A Convolutional Neural Network-based Automatic Identification and Intervention Model for Health Surveillance Data during Postpartum Recovery Periods

Abstract Women suffer great psychological pressure on the postpartum recovery period, which can cause certain psychological diseases in the long run if not paid attention to. Based on the research related to the principle of health parameter detection and the feature extraction method of pulse wave data, the study was conducted by extracting the physiological signal features of normal pulse, using the improved support vector machine (OC-SVM) for abnormality detection, and adding the attention-based two-stage long and short-term memory network (DA-LSTM) to the AE, which adaptively directs the weights of the input sequences in the encoding/decoding stages, respectively allocation and selecting the hidden state of the encoder in the time step, respectively. Then, based on the experimental data, the development of the health monitoring system was carried out from three major modules, namely, the main control module, the front-end acquisition and processing module, and the auxiliary module, to realize the intervention for postpartum recovery. Using this paper to carry out a three-month intervention experiment for postpartum women, it is found that the experimental group after the experiment of each index value has decreased and the rate of decrease is large, the experimental group somatization from (1.26 ± 0.13) to (1.09 ± 0.58), the value of the decrease is large, compared with the experiment before the significant difference (P < 0.05), to help women recover their health level more quickly after childbirth.