Abstract: In this paper, an intelligent grasping system combining deep learning and ROS is proposed. The system is based on the YOLOv5 algorithm model, and MobileNetV3 is introduced as the network backbone, which effectively improves the lightweight degree of the model. The Conv of the neck network was replaced by the group shuffle convolution(GSConv) to reduce the complexity of the model and enhance the effect of the attention mechanism. The C3 module in the neck network is replaced with VoVGSCSP to further reduce network complexity. The parallel convolutional block attention module(PCBAM) is added before the last detection head to enhance the model′s sensitivity to the target object. Compared with the YOLOv5s baseline model, the recall rate of the improved model is increased by 3.1%, the number of parameters is reduced by 49.6%, and the amount of computation is reduced by 65.8%. Experimental results show that the system has high real-time detection performance, and can accurately identify and grasp objects in chaotic scenes.