Rapid urbanization and improved living standards have led to increasing complexity and volume of household waste, making efficient sorting a critical challenge for urban management and environmental sustainability. In this paper, we propose an intelligent waste classification system that integrates visual recognition with mechanical sorting. The design incorporates a dual-conveyor zigzag mechanism for primary and secondary waste separation, a Raspberry Pi 4B deployed to perform object detection via the YOLOv5s model, and a dual-servo gimbal that directs waste into designated bins based on predictions from a PyTorch-trained model converted to TFLite. Utilizing the lightweight YOLOv5s model, the system achieves an \(\textrm{mAP}_{50}\) of 92.4% along with high category-specific average precision (AP): 99.2% for kitchen waste, 92.9% for hazardous waste, 91.6% for recyclable waste, and 85.9% for other waste. Experimental results confirm the system’s robust and real-time performance, highlighting its potential to improve automated urban waste management and promote sustainable environmental practices.