Objective A grasping detection network that balances detection speed and detection accuracy was proposed for how a robotic arm can sense the position information of a target object and complete a grasping operation in a planar grasping scenario. Methods Based on the feature that the input size and the output size are the same in the grasp detection task a grasp detection network was designed using semantic segmentation ideas. The input was a randomly cropped depth image and the output was images of the same size containing features like grasp confidence grasp angle and grasp width. To improve the efficiency of the planar grasping task the network structure was improved by replacing the backbone structure of the network using the U-net and Deeplabv3 algorithms in addition to adding an attention mechanism to the network structure taking into account the speed and accuracy of the detection network. Comparative experiments concluded that the detection network with the attention mechanism has a better balance of detection speed and detection accuracy and can achieve the grasping task transmitting the grasping poses to the ROS system and performing the grasping operation through a series of coordinate transformations and motion planning. Results After adding the attention mechanism the inference time of the detection network was still in the millisecond range and the maximum grasping confidence was improved by 7. 2%. The detection speed of the networks using U-net and Deeplabv3 was slower the grasping confidence ??Qmax of the U-net network was improved by 18. 8% and the accuracy Acc of the Deeplabv3 network was improved by 12. 8%. Since the grasp detection network should consider both detection speed and accuracy the detection network with the addition of the attention mechanism has better performance. Conclusion The experimental results show that the grasping network with the attention mechanism has a better balance of detection speed and accuracy and can successfully grasp the target objects in the scene and the method has some practical application value.