Article: Student facial expression recognition and learning status analysis based on LFM-FD model Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.88 pp.1 - 29 Abstract: In online education, students' home lighting conditions are variable, and traditional methods suffer from a decline in facial feature extraction accuracy under low light conditions. This paper proposes a face detection model that integrates local binary pattern texture features with FaceNet deep embedding features. Based on multi-task cascaded convolutional networks, low light texture perception is enhanced through local binary patterns, identity consistency constraints are provided through FaceNet, and deep feature fusion is performed through a 50 layer residual network to improve the robustness of keypoint detection in dim environments. In addition, the eye aspect ratio and mouth aspect ratio are calculated, and a three-path head pose estimation algorithm is integrated to achieve multidimensional quantitative analysis of the learning state. Experiments show that the key point detection accuracy under extremely low lighting conditions reaches 95.03%, with an average error of 4.12 pixels. The recognition accuracy for fatigue and other states in the educational dataset exceeds 95%. The classroom video analysis successfully plotted the time series variation curves of different students' attention levels. The contribution of this study lies in proposing a face detection model for low-light environments and constructing a multi-dimensional feature system that integrates facial expressions and head poses, and verifying the effectiveness of this solution in the online education scenario. Inderscience Publishers - linking academia, business and industry through research

Title: Student facial expression recognition and learning status analysis based on LFM-FD model

Authors: Xirun Wang; Liang Bai

Addresses: School of Marxism, Jilin University of Architecture and Technology, Changchun, 130114, China ' Academic Affairs Office, Jilin University of Architecture and Technology, Changchun, 130114, China

Abstract: In online education, students' home lighting conditions are variable, and traditional methods suffer from a decline in facial feature extraction accuracy under low light conditions. This paper proposes a face detection model that integrates local binary pattern texture features with FaceNet deep embedding features. Based on multi-task cascaded convolutional networks, low light texture perception is enhanced through local binary patterns, identity consistency constraints are provided through FaceNet, and deep feature fusion is performed through a 50 layer residual network to improve the robustness of keypoint detection in dim environments. In addition, the eye aspect ratio and mouth aspect ratio are calculated, and a three-path head pose estimation algorithm is integrated to achieve multidimensional quantitative analysis of the learning state. Experiments show that the key point detection accuracy under extremely low lighting conditions reaches 95.03%, with an average error of 4.12 pixels. The recognition accuracy for fatigue and other states in the educational dataset exceeds 95%. The classroom video analysis successfully plotted the time series variation curves of different students' attention levels. The contribution of this study lies in proposing a face detection model for low-light environments and constructing a multi-dimensional feature system that integrates facial expressions and head poses, and verifying the effectiveness of this solution in the online education scenario.

Keywords: LFM-FD; facial expression recognition; learning state analysis; convolutional neural network; feature extraction.

DOI: 10.1504/IJICT.2026.155565

International Journal of Information and Communication Technology, 2026 Vol.27 No.88, pp.1 - 29

Received: 09 Feb 2026
Accepted: 11 May 2026
Published online: 05 Aug 2026 *

Title: Student facial expression recognition and learning status analysis based on LFM-FD model

Keep up-to-date