Prediction of Phenotypic Parameters of Sugarcane Plants Based on Multispectral Drone Imagery and Machine learning

Measuring phenotypic parameters is important in evaluating sugarcane productivity. Existing manual measurements are considered less efficient so better alternative methods is needed. This research aims to explore the potential of using multispectral drone imagery and machine learning to estimate phenotypic parameter of sugar cane plants that are efficient, accurate, inexpensive, and support sustainable agricultural practices. Spectral data captured by drones, including Red, Red-Edge and NIR, are used as input to estimate phenotypic parameters which include Brix values, number of stands, diameter and plant height. Based on the results of the development of the machine learning model, the ANN algorithm model is most effective in predicting the Brix value with R2 0.75 and RMSE 0.06 and the number of stands with R2 0.63 and RMSE 2.29. All models cannot predict plant diameter and height well. The best model was obtained using the RF algorithm with R2 0.26 and RMSE 0.65 for diameter and height, R2 0.33 and RMSE 17.27 for sugarcane plant height. The SVR algorithm model shows inconsistent performance compared to other algorithms. This indicates that the effectiveness of an algorithm depends on the specific parameters being predicted and there is no algorithm that is dominant for all phenotypic parameters.