Crop Classification with Multi-Frequency Multi-Temporal Polarimetric and Interferometric SAR Data

Abstract

The combination of radar-derived features extracted from a dataset formed by multi-temporal polarimetric images and interferometric products acquired at various frequency bands has been evaluated for crop classification in this work. Fully polarimetric images acquired by the F-SAR airborne sensor at three frequency bands (L, C and X) and at three separate dates during the CROPEX14 campaign are exploited for this purpose. The images include consistent repeat-pass acquisitions with different baselines. Results show that all data dimensions contribute to improve the classification accuracy. The main novelty of this work is the joint use of different frequency bands (in groups of two and three) and specific interferometric data suitable for crops (short temporal baselines and large spatial baselines). The achieved performance with different combinations of input features is analysed both globally and for specific crops. Among other conclusions, L- and X-band show the best performance of all band pairs. Since they are the most extreme, their complementary sensitivity is maximised when used together.