Maize field area detection in East Java, Indonesia: An integrated multispectral remote sensing and machine learning approach

Abstract

An accurate and high quality of agricultural monitoring and statistics commonly requires a huge amount of resources in terms of human, cost, and time. In this paper, we introduce a cost-efficient, scalable, and accurate framework for multilabel classification of the maize (corn) field area using remote sensing approaches. Official statistical survey results are used to provide the ground truth labels. Five vegetation indices, which include the enhanced vegetation index (EVI), normalized difference vegetation index (NDVI), normalized difference water index (NDWI), normalized difference built-up index (NDBI), and visible atmospherically resistant index (VARI), are used to enhance the multitemporal features and predictor variables. We train an ensemble machine learning model, random forest (RF) as the classifier. Experiments are carried out to detect maize field areas in ten regencies of East Java, Indonesia using multispectral imagery data acquired by Landsat-8, Sentinel-1, and Sentinel-2 satellites. The results show that our proposed approach gains a promising accuracy of up to 87 percent in detecting maize field area. We believe that our framework could be beneficial to support and improve the quality of official statistics in the agricultural sector.