Abstract: Surface electromyography (sEMG) has been currently widely used in rehabilitation engineering related to the design and application of prosthetic limbs.Aiming at the problems of the current sEMG prosthetic hand controlled by sEMG signals, such as poor real-time performance, high cost, and insufficient portability, a set of split low-latency prosthetic hand system based on sEMG signals was designed.First of all, the whole system was divided into two modules, with K210 as the central master for pattern recognition and a mechanical prosthetic hand controlled by STM32F103, using 2.4 G wireless transmission as the communication connection between the two parts, and using dual-channel EMG acquisition electrodes to collect data from the anterior segment of the elbow.The muscle signals on both sides were transmitted to the main control chip equipped with K210, and then the sliding window was used to judge the initial gesture action, and the high-performance design of the data window sliding was carried out so that the whole system could realize low-latency signal processing, and then The support vector machine pattern recognition algorithm was used to identify the action type of the electromyographic signal, and finally the action type corresponding to the identification result was sent to the mechanical prosthetic hand control terminal through wireless transmission.The experimental results showed that the real-time wearable mechanical prosthetic hand system with a split design had a comprehensive recognition rate of 95.28% for gestures based on SVM, and the real-time response time to actions was less than 110 ms.