Detection of different Crops Types using RapidEye Imagery over Fresno, California

Monitoring the crops periodically and implementing sustainable agricultural practices are the considerable conditions for implementing efficient methodologies for Cropland Mapping. Detection of crop type, acreage calculation, and yield detection are the key research points that are under consideration of the research in today’s era. In this research work, the authors have focused on exploring the ability of Rapid-Eye data satellite images for crop identification/classification. A hybrid crop classification model was proposed for temporal and multispectral RapidEye imagery over the area of Fresno county of California state, United States. The studied area is enriched in a specific crop such as grapes, almonds, alfalfa, and dbl crop winwht/corn that covers 47.34%, 26.93%, 8.79%, and 3.51% of the area respectively. The ground truth data can be downloaded from the United States Department of Agriculture, CropScape-Cropland Data Layer by specifying a few parameters such as year, and area of interest (based on region, state, and county). The proposed classification model was designed by introducing the combination of a convolutional 1-D layer and two Gated Recurrent Unit layers and is aimed to obtain the best performance in terms of quantitative parameters. The authors introduced these layers to predict class probabilities based on a temporal sequence of observations. The proposed model outperforms for time-series data and resulted in a noticeable improvement in an overall accuracy of 96.31% and a kappa coefficient of 0.95.