<p>The hand gesture recognition methods have achieved significant success in the field of computer vision. Graph convolutional networks (GCNs) are crucial techniques for hand gesture recognition tasks, especially for extracting features from graph-structured data. However, existing GCNs suffer from limitations such as excessive reliance on predefined skeleton topological graphs and a lack of flexibility in handling large temporal convolution kernels, which significantly constrain their expressive power and robustness. In this paper, an adaptive bottleneck layer multi-scale graph convolutional hand gesture recognition method based on skeleton data is proposed (BM-GCN). The adaptive spatial module optimizes the skeleton topological graph structure and parameters, enhancing the model’s flexibility. The bottleneck layer multi-scale temporal module improves the temporal modelling capabilities while reducing channel width to save computational costs and parameters. To verify the effectiveness of the proposed method, experiments were carried out on the large skeletal hand gesture recognition datasets DHG-14/28 and SHREC’17. The results demonstrate that the proposed method delivers outstanding performance on the SHREC’17 Track dataset, achieving accuracies of 98.9% and 96.9% in the 14-gesture and 28-gesture recognition tasks, respectively. Similarly, on the DHG-14/28 dataset, it achieves accuracies of 97.9% and 96.6% for the 14-gesture and 28-gesture recognition tasks, respectively.</p>