Multi-object detection Intelligent Sorting Model of Rural Construction Waste Based on Improved YOLOv8

The rapid and efficient sorting of construction waste in rural areas is crucial for promoting its resource utilization. Traditional manual sorting methods are inefficient and costly, which severely restricts the automation of this process. Current automated systems also face challenges such as low recognition accuracy, poor adaptability, and high computational demands in complex rural environments. To address these gaps, this study aims to develop a lightweight, robust detection model based on an improved YOLOv8 architecture for multi-object detection intelligent sorting of rural construction waste. Methodology: We introduce the FasterNet backbone network to reduce computational complexity and embed the Spatial Pyramid Pooling Fast with Cross Stage Partial Connections (SPPFCSPC) module to enhance multi-scale feature fusion. Additionally, the Attention Mechanism (AM) is integrated to improve the perception of occluded and small targets. Results: Experimental results show that the proposed model achieves a sorting accuracy of 96.7%, a missed sorting rate of 6.2%, and an average sorting speed of 510 kg/hour. It also demonstrates a lower false detection rate under various interference conditions compared to baseline models. Novelty: The main contributions include the integration of FasterNet for lightweight design, the use of SPPFCSPC for enhanced multi-scale feature extraction, and the incorporation of a AM for improved small and occluded target detection. The model maintained stable performance and demonstrated generalization capability in real-world deployments, providing an efficient and reliable solution for intelligent construction waste sorting in rural areas. The model’s ability to physically sort different compositions of construction waste has been validated through field tests, where it consistently separated bricks, concrete, wood, plastic, metal, and mixed debris with an overall accuracy above 94% and a missed sorting rate below 7%. The key novelties of this work are threefold: (1) the integration of FasterNet as a lightweight backbone specifically for rural edge computing scenarios; (2) the incorporation of SPPFCSPC for enhanced multi-scale feature fusion under complex debris conditions; and (3) the introduction of coordinate attention and part-whole spatial constraints from YOLO-Parts to improve occluded and small-target detection, which has not been previously applied to rural construction waste sorting.