Against the backdrop that the annual output of shiitake mushrooms in China has exceeded the 10 million-ton scale,the continuous surge in production has made quality sorting a key bottleneck restricting the cross-scale development of the industry.The traditional manual sorting mode has obvious drawbacks,including the industry pain points of low efficiency,strong subjectivity,and poor process traceability.To address the demand gap of modern smart agriculture for the production of high-quality and high-value-added agricultural products,while resolving consumers' concerns about food quality,this study aims to propose an intelligent quality evaluation scheme for shiitake mushrooms that integrates accurate sorting and full-chain traceability.An integrated information collection system of "RFID + multimodal perception + edge computing + YOLOv8" is constructed:Each batch of shiitake mushrooms is assigned a unique RFID electronic tag to bind data throughout the entire process of cultivation,harvesting,and circulation,realizing a data closed loop from the source to the supermarket shelves.Multimodal data are collected using a USB industrial camera (resolution: 4K×3K),temperature and humidity sensors,weight sensors,and odor sensors,forming a multi-dimensional data matrix of "images + sensor data + RFID".Images are labeled for front and back sides according to refined industry standards,and the dataset is augmented through affine transformation and contrast transformation.Transfer learning is conducted based on the pre-trained YOLOv8 model,with the introduction of an attention mechanism for model optimization.Meanwhile,non-image sensor data are preprocessed and encoded before undergoing feature-level + decision-level fusion with image data to achieve quality identification and classification.Verified by multiple comparative experiments,the core performance of the intelligent sorting system is excellent:①Dual breakthroughs in sorting accuracy and efficiency: The mean Average Precision (mAP@0.5) for quality identification reaches 95.1%,the inference speed per sample is improved to 50ms,and the average F1-score attains 95.3%.②Significant comparative advantages: Compared with other research methods such as Faster R-CNN,the number of parameters of the proposed system is 15.5 times fewer than that of Faster R-CNN,with a substantial improvement in inference speed.Although the improved YOLOv8-Seg achieves higher accuracy,its large number of parameters makes it unfavorable for deployment on Raspberry Pi.③Remarkable improvements over traditional manual sorting:The system achieves dual significant enhancements in recognition accuracy and processing efficiency,effectively addressing the efficiency bottleneck and subjective bias of manual sorting.④Improved traceability function:Relying on the full-chain data binding capability of RFID electronic tags,the system can accurately trace key information of each batch of shiitake mushrooms,such as origin,cultivation environment,and circulation path,realizing real-time association between the sorting process and traceability information.⑤Strong practicality in complex environments:Through multimodal data fusion technology,the limitation of single-dimensional data recognition is broken,while the accuracy remains above 89% after fusing other data.

This study successfully realizes the integrated integration of accurate quality sorting and full-process quality traceability,further ensuring the stability and reliability of quality classification,which fully meets the technical requirements of smart agriculture for standardized sorting.