<p>In the task of sorting waste using robots, it is necessary to determine the location and category information of waste. The key to the efficient work of the waste sorting robot lies in the accuracy of the target recognition. Traditional waste detection algorithms suffer from high computational cost, low detection accuracy, and poor adaptability, which cannot meet the actual detection needs. Therefore, this paper proposes an efficient and lightweight algorithm YOLO-VG based on the current mainstream algorithm YOLOv8s for waste detection and classification. The algorithm replaces the conventional convolution in the backbone network with GSConv to reduce redundant information and improve the model’s inference process. ODConv replaces the regular convolution in the neck network to enhance the model’s adaptability and generalization ability, thereby reducing the risk of overfitting. Additionally, the VoV-Ghost structure is introduced in the neck network to replace the original C2f, making the model more lightweight and efficient, meeting the requirements of real-time object detection in embedded devices or resource-constrained environments. Finally, the ECA attention mechanism is introduced to improve the ability of information interaction between feature channels in the model, thereby better capturing important information between images or features. Experimental results on the recyclable waste dataset demonstrate that the proposed YOLO-VG achieves a 24.6% improvement in computational efficiency, a 20.4% reduction in model size, and an mAP0.5 of 88.4%, surpassing the performance of the original YOLOv8s. These results indicate that YOLO-VG not only demonstrates excellent detection performance and stability, but also exhibits significant potential for widespread application in the field of waste sorting.</p>