Pritam Das, Mahesh Kothari, Pradeep Kumar Singh, Manjeet Singh, Naveen Jain and Vinod Kumar
Abstract
Accurate and timely information on crop areas is essential for informed decision-making in agriculture, from farm management to food security assessments. Synthetic Aperture Radar (SAR) data has emerged as a game-changer in this field due to its ability to image landscapes through cloud cover and its sensitivity to vegetation structure. This review paper delves into the various methodologies employed to utilize SAR data for effective crop area estimation. We begin by the significance of crop area information and progressively dive into the crucial image processing techniques used in SAR data analysis. These techniques encompass preprocessing steps to remove noise and artifacts, segmentation to delineate individual crop fields, and feature extraction to capture relevant crop characteristics that influence the radar backscatter. Following this, we examine the classification algorithms employed to identify and differentiate crop types within the segmented regions. We discuss both established supervised machine learning algorithms like Support Vector Machines (SVM) and Random Forest (RF) and unsupervised machine learning algorithms like K-means clustering and cutting-edge deep learning approaches like Convolutional Neural Networks (CNNs) for robust and automated crop classification. Furthermore, we explore the concept of time series analysis in the context of SAR data. By incorporating data acquired at different points throughout the growing season, these methods leverage the dynamic changes in backscatter associated with various crop growth stages. This improves differentiation between different crop types and facilitates more precise area estimation. Finally, we acknowledge the limitations and challenges associated with SAR-based crop area mapping, such as the sensitivity to soil moisture content and the potential confusion between crops with similar characteristics.