Accurate swimming motion recognition is crucial to underwater exoskeleton control to provide appropriate assistances to divers’ motions. Because it is muscles that actuate joint motions, muscle deformation sensing is a promising alternative to traditional joint kinematics sensing in swimming monitoring. However, most previous studies directly fed the muscle deformation into “black box” machine learning models leading to opaque mechanism. Aiming to intuitively reveal the intrinsic relationship between muscle deformation and swimming motion postures, this paper characterizes the standard muscle synergy curves of four swimming modes based on the previously formulated Gaussian-based physical model, and then each muscle feature sample can be mapped into the constructed synergy curve based on their Euclidean distance, thereby enabling multimodal swimming motion recognition. Results show that average mode classification accuracy is 93.9% and phase estimation error is 6.56%. This paper validates muscle deformation sensing for swimming motion recognition with an intuitive and transparent mechanism, which is expected that can enhance swimming motion monitoring for underwater wearable robotic control.