在教学过程中，老师可以通过学生的课堂行为对教学活动的展开与教学策略进行调整。由于老师的精力有限，传统办法通过人工注意每个学生的学习状态既耗费精力又不能保证面面俱到。通过更加智能化的手段，实现学生课堂行为识别，成为了一个亟需解决的问题。基于此，本文采用深度学习的方法对学生的课堂行为进行识别。整个研究包括行为识别算法的比较、数据预处理、学生课堂行为识别、探究不同深度学习网络结构、学习效率函数，以及各种参数选取对检测结果的影响。论文主要工作如下：1）使用UCF-101标准数据集，对经典的以及近年来表现比较好的四种行为识别方法进行复现和对比。选择出效果比较好的模型进行下一步学生行为识别的实验。2）在选出较好的方法后，选择ResNet50与ResNet101这两个不同的深度学习网络对学生行为进行学习，比较出更好的网络模型。3）使用挑选出来的深度学习网络，尝试不同的学习计划方法与参数，对学生行为数据集进行训练和验证，并留下每次的训练模型。

In the process of teaching, teachers can adjust teaching activities and teaching strategies through students' behavior. Because the teacher's energy is limited, the traditional way to pay attention to each student's learning state through the manpower not only consumes energy but also can't guarantee everyone. It has become an urgent problem to recognize students' classroom behavior by more intelligent means. Based on this, this paper uses the method of deep learning to identify students' classroom behavior. The whole research includes the comparison of behavior recognition algorithms, data preprocessing, students' classroom behavior recognition, and explores the influence of different depth learning network structure, learning efficiency function, and various parameters selection on the detection results. The main work of this paper is as follows: 1) Use ucf-101 standard data set to reproduce and compare four classic and good performance behavior recognition methods in recent years. Select a better model to carry out the next step of student behavior recognition experiment.2) After choosing a better method, choose resnet50 and resnet101, which are two different deep learning networks, to train the students' behavior and get a better network model.3) Using the selected deep learning network, different learning plan methods and parameters are adjusted to train and verify the student behavior data set. And leave each training model.