To address the issues of a large number of parameters and false or missed detections by garbage-detection models in complex environments， this paper proposed a lightweight garbage-detection model based on an improved YOLOv8n. First， a lightweight network structure， MobileNet V3_ECA， was introduced as the backbone of YOLOv8n， which enhanced the model’s ability to represent garbage-feature regions and reduced the model’s parameter count. Second， the Context Anchor Attention （CAA） mechanism was integrated into the backbone to strengthen the extraction of garbage-related features. Next， Omni-Dimensional Dynamic Convolution （ODConv） replaced the standard convolutions in the neck network， refining local feature mapping and enabling the fusion of fine-grained local garbage features. Finally， the Wise Intersection over Union （WIoU v3） bounding-box loss function was adopted to improve the regression performance of the network’s bounding boxes. Compared with the original YOLOv8n， the improved model is improved by 1.1% in mAP@0.5， the detection speed is increased by 11.7%， and the parameter Params， model size and floating-point operation FLOPs are reduced by 70.8%， 66.1% and 70.7%， respectively. Experimental results demonstrate that the improved model can effectively improve the detection accuracy and significantly reduce the complexity of the model， which has important engineering significance for the deployment and application of the model to the edge detection equipment.