The dexterity and coordinated movement of the hand play a very important role in people's daily life and interpersonal communication. For patients with hand amputation, there are many inconveniences in life. However, prosthetic limb can help them to carry out daily life in accordance with their own intention to meet the basic grasp and interpersonal gesture needs of daily life, which can greatly improve the quality of life of patients. The bionic manipulator is such a humanoid prosthetic limb for the rehabilitation of amputees and is controlled by the surface electromyography (sEMG) signals on the surface of the human body. In this paper, a wearable and portable gesture recognition system based on pattern recognition algorithm is established to control the bionic manipulator in real time. We enhanced the accuracy in each step of signal acquisition and processing. By analyzing the mechanism and characteristics of sEMG signals, we determined the gestures to be identified, the areas of muscle, the number of electrodes, and the placement positions. Our self-developed acquisition device was used to acquire 4-channel forearm sEMG signals and data preprocessing, window sliding, feature extraction, gesture classification were conducted.