In the preschool education period of young children, the cultivation of good learning habits is launched for young children, and the value is gradually increased along with the growth of young children’s age, which ultimately leads to the development of children. The DSBC dataset is manually completed after collecting and classifying the learning behaviors of young children in this paper. The dataset is expanded by using data enhancement, and a bilinear convolutional neural network-based model for learning behavior habit detection for young children is proposed. The embedded attention module is utilized to create a simplified bilinear attention network model, and the SLBAN model is employed in real preschool teaching work. After the experimental analysis, the model of cultivating young children’s learning habits proposed in this paper produced a 27.65% amount of change in young children’s self-efficacy. Meanwhile, the F-test value was 265.255, P<0.01, which showed a significant positive influence effect. In the analysis of young children’s interest level in learning, the overall interest level of student #1 in the classroom for 40 minutes remained around 0.5, which indicates a high level of interest in learning.