Polarized SAR image classification method based on superpixels and topic model

Abstract

A polarized SAR image classification method based on superpixels and a topic model comprises steps of: inputting a polarized SAR image to be classified, and subjecting the same to refined Lee filtering; acquiring the sample points of the image processed the step 1 by using uniform sampling to obtain a sample point set; extracting three types of features from sample point set and normalizing the three types of features separately to obtain feature sets F1, F2 and F3; clustering the feature sets F1, F2 and F3 separately to form visual dictionaries V1 and V2 , V3, and merging the visual dictionaries V1 and V2 , V3 to form a multi-feature visual dictionary V; performing over-segmentation on the image processed in the step 1 to obtain a plurality of superpixels, and performing sparse coding on each of the superpixels according to the dictionary V; performing feature learning on the sparse coding of the superpixels by using the topic model, and performing classification by using a SVM classification method to obtain a final classification result. The polarized SAR image classification method can represent the heterogeneous region of the polarized SAR image.

Description

technical field [0001] The invention belongs to the technical field of image processing and remote sensing, and relates to a polarization SAR image classification method based on superpixels and topic models. Background technique [0002] Compared with Synthetic Aperture Radar (SAR) images, polarimetric SAR images, as multi-channel SAR images, are rich in more polarization information. The classification of ground objects is the basic task of image processing, and it is also a national development. The major demand has attracted more and more people's attention. However, for heterogeneous areas such as urban areas and forests, due to the hybridity of ground objects, it is difficult for traditional object decomposition-based methods to divide images into semantically consistent ground object areas, which is also a challenge for polarimetric SAR image classification. [0003] Polarimetric SAR images contain rich polarization scattering information. Traditional polarization SA...

AI Technical Summary

Problems solved by technology

[0004] Although the above polarimetric SAR classification algorithm considers the scattering characteristics and spatial information of polarimetric SAR, there are still many defects: (1) For polarimetric SAR heterogeneous areas, such as urban areas and forests, since urban areas are composed of multiple buildings Formation, the scattered echo will form a strong change of light and dark, and this change of light and dark occurs repeatedly, forming an urban area. Because the existing algorithm does not consider the high-level semantic information of the image, it is difficult to divide heterogeneous areas into similar areas with consistent semantics

(2) Due to the lack of fusion of scattering features and image characteristics, it is difficult for traditional algorithms to learn high-level discriminative features, and to classify polarimetric SAR images, high-level semantic features and classification methods that consider polarization characteristics have yet to be explored

[0005] To sum up, there are still some defects in the existing polarization SAR image classification methods. It is difficult to divide the polarization SAR heterogeneous regions into regions with consistent semantics, and it is difficult to effectively classify the polarization SAR ground objects.

Images