Measurement of abnormal heartbeat rhythm of a fetus to detect arrhythmia using fetal-electrocardiogram (f-ECG) signals is one of the most convenient methods, used to quickly assess the health status of the fetus. However, instruments used to record the clean f-ECG under free-living conditions have several operational difficulties. Therefore, this study presents a new integrated Stockwell transform (S-transform)-based learning system to evacuate the clean f-ECG signals from the mother’s abdomen and automatically detect the arrhythmic fetuses. In this work, an empirical mode decomposition (EMD) algorithm is employed to decompose the raw signals into various intrinsic mode functions (IMFs) and a correlation criterion is set to eliminate noisy IMFs. The remaining noise-free IMFs are utilized to reconstruct the denoised f-ECG signals. The S-transform is used to convert the clean f-ECG signals into time–frequency (T–F) images. The obtained images are given to two prebuilt Alex-Net and Squeeze-Net along with a newly developed lightweight deep convolutional model (DCM). The presented system has yielded an average detection accuracy of 95.31%, an area under the curve of 99.57, and a specificity of 96.29%. The comparison results depict that the developed system provides higher detection performance compared to the state-of-the-art techniques. The proposed learning system efficiently detected the arrhythmic fetus, which may enhance medical measurement performance by reducing operational cost/difficulties of various f-ECG recording instruments and can assist obstetricians in deciding on delivery. Our presented system is suitable to design abdominal wearables for continuous monitoring of the fetus.