Due to the change of the ridge spacing of sugarcane plants, it is difficult for the conventional cultivator to control the quality of soil cultivation on both sides of the ridge, which causes inadequate soil cultivation, i.e., the “crater”phenomenon. We thus developed a multi-functional sugarcane intertillage cultivator and proposed a machine learning-based method to locate and identify sugarcane seedlings and calculate the coordinate classification during the intertillage period. Based on the YOLOv4 network, the method established a recognition model, which identifies and locates the area between the roots and the soil and accesses the coordinates. Then, the Support Vector Machine divided the coordinate data into two groups and processed them separately in real-time to obtain the tilt value. Finally, the data were adjusted according to the tilt value to modify the quantity and direction of soil supply. The results of the present study indicate that in use of the YOLOv4 recognition model, the recognition accuracy can reach 95.50%, and the classification accuracy using the SVM is 92.60%, which realizes the real-time dynamic recognition and classification calculation of sugarcane seedlings in the intertillage period, and provided data foundation for the development of intelligent sugarcane protection and joint operation machinery.