<p>By using a student classroom behavior detection platform to detect student behavior, it can provide schools and teachers with more comprehensive classroom performance analysis methods, so that teachers' evaluation of students no longer depends on grades, but on the overall focus of students in class. However, existing object detection algorithms are unable to fully meet the multi student detection tasks in real classrooms, such as student seat occlusion, complex backgrounds, and diverse and frequently changing student movements, all of which pose significant challenges to the algorithms. Based on this, this paper has made improvements on existing object detection algorithms to better solve the detection difficulties caused by multi student occlusion and complex environments. This paper focuses on the occlusion and multi-scale problems of existing object detection algorithms in real student classrooms. This study proposes three strategic improvements to the YOLOv11 algorithm, introducing the Metaformer architecture and convolutional gated channel attention module, optimizing the backbone network of YOLOv11, and enhancing the model's ability to extract global and local information from images. Design a downsampling module based on linear deformable convolution and YOLOv7 downsampling, replacing the original downsampling structure to preserve more detailed information and cope with the complexity and background interference of student behavior in the classroom. Replace the detection head of the original YOLOv11 model with a dynamic attention mechanism object detection head, and further improve the detection accuracy and stability of the model in dealing with occlusion problems and complex backgrounds through dynamic multi-scale feature fusion and adaptive deformable convolution mechanism.</p>