Lower limb exoskeletons have been shown to have the potential to help patients rehabilitate and improve their walking ability. This paper propose a lightweight rigid-flexible coupled exoskeleton robot and design a gait trajectory planning algorithm for lower limb exoskeleton based on adaptive oscillator and deep neural network. The main body of the exoskeleton robot is made of lightweight materials with an overall mass of only 7 kg. At the same time, the knee motor is moved up and placed between the hip and knee joints, and the knee motion is driven by the Bowden cable, which reduces the rotational inertia of the knee joint and improves the comfort of the wearer. In the control method, the starting point of each gait cycle is determined by gait feature event judgment through deep neural network. Real-time estimation of gait phase is performed by adaptive oscillator for exoskeleton trajectory planning. Meanwhile, level ground walking experiments of normal human body and clinical verification experiments of patients were carried out, and the experimental results demonstrated the effectiveness of lower limb exoskeleton robot-assisted rehabilitation.