Abstract: Accurate recognition of an amputee's lower limb motion intention is the key to improving the human-computer interaction performance of lower limb prosthesis and reducing the movement energy consumption of prosthetic users. In this study, a motion intention recognition method for lower limb prosthesis is designed based on surface electromyography (sEMG) signals from the lower limb surface of healthy subjects in different motion modes and kinematics signals such as angular velocity and acceleration signals collected by 6-DOF inertial measurement unit (IMU). The sEMG of the healthy side and the kinematic data of both lower limbs of hip amputees are used to validate the method's feasibility and validity. The results show that the proposed method can select the most appropriate minimum feature subset for the classification from the multi-source sensor data of healthy subjects and achieves 99.2% recognition accuracy in the fine K-nearest neighbor (KNN) classifier for six different motion intentions: standing, walking on flat ground, up and down stairs, and up and down slopes. Simultaneously, the minimum feature subset can still be selected from the hip amputee's multi-source sensor data, and the recognition accuracy can reach 99.8%, which shows the effectiveness and universal applicability of the proposed method.