In the field of object detection, the YOLO series of object detection networks are widely used in mobile grabbing robots such as agricultural picking, logistics and warehousing, road garbage cleaning and collection due to their powerful performance, which improves the production efficiency of work and reduces labor costs. The object detection network used in mobile grabbing has more real-time and lighter requirements. In order to reduce the complexity of the model and save computing resources by reducing the complexity of the network model, so as to facilitate its use on limited resources and mobile devices, this paper proposes an improved lightweight network detection model based on the YOLOV5 framework, and we optimize the network in multiple dimensions, including introducing the CAFM attention mechanism to enhance the feature representation ability, and adjusting the network structure to balance the computing resource consumption and detection accuracy. Experiments show that the improved network has superior performance compared with the traditional lightweight model.