A Deep Learning-Based Method for Object Workpiece Recognition and Grasp Detection

Abstract

Accurate and stable target detection is crucial for robotic grasping tasks under uneven lighting conditions. To address this, this paper proposes a target object detection network (YOLO-Net) that integrates feature fusion and attention mechanisms. First, a deep learning-based object detection model is developed to effectively mitigate the interference caused by uneven lighting, accurately extracting the features of the target objects. Next, the pose of the target object in the world coordinate system is obtained through hand-eye calibration. Finally, robot modeling and control are implemented within the ROS system to guide the robot in precisely grasping and placing the target object. Experimental results demonstrate that the proposed method effectively handles the interference caused by uneven lighting, achieving a recognition accuracy of 92.2% and an overall average grasping success rate of 93.75%, confirming the feasibility and effectiveness of the approach.