Brain-computer interfaces (BCIs) are an emerging human-machine interface technology that establishes a direct connection between the human brain and machines, thereby offering significant application potential in the field of sports rehabilitation training. This paper focuses on the overall control scheme design of a sports rehabilitation training system, which primarily consists of three components: the host computer, the underlying motion control system, and detection feedback. Magnetic control damping force is incorporated to enhance muscle strength in the patient’s injured limb. In terms of rehabilitation training action matching, preprocessing of electroencephalographic (EEG) signals is performed by removing baseline drift, power frequency interference, and electrooculographic (EOG) interference. Additionally, for the traditional dynamic time warping (DTW) algorithm, a similarity measurement method tailored to the specific requirements of rehabilitation training action matching is introduced to impose temporal constraints on the matching points of two action sequences. This leads to the proposal of an improved DTW algorithm for efficient matching and recognition of rehabilitation training actions. Furthermore, based on the proposed rehabilitation training action matching model, an evaluation method and scoring formula for rehabilitation training actions are proposed. The designed rehabilitation training action matching model achieves an identification accuracy rate of 79.00% or higher for three randomly selected matching action groups.