In recent years, there have been numerous studies on lightweight gesture recognition models, but often at the cost of sacrificing recognition accuracy; Therefore, a lightweight gesture recognition model LAW-YOLO based on the YOLOv8 model is proposed;, Firstly, a lightweight adaptive weight convolution(LAWC) is designed and a feature extraction network is introduced to accurately focus gesture regions in complex backgrounds through an adaptive weight mechanism, effectively reducing computational costs; Secondly, the bidirectional feature pyramid network(BiFPN) is used to optimize the model feature fusion network and reduce redundant channels, significantly reducing the number of model parameters and computation with minimal accuracy loss; Finally, the self-calibrated module(SC2f) incorporating self-calibrated convolution(SC-Conv) is used to replace the C2f module in front of the detection head, enhancing the performance of the model feature fusion network and compensating for the accuracy loss caused by lightweight design; The improved model was validated on the HaGRID gesture dataset, and the number of model parameters and model size were reduced by 41.20% and 41.94% respectively compared to before the improvement. The recognition accuracy reached 98.63%, significantly improving the computational efficiency and recognition accuracy of the model.