Accompanied by the significant progress of deep learning in various fields, target detection, an important branch in the field of deep learning, is gradually being applied in educational scenarios. In this paper, using the GCT-YOLOv5 algorithm and the Lasso-LARs algorithm, we studied how to improve the interactive efficiency of college English classroom teaching Research first combines the YOLOv5 model with the GCT unit, constructs the GCT-YOLOv5 target detection model for students’ classroom behaviors, and compares it with Fast-RCNN, SSD, YOLOv3, YOLOv4, and YOLOv5 in the public dataset and the student classroom behavior dataset for comparison experiments, respectively. Then, the LASSO-LARS algorithm is used to construct the classroom behavior scoring model, adjust classroom teaching interactions according to the scoring results, and analyze the effect of the adjustment. On the classroom behavior dataset, the GCT-YOLOv5 model is 98.81%, second only to Fast-RCNN, and improves 0.65% relative to the original model YOLOv5. Its inference time compared to YOLOv5 inference time increased by only 0.13h, shorter than other models. The GCT-YOLOv5 model in this paper has better applicability and timeliness in general. Teaching interaction adjustment based on the results of classroom behavior analysis can improve students’ classroom performance as well as their academic performance, i.e., enhance the efficiency of teaching interaction and provide a way of thinking about how to monitor learners’ learning status in a smart classroom.