Lightweight Convolutional Networks for Uterine Contraction Classification Based on Continuous Wavelet Transform

Accurate classification of uterine contraction (UC) signals is critical for real-time obstetric monitoring, yet remains challenging due to their non-stationary nature and susceptibility to noise. This study proposes a time-frequency domain approach that transforms raw UC signals into two-dimensional representations via continuous wavelet transform (CWT), thereby capturing both temporal and spectral dynamics. These representations are processed by a compact convolutional neural network (CNN) designed for efficient feature extraction and low computational overhead. Experimental evaluations on a curated UC dataset demonstrate that the proposed method substantially outperforms conventional one-dimensional baselines in classification accuracy, sensitivity, specificity, and AUC. The findings underscore the advantage of integrating time-frequency analysis with lightweight deep architectures for robust UC signal interpretation in resource-constrained clinical settings.