The Research On Full Polarization Image Classification Based On Multi-band

As an active microwave remote sensing technology, Synthetic Aperture Radar (SAR) is better than the traditional optical remote sensing for its all-day and all-weather monitoring. It can provide people abundant information to study object types. As the time goes, the technology of SAR classification has improved a lot, the study of areas where the climate condition is not so good classification has a new way. And we also can use different bands of polarization SAR data to do the classification, it really helps a lot to it.In order to explore how to use the multi-band polarization data to SAR classification, this paper select Gen he city, Inner Mongolia as the study area, and combining respective the two advantages, use two different bands to classify the study area. The main work is as follows:1)First of all, to do some pretreatment of the two bands full polarization data, including geometric correction based on range Doppler model, and filtering to eliminate coherent noise. And then the image registration work, to resample the airborne based on the spaceborne data.2)Then in this paper to do the work of extracting the respective polarization feature of two wavelengths, such as the scattering angle, scattering entropy of the data image and so on. And then we will extract the features of both two bands based on noncoherent polarization target decomposition model of image decomposition, and do some analysis about it. As a conclusion, we can see that band P has better sensitivity for forests and at the same time band C is more suitable for the conclusion of the grass.3)Using two different ways to classify the SAR data image, we can see the supervised classification has a better result, and then we will evaluate the accuracy of the classification of this method.4)In this paper, the difference of two bands data penetrating will be studied.Because of the difference of the wavelength, they have different results on the image. Finally combine the two bands' respective advantages, extract the polarization feature of each data and then make these vectors into a high dimensional space for data fusion. Then we use the supervised classification way for the fusion data to get the final result. We analysis the final result of classification, and compare it with the results of single band.Results showed that: the data after fusion of the full polarization data has a better accuracy of the classification results than the two single band based results.It has an obvious improvement according to the results. False results and confusing targets existing in the image have been cut down effectively. The accuracy has improved from 50% to more than 80%. It has verified the method of data fusion classification, and also offers great potential for further study.