In the field of abnormal behavior recognition in smart classroom, the proportion of abnormal behavior samples is low, the feature extraction of abnormal behavior is not accurate, and the temporal and spatial characteristics of behavior are not considered. To solve these problems, this paper proposes an enhanced small sample abnormal behavior recognition method in the core area for smart classroom. Firstly, a recognition framework consisting of a spatially adaptive core region selection module and a spatio-temporal correlation unit of long and short-term feature maps was designed. Then, considering the temporal and spatial refinement matching mechanism, a spatio-temporal refinement few-shot loss function was designed in the framework, and three training techniques were added to the training: gradient pre-interruption, enhancing local feature random data, and optimizing the loss function. Finally, the effectiveness of the proposed recognition algorithm is verified by selecting the public behavior data set HMDB51 and the public course video data set from the National Education Resource Service platform. Experiments show that the algorithm proposed in this paper performs superior in the problem of abnormal behavior recognition in smart classroom and has certain reference value.