Sugarcane is a significant and globally important economic crop, with a close correlation between plant height and yield. Traditional methods for determining sugarcane height characterization rely on labor-intensive and time-consuming manual detection. Therefore, this study collected image datasets of sugarcane crops in various scenarios and proposed lightweight enhanced PSPNet and YOLOv5s models for segmenting sugarcane bodies and detecting the tip positions. Additionally, an automated measurement robot was developed, deploying the improved models for real-time inference on images captured by a depth camera and implementing a gimbal automation system. Finally, sugarcane positions and depth information were utilized to calculate sugarcane height.The experimental results show that the average absolute error of the sugarcane height measurement method is less than 2.4 cm, the accuracy is higher than 97.61%, the success rate is higher than 93%, and the detection time is less than 13.2 s. In addition, the proposed sugarcane trunk segmentation and top detection model parameters are significantly reduced, with accuracies maintained at around 87% and 97%, respectively.