Fig 1.
The yellow box represents the study area of this paper, the blue box represents the coverage of Sentinel-1 data from the ascending orbit, and the red box represents the coverage of Sentinel-2 data.
Table 1.
Winter wheat growth period and SAR image parameters.
Table 2.
Number of samples.
Fig 2.
It mainly includes data preprocessing, feature extraction and feature discussion and analysis.
Fig 3.
(a) is an average backscattering coefficient map under VH polarization over multiple periods; (b) is an average backscattering coefficient map under VV polarization over multiple periods; (c) is the SARadd map according to the Formula (2) and (d) is the NDVI map at maturity. All the legends are marked in the lower right of the figures.
Fig 4.
Statistical map of the mean values of the four classes.
(a) is the mean and the deviation of backscattering coefficient under VH polarization; (b) is the mean and the deviation of backscattering coefficient under VV polarization; (c) is the mean and the deviation of radar index SARadd; (d) is the mean and deviation of the NDVI at the maturing. The numbers in the figure (a), (b) and (c) represent backscattering coefficient. The numbers in the figure (d) represent NDVI.
Fig 5.
Decision tree constructed for extracting the winter wheat.
The black diamond box indicates the setting of the NDVI threshold extracted at maturity and the constructed radar index SARadd threshold, respectively. Together, they complete the classification of the ground objects.
Fig 6.
A, B, C, D, E, F, G, H, I represent regreening, erecting, jointing, booting, heading, flowering, seeding, waxing, maturing of the winter wheat, respectively. (a) and (b) represent the temporal backscattering coefficient curves under VH and VV polarizations, respectively.
Table 3.
Results of J-M distance.
Fig 7.
SAR and optical image for classification.
(a) is the muiti-temporal composite SAR pseudocolor image. The channels are R (VH backscattering on March 31, 2022), G (VH backscattering on April 24, 2022), and B (VH backscattering on May 06, 2022). (b) is the Sentinel-2 image which aquired on May 4, 2022. It is composed of B1, B2 and B4.
Fig 8.
Decision tree classification results and zoomed-in regions a, b, c.
Fig 9.
(a) is the classification result from multi-temporal composite SAR pseudocolor image, (b) is the classification result from single Sentinel-2 image.
Fig 10.
The extraction area of winter wheat and relative error with official statistics data.
(a), (b) and (c) indicate the classification result of the decision tree method of this paper, Single Sentinel-2 image and Multi-temporal composite SAR pseudocolor image, respectively.
Table 4.
Classification accuracy of different classification methods.
Fig 11.
Classification results with different features.
(a) is the classification result only using SARadd feature. (b) is the classification result using both SARadd feature and NDVI.
Table 5.
Classification scheme and accuracy.
Fig 12.
Comparison of spatial details.
(a) is the result from Sentinel-2 RGB image. (b) is the result from multi-temporal composite SAR pseudocolor image. (c) is the result from single Sentinel-2 image. (d) is the classification result of scheme1. (e) is the classification result of scheme 2. Region a, Region b, Region c are randomly selected subregions used to demonstrate the advantages of the method in this paper.