Abstract: Blueberry ripeness detection aims to automatically determine blueberries ripeness based on input images captured in a natural environment, thereby providing technical support for intelligent blueberry harvesting．On basis of collecting blueberry images in field to build a dedicated dataset, a multi-ripeness detection model was trained based on YOLOv11n, and optimized model was deployed to K230 embedded platform．Experimental results showed that this model achieved high accuracy, with precision rate, recall rate, mAP_0.5, and mAP_0.50～0.95 indicators reaching 90.1%, 81.0%, 88.1%, and 64.9% respectively．Its parameter quantity was only 2582737, and it achieved a single-frame image inference speed of 9.5 ms, an inference speed on embedded device of only 240 ms．This system was both high accuracy and real-time performance．Its lightweight characteristics and embedded compatibility provided a new method for practical application of intelligent blueberry harvesting robots．