Research on Brain-Controlled Robotic Arm Based on Improved Faster-RCNN Target Detection Model

Abstract

Aiming at the problem of poor target detection in brain-controlled system, which leads to limited applications, this paper proposes a brain-controlled robotic arm system based on an improved Faster-RCNN target detection model. Firstly, the dataset is obtained through data enhancement for improve the generalization ability of the model, meanwhile, the Faster-RCNN model is improved, which increases the recognition accuracy by 4.38% compared with the previous improvement. Then, the improved model is used to detect the type and position of targets in real time, and the Filter Bank Canonical Correlation Analysis (FBCCA) algorithm is used to extract the features of the Steady-State Visual Evoked Potential (SSVEP) paradigm Electroencephalogram (EEG) signal to control the robot arm to grasp the desired object. Finally, 12 subjects were recruited to test the system. The test results showed that the designed brain-controlled robotic arm grasping system based on the improved Faster-RCNN target detection had a grasping average accuracy of 92.33%, which meet the practical application requirements of medical auxiliary equipment, medical rehabilitation and other fields.