High Accuracy Pre-Harvest Sugarcane Yield Forecasting Model Utilizing Drone Image Analysis, Data Mining, and Reverse Design Method

Abstract

This article presents a new model for forecasting the sugarcane yield that substantially reduces current rates of assessment errors, providing a more reliable pre-harvest assessment tool for sugarcane production. This model, called the Wondercane model, integrates various environmental data obtained from sugar mill surveys and government agencies with the analysis of aerial images of sugarcane fields obtained with drones. The drone images enable the calculation of the proportion of unusable sugarcane (the defect rate) in the field. Defective cane can result from adverse weather or other cultivation issues. The Wondercane model is developed on the principle of determining the yield not through data in regression form but rather through data in classification form. The Reverse Design method and the Similarity Relationship method are applied for feature extraction of the input factors and the target outputs. The model utilizes data mining to recognize and classify the dataset from the sugarcane field. Results show that the optimal performance of the model is achieved when: (1) the number of Input Factors is five, (2) the number of Target Outputs is 32, and (3) the Random Forest algorithm is used. The model recognized the 2019 training data with an accuracy of 98.21%, and then it correctly forecast the yield of the 2019 test data with an accuracy of 89.58% (10.42% error) when compared to the actual yield. The Wondercane model correctly forecast the harvest yield of a 2020 dataset with an accuracy of 98.69% (1.31% error). The Wondercane model is therefore an accurate and robust tool that can substantially reduce the issue of sugarcane yield estimate errors and provide the sugar industry with improved pre-harvest assessment of sugarcane yield.