mechanism module.The recognition accuracy of DRSN network is 84.12%, which is about 4.17% higher than the original ResNet network (79.95%).The application of this research can assess the concentration of college students in online learning and measure the degree of emotional engagement of college students.Through the analysis of model result data, teachers can understand the course content that college students are interested in, adjust teaching plans according to the learning status of students, help realize personalized teaching, and consolidate teaching results.
Sun et al. (Thu,) studied this question.