To address the time-consuming and labor-intensive nature of traditional manual physiotherapy and rehabilitation, this study proposes a novel five-degree-of-freedom (5-DOF) rehabilitation robot system integrated with a deep learning-based dynamic treatment point recognition algorithm. First, a 5-DOF rehabilitation robot hardware architecture was designed and implemented, incorporating a multifunctional intelligent physiotherapy probe capable of delivering both thermal stimulation and vibration therapy. Second, a deep learning framework was developed for accurate localization and dynamic tracking of back treatment points (BTP). This framework utilizes a human pose estimation model to extract key anatomical landmarks and constructs a joint distribution vector and coordinate system to enable real-time identification of treatment points. Finally, based on established rehabilitation methodologies, vertical compliance control strategies, targeted heating, and rhythmic vibration functions were incorporated to deliver precise therapeutic interventions. Experimental results demonstrate that the proposed robotic system achieves accurate identification and dynamic tracking of target treatment points, while maintaining advantages such as low cost and lightweight design, making it suitable for use in various rehabilitation settings including hospitals and home environments.