Abstract:
Abstract: Because of the long-time influence of rain, cloud and fog in summer and autumn, the information of agricultural land is difficult to obtain instantly and accurately in local agricultural areas in North China. Radar remote sensing has advantages of all-time, all-weather and high penetration etc, and it can be widely used in cloudy regions. In applications of multi-polarization radar data, polarization decomposition model can get effective polarization characteristics to extract land features accurately. The Freeman decomposition model is a polarization decomposition model used frequently, but it can only be used in the circumstances that satisfy the demand of reflection symmetry, which limits the use of the model to further improve the classification accuracy of remote sensing to a certain extent. On the basis of analyzing the reason that the Freeman decomposition model is not suitable for agricultural area, this paper proposed an automatic extracting method of agricultural land of SAR image using optimized three-component decomposition model (OTDM) which is the improvement of the Freeman decomposition model. In this study, firstly, the orientation processing was joined into polarization decomposition model to inhibit the production of negative power. And the Freeman decomposition model was optimized by introducing volume scattering parameter, secondary scattering parameters and Bragg scattering parameters, so as to improve the performance of Freeman decomposition model which has the shortcoming of lacking adjustable parameters, and make the decomposition results adaptable to the scattering characteristics of different surfaces of agricultural area. Then, combining OTDM and fuzzy C-means clustering (FCM), after land feature categories were merged, agricultural land information was extracted in an automatic way. The results of experiments indicated that when parameters were equivalent to 3, 1.75, 10 and 0.001, respectively, the classification from FCM achieved the better result. Finally, both of the H-Alpha-Lambda and OTDM-FCM were applied in an experiment and compared with the ground samples to verify the effectiveness of OTDM-FCM. In this experiment, the study area was located in Zaoqiang County, Hebei Province in Huang-huai-hai Plain, and Radarsat-2 images were used as the radar data source. The experiment was carried out under the circumstances of full and partial cover crop by selecting the images in appropriate time. The final results of the experiment indicated that under the circumstance of full cover crop, overall accuracy and Kappa coefficient of OTDM-FCM were 96.12% and 0.857, respectively, while the results of H-Alpha-Lambda were 87.43% and 0.520, respectively; under the circumstance of partial cover crop, overall accuracy and Kappa coefficient of OTDM-FCM were 97.53% and 0.902, respectively, while the results of H-Alpha-Lambda were 80.16% and 0.307, respectively. It could be concluded that the classification extraction accuracy of OTDM-FCM was superior to H-Alpha-Lambda classification under the circumstances of both full and partial cover crop. Therefore, under the conditions of different imaging time and different extents of crop covered, OTDM-FCM classification algorithm could effectively extract the information of agricultural land, and it was shown that OTDM-FCM classification had certain feasibility and applicability in the extraction of agricultural land depending on SAR image information. This method put forward in this paper could provide a new thinking for the application of SAR image in the extraction of agricultural land information.