<p>Accurately recognizing students’ micro-expressions is crucial for the timely adjustment of teaching strategies and the improvement of education quality. However, in classroom teaching scenarios, the weak intensity and brief duration of students’ micro-expressions, as well as students’ head movements and uneven classroom lighting, pose difficulties in extracting micro-expression features. To solve this problem, a novel dynamic-static bidirectional interaction collaborative network (DSBICNet) is proposed in this paper. The network firstly proposes a dual attention guided feature fusion module(DAGFFM) in the dynamic branch, which characterizes the micro-expression motion features by fusing the optical flow and pixel differences between the onset frame and apex frame, and proposes a gated cross-layer feature transfer mechanism(GCFTM) to screen the key motion features and inject them into the deeper layers of the network, which effectively suppresses the noise interference and mitigates the problem of the micro-expression signal attenuation caused by continuous downsampling. Secondly, the mixed attention Transformer blocks for generating micro-expression saliency maps and the token re-allocation Transformer blocks guided by the saliency maps are innovatively designed in the static branch, which accurately extracts fine-grained static features from saliency regions such as the eyes and mouth corners of the apex frame. Finally, a bidirectional interactive attention fusion module (BIAFM) is proposed to deeply and interactively fuse dynamic and static features. Experimental results on the public micro-expression datasets and the self-constructed student micro-expression dataset in classroom scenarios show the superior performance of DSBICNet.</p>