Sugarcane and Cassava Classification Using Machine Learning Approach Based on Multi-temporal Remote Sensing Data Analysis

Abstract

Crop identification and mapping provide valuable information about crop acreage and aid in monitoring and decision-making for government and agro-industrial businesses. Multi-temporal remote sensing is widely used for the phenological study of crops, especially in sugarcane and cassava. The development of crop growth monitoring is one of the capabilities of multi-temporal data. This study integrated the phenological characteristics from remote sensing data to develop the optimal classifier for the reliable classification of sugarcane and cassava. Sentinel-1 and Sentinel-2 images from October 2017 to September 2019 were used to classify the crop types. The random forest (RF) classifier provided the highest overall model performance accuracy than other algorithms when using the specific NDVI dataset. Furthermore, the ground truth data collected from an unmanned aerial vehicle (UAV) field survey was used to assess the classification performance and resulted in 68% accuracy for sugarcane and cassava classification. Multi-temporal remote sensing can aid in the mapping of sugarcane and cassava. The developed approach can be used for crop mapping, management, and estimation of crop production on a regional scale. Expectedly, our development could be practically adopted on a multi-crop plantation scale. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.