In recent years, with the popularization of intelligent devices, there have been more and more low-headed individuals in the classroom, seriously affecting the quality of classroom learning. To address this issue, this article first uses high-definition cameras to capture the real-time situation of the classroom, and uses the obtained image information as input samples to input into the neural network algorithm; Secondly, a combination of convolutional neural network (CNN) and recurrent neural network (RNN) was used to improve the accuracy and robustness of head rate detection. Subsequently, this article used a large number of classroom datasets for training and obtained an efficient and reliable head rate detection model through feature extraction, classification, and regression operations on the sample data. Finally, this article provides a detailed analysis and validation of the experimental results. Through testing multiple different datasets, we demonstrate the excellent performance of the proposed neural network model in detecting classroom head rates. Compared with traditional methods, this model has higher accuracy and robustness, and can effectively adapt to changes in the head posture of different students and environments.