The comparison of numerous machine learning algorithms performance in classifying rice growth stages based on Sentinel-2 to enhance crop monitoring in national level

Abstract

The rice monitoring based on Sentinel-2 (SC-S2) has been developed for over nine months. It has been observed as the first and only system which generate rice growth stages maps in 10 m spatial resolution using machine learning in Indonesia. However, the SC-S2 use Support Vector Machine to separate the rice growth stages, which may have poor performances. The objective of this study is to investigate the performance of other classifiers to increase the performance of SC-S2. We used survey data from the field campaign in 2018 and synchronized with Sentinel-2 bands. The model dataset was trained using 61 machine learning algorithms to create 61 rice growth stages models. The models were applied to the Sentinel-2 image of part of Indramayu area. The accuracy, computational time and visual inspection score were collected, and the final score was calculated. The results are the highest final score is Shrinkage Discriminant Analysis, with overall accuracy 88.1% (p<0.001) and the average accuracy of all classifiers is 76.2% (p<0.05). The implication of this study is to propose some changes in the classification process into the SC-S2 for increasing the overall performance, which will provide better information for agricultural policymakers.

Parcel-Level Mapping of Horticultural Crop Orchards in Complex Mountain Areas Using VHR and Time-Series Images

“…Studies have shown that, among these satellite products, the spatial resolution of optical images used for classification has little effect on the accuracy of classification in areas with large parcels, while Sentinel-2 images can provide better classification results than Landsat and MODIS images for areas with fragmented farmland [12]. Many scholars have carried out research on crop classification based on Sentinel-2 images [13][14][15][16].…”