In this paper, for the common behaviors of students in classroom teaching, deep learning and the improved student behavior detection algorithm based on YOLOv5s are used to transform the source data into another high-level and high-abstract data expression to realize the detection of students’ behavior in class. Then, taking into account the real conditions of college students and the internal logic of deep learning and classroom teaching, we sort out the target conditions of college students’ deep learning and their brand-new requirements for college classroom teaching and innovate the art teaching mode in colleges and universities. The innovative art teaching mode is applied in the art classroom teaching of W College of Fine Arts, and the teaching effect is analyzed. The results show that the mean values of students’ evaluation in the five core literacy skills of image literacy, art expression, aesthetic judgment, creative practice, and cultural understanding increased by 12.6, 14.95, 17.66, 12.27, and 19.1 points after the innovation compared with those before the innovation. The above data show that the innovation of art teaching mode can enhance students’ motivation to learn and stimulate their creative passion, thus cultivating their creative ability, making students learn to inherit and carry forward the traditional culture of the Chinese nation, and enhancing their cultural self-confidence and national pride.