Research on Learning Behavior Detection Based on Deep Learning

In recent years, with the rapid development of deep learning technology, its application in various fields is more and more extensive. Within the realm of education, the detection of learning behavior based on deep learning has emerged as a prominent research area. Learning behavior detection aims to understand students’ learning status, habits and difficulties by analyzing the data generated by students in the learning process, such as learning logs, online answers, reading materials, etc., so as to provide support for educational decision-making. This research focuses on the detection of learning behavior based on deep learning, aiming to improve the accuracy and efficiency of learning analysis. In the past, traditional methods of learning behavior detection often relied on manual extraction of features and design rules, however, these methods often struggled to capture complex learning patterns and individual differences. Through the construction of multi-layer neural networks, deep learning has the capability to automatically acquire feature representations from extensive learning data. This enables a better understanding and capture of the underlying patterns and principles of learning behavior. In this study, aiming at the problems of low detection rate and object occlusion in target detection, we proposed an improved CaR-CNN human target detection algorithm. In this algorithm, a switchable void convolution module was introduced into the backbone network, which convolved features with different void convolution rates, and a block feature erasable module was introduced into the network structure. Batch feature erasing enhances local feature learning by randomly erasing the same region, and then strengthens the network’s learning expression of residual features. In addition, aiming at the problem of target occlusion, a Re-Repulsion Loss function was proposed. The Re-Repulsion Loss function mainly consists of attraction loss and Repulsion Loss. The original attraction loss and DIoU Loss were weighted and fused. It can make the target detection result more accurate. However, deep learning models have powerful presentation capabilities to mine the data for distinguishing features between individuals. In conclusion, learning behavior detection based on deep learning represents a cutting-edge area of research within the field of education. By leveraging the inherent strengths of deep learning, we can effectively analyze students’ learning behavior with greater accuracy, thereby providing robust support for educational decision-making. Future research can further explore more complex deep learning models, integrate multi-source learning data for comprehensive analysis, and expand the application of this technology to personalized education and other fields, so as to promote the intelligent development of education.