Integrated transfer learning classification method and system for single-view fully polarized SAR data

Abstract

The present invention provides an integrated migration learning classification method for single-view full-polarization SAR data, which adopts full image segmentation and partial window segmentation, and utilizes image space context information to concentrate source domain labeled samples and expand target domain labeled samples ; Then set the label sample selection process in the source domain by combining the scatter vector similarity coefficient and the fitness function, and adjust the source domain sample distribution used for training the evaluator model according to the target domain category center; introduce the reference information for the reliability of the weak classifier as the weight factor , use the weighted form to use the migration weak classifier for category soft prediction, give the unlabeled pixel pseudo-label of the target domain with the same prediction result, so as to expand the labeled sample size of the target domain and train the new weak classifier; the migration weak classifier generated by iterative utilization The set of classifiers completes the integrated classification of all pixels in the target domain. The invention can significantly reduce the high dependence of the radar image classification task on the label information of the target domain, and improve the classification efficiency and automation level.

Description

technical field [0001] The invention belongs to the technical field of fully polarimetric radar remote sensing image processing, and designs a new transfer learning method and system based on fully polarimetric synthetic aperture radar (Polarimetric Synthetic Aperture Radar, POLSAR) single-view data for integrated classification of ground objects. Background technique [0002] Remote sensing technology can conduct long-distance observation without direct contact with the target, and is an ideal way to obtain regional surface information. Supervised classification of remote sensing images is the process of using pixel samples with class labels to establish a recognition model to infer the category of other pixels in the image. It has great application potential in land use management and land cover change. It provides important support for planning land use, assessing the impact of natural disasters, and organizing rescue and reconstruction in disaster areas. Among various r...

AI Technical Summary

Problems solved by technology

However, the transfer learning technology at this stage mainly focuses on conventional pattern recognition tasks, and fails to combine the spatial context information in remote sensing images and the characteristics of radar scattering vector data to design algorithms, which can easily cause negative transfer effects when directly used in SAR images; , Polarization Synthetic Aperture Radar (POLSAR) uses multi-channels to obtain the backscattering information of ground objects and targets, and the classification results are easily affected by speckle noise

A common way to reduce the impact of noise is to use multi-view processing, that is, to use the spatial set average to obtain the second-order statistics of the scattering vector as the input features of the classifier, but this processing artificially reduces the spatial resolution and loses a lot of spatial details.

Images