Pulse Signal Recognition Based on EfficientNetV2_ResBlock Network

In the process of pulse signal recognition analysis, methods in the time domain, frequency domain, and time-frequency domain are difficult to deeply extract the nonlinear features of pulse signals. Moreover, traditional machine learning requires manual feature definition and cannot perform self-learning of features. Therefore, this paper proposes a pulse signal recognition analysis method based on the EfficientNetV2_ResBlock network. Compared with previous methods that use features for pulse recognition, this paper uses two-dimensional images for pulse recognition, fully retaining the feature information of the pulse signals and improving recognition accuracy. Using GAF (Gramian Angular Field), the one-dimensional pulse time series is converted into two-dimensional images. The Efficient-NetV2_ResBlock network model is used to analyze and recognize pulse signals and to establish a pulse signal classification model. Experimental results show that the proposed model achieves an average recognition accuracy of 97.9% for pulse recognition, with the recognition accuracy for abrupt pulse and floating pulse reaching 100%. Compared with other convolutional neural networks, the proposed model has higher accuracy, shorter training time, and fewer network parameters.