Surface electromyography (sEMG) signals reflect the local electrical activity of muscle fibers and the synergistic action of the overall muscle group, making them useful for gesture control of myoelectric manipulators. In recent years, deep learning methods have increasingly been applied to sEMG gesture recognition due to their powerful automatic feature extraction capabilities. sEMG signals contain rich local details and global patterns, but single-scale convolutional networks are limited in their ability to capture both comprehensively, which restricts model performance. This paper proposes a deep learning model based on multi-scale feature fusion—MS-CLSTM (MS Block-ResCBAM-Bi-LSTM). The MS Block extracts local details, global patterns, and inter-channel correlations in sEMG signals using convolutional kernels of different scales. The ResCBAM, which integrates CBAM and Simple-ResNet, enhances attention to key gesture information while alleviating overfitting issues common in small-sample datasets. Experimental results demonstrate that the MS-CLSTM model achieves recognition accuracies of 86.66% and 83.27% on the Ninapro DB2 and DB4 datasets, respectively, and the accuracy can reach 89% in real-time myoelectric manipulator gesture prediction experiments. The proposed model exhibits superior performance in sEMG gesture recognition tasks, offering an effective solution for applications in prosthetic hand control, robotic control, and other human–computer interaction fields.

Gesture recognition based on surface electromyography (sEMG) signals typically involves extracting features from the signals and then incorporating recognition models to increase the accuracy of classification. In this paper, drawing on the properties of sEMG signals and statistical principles, we propose a novel feature extraction method called “multiple mapping”, designed to construct a high-performance representation of sEMG signals. The multiple mapping approach incorporates sequential mappings, including the sliding average power, lg (base-10 logarithm) mapping, a linear compression principle, and a sigmoid normalization function. The sliding average power captures the intensity variations in sEMG signals across various channels, allowing differentiation between the activity levels of distinct muscle groups. lg mapping adjusts the distribution of the sEMG signals to improve their uniformity, enhancing feature stability and facilitating comparisons. The linear compression and sigmoid normalization emphasize the signals’ central characteristics while compressing the extremes at both ends. Then, the sEMG signals obtained from multiple mapping are transformed into Sem grayscale maps, which are subsequently processed using the deep neural network ResNet50 for gesture recognition. Extensive experiments on three public datasets were conducted, and the average recognition accuracy was 95.26%, 90.81%, and 96.72%, respectively, while it was 96.8% in self-collected recognition tasks. The results demonstrate that the multiple mapping method significantly improves the feature extraction performance for sEMG-based gesture recognition, offering a promising direction for applications in prosthetic gesture control and muscle–computer interaction.