Robots equipped with visual grasping capabilities are well-suited to handle variations in object shape and position. This paper introduces a real-time, object-agnostic grasping synthesis method, YOLOv10-GGCNN, which facilitates closed-loop grasping control. The approach leverages an end-to-end detection model, YOLOv10, for visual processing, and the Generative Grasping Convolutional Neural Network (GGCNN) for grasping prediction. These two models are integrated within the ROS2 system, where YOLOv10 is tasked with recognizing and selecting specific objects from RGB image data, extracting bounding boxes via image segmentation, and converting this information into depth data for GGCNN. GGCNN then computes the optimal grasping point and trajectory based on the highest quality score. This method effectively mitigates interference from non-target objects during grasping point generation, improving the accuracy of the GGCNN model and enabling precise and rapid grasping of specific objects.