Deep CNN fusing attention and LGBM classifier for high-accuracy hand gesture recognition

Gesture recognition has been widely used in automation and medical and smart-home. The adaptability and accuracy of traditional machine learning algorithms are limited by feature extraction and classifier selection, and the recognition accuracy is low as well as the robustness is poor. In our work, an improved CNN (Convolutional Neural Networks) combined with an LGBM (Lightweight Gradient Boosting Machine) hand gesture recognition method is proposed to improve the performance of hand gesture recognition. The method designs a 4-layer CNN structure and integrates a CBAM (Convolutional Block Attention Module) in the network to enhance the model's ability to learn and extract features. Furthermore, the extracted features are fed into an LGBM classifier for gesture classification, which improves the accuracy of gesture classification and reduces the risk of model overfitting. The accuracy of the 16 types of static gesture recognition reaches 97.89% in our dataset. An improvement of more than 10% is obtained in comparison with traditional machine learning classification algorithms. It demonstrates promising prospects for gesture recognition.