Crop classification is a crucial prerequisite for the collection of agricultural statistics, efficient crop management, biodiversity control, the design of agricultural policy, and food security. Crops are characterized by significant change during the growing season, and this information can be used to improve classification accuracy. However, capturing variation in vegetation cover requires a reliable source of valid data. Sentinel-1 radar images are a good candidate, as they supply information about Earth’s surface every six days, independent of weather and light conditions. In this paper, we present a method for crop classification based on radar polarimetry. We propose a set of multi-temporal indices derived from time series Sentinel-1 images that aim to characterize crop phenology. A big data, object-oriented classification technique is developed and tested on 16 crop types for the whole of Poland. Our analysis found that overall accuracy varied (regionally) from 86.36 to 89.13% in 2019, and from 85.95 to 89.81% in 2020. F1 scores for individual crops varied from 0.73 to 0.99, and the use of our multi-temporal phenological indices increased F1 scores by about 0.14 compared to calculations using only basic parameters. Results obtained for the whole country demonstrate the efficacy of the method and its resistance to environmental conditions.
Woźniak E., Rybicki M., Kofman W., Aleksandrowicz S., Wojtkowski C., Lewiński S., Bojanowski J., Musiał J., Milewski T., Slesiński P., Łączyński A. (2022), Multi-temporal phenological indices derived from time series Sentinel-1 images to country-wide crop classification