A multi-scale lightweight garbage classification target detection algorithm is proposed in view of the fact that the garbage in scenic spots is concentrated and the target scales are different, the cost of manual picking is high, the efficiency is low, and the traditional detection algorithm is difficult to effectively identify the garbage types. Firstly, C2f-MS module is applied to the backbone network to reduce the number of model parameters and enhance its ability to handle complex multi-scale scenes; Secondly, the neck network is redesigned, and a multi-scale feature fusion network, MergenNet, is proposed to enhance the extraction ability and speed of targets with different scales, and improve the extraction ability of small target features; Finally, the improved QFL Shape IOU is used as the loss function of the model to improve the positioning ability of objects in overlapping environments; The experimental results show that in the domestic waste classification detection dataset, the parameters of the improved algorithm model are reduced by 11%, FLOPs are reduced by 4%, and Map is increased by 3.3% compared with YOLOv8n. Through the scenic spot robot capture experiment, the improved algorithm has good use value in processing scenic spot garbage tasks.