Machine learning for sugarcane mapping based on segmentation in cloud platform

Abstract

Sugarcane mapping is very important in precision agriculture. Mapping can also help enforce policies in land provision of sugarcane agriculture. Currently, sugarcane mapping is often done conventionally, this is less effective because it is expensive and labor-intensive. Reliable mapping at a low cost and fast can use remote sensing and artificial intelligent. This study proposed an automatic sugarcane classification method based on segmentation on a cloud platform. The proposed method uses machine learning algorithms, namely Classification and Regression Trees (CART), Random Forest (RF), Support Vector Machine (SVM) for classification. Then, the segmentation algorithm used a combination of Simple Non-Iterative Clustering (SNIC) for cluster identification and Gray-Level Co-occurrence Matrix (GLCM) for texture extraction. Segmentation is divided into several intervals to get the best results from the classification method. All stages are performed on the Google Earth Engine (GEE) cloud platform based on the use of Landsat 8 images in 2019-2020. The results obtained the best classification is Random Forest (size segmentation=8, Overall Accuracy=99.0%, Kappa Coefficient=0.99), Classification and Regression Trees (size segmentation=8, Overall Accuracy=97.5%, Kappa Coefficient=0.96) and, Support Vector Machine (size segmentation=15, Overall Accuracy=75.0%, Kappa Coefficient=0.58).