Sugarcane (Saccharum spp.) is one of the most significant sugar and bioenergy crops globally. Traditional methods for detecting effective tillers in sugarcane rely on manual field surveys, which are not only time-consuming but also expose workers to harsh environmental conditions. Therefore, we constructed a dataset of 2274 images and proposed ST-YOLO models for effective sugarcane tiller detection. Experimental results revealed that ST-YOLOs achieved an improvement of 0.047 in mAP_@50:5:95 and a 7.32 % reduction in parameters compared to the baseline model YOLO 11s. The 10-fold cross-validation results demonstrated the model achieved an average mAP_@50:5:95 of 0.798 and an RMSE of 0.006, confirming its robustness. The Pearson correlation coefficient (r) between machine and manual counting reached 0.902. ST-YOLOn maintained high detection precision while reducing computational consumption. Additionally, a sugarcane effective tiller detection application was developed by PyQt5, supporting real-time image detection. This approach provides an efficient solution for automated monitoring of sugarcane effective tillers, with primary applications in smart breeding and precision management.