A Polarization SAR Image Classification Method Based on Multi-Feature Fusion

Abstract

The invention discloses a polarization SAR image classification method based on multi-feature fusion. The present invention firstly extracts the polarization feature vector of the polarimetric SAR image to be classified to obtain a high-dimensional polarization feature set; extracts the morphological section feature vector of the image SPAN processing result to obtain a high-dimensional morphological feature set; After the features are subjected to the dimensionality reduction processing of the locality-preserving discrimination analysis, the image pixels with known category labels are selected to form the training sample set, and then the remaining pixels of the entire image are selected as the classification sample set; the two types of low-dimensional features are respectively passed through the The SVM of the maximum posterior probability obtains the category label and the corresponding posterior probability of the pixel points in each case; the summation criterion or the adaptive weighted summation criterion is used to combine the posterior probability vectors of each pixel in the two situations According to the principle of maximum a posteriori probability, the final classification result of high-resolution polarimetric SAR images is obtained. The implementation of the invention helps to improve the classification accuracy and efficiency of resolving polarimetric SAR images.

Description

technical field [0001] The invention belongs to radar image processing and interpretation technology, that is, ground object classification processing is performed on high-resolution polarization synthetic aperture radar images, and in particular relates to a polarization SAR image classification method based on multi-feature fusion. Background technique [0002] In recent years, my country has successfully applied Synthetic Aperture Radar (SAR) in various fields. The successful application of SAR depends on effective SAR image processing and interpretation technology, and its premise is the classification technology of SAR image. Through classification, the ground object information in SAR images can be effectively obtained, and it can provide assistance for various applications such as urban planning, crop and forest observation, disaster assessment, and ground target recognition. Therefore, SAR image classification technology occupies a very important position. [0003] ...

AI Technical Summary

Problems solved by technology

Typical dimensionality reduction methods include linear dimensionality reduction methods and manifold learning methods. Linear dimensionality reduction methods are based on the Gaussian hypothesis, but real data are often in complex nonlinear spaces, requiring nonlinear methods to explore potential structures, and The linear dimensionality reduction method does not have the ability to cover the geometry and local distribution structure in the data space, and this information is crucial for classification; the manifold learning method in the nonlinear dimensionality reduction method assumes that the features can be retained at a low In the three-dimensional manifold structure, however, the inherent unsupervised and non-discriminative nature of manifold learning makes it difficult to apply in practice, and the feature vectors of different categories cannot be separated optimally, which is easy to cause misclassification and misclassification; (4) combining The most prominent problem in the two types of features is how to effectively combine different types of features, which requires the so-called information fusion technology

The difficulty of information fusion is that the space of different features is different, and the corresponding kernel functions are also different; the second is that the original feature information may be destroyed, especially when the magnitude of the two features differs greatly; The increase in capacity will easily lead to a decrease in computing efficiency

At present, there is no in-depth study on the information fusion method in the interpretation of polarimetric SAR images

Images