Polarized SAR classification method based on clustering refinement residual error model

Abstract

The invention discloses a polarized SAR classification method based on a clustering refinement residual error model. The method comprises the following steps of (1) constructing the clustering refinement residual error model; (2) preprocessing a polarized SAR image to be classified; (3) generating a training data set and a test data set; (4) carrying out information fusion processing on the deep and shallow layer of a network; (5) carrying out reclassification processing on the small graph spot of an initial classification graph; and (6) classifying test data and acquiring a test result. In the invention, through carrying out fusion processing on the deep and shallow layer information of the polarized SAR image in a refinement residual error network, the abundant texture characteristics ofthe polarized SAR image are extracted, characteristic information integrity is reserved, a training speed is increased; and a clustering layer is used to reclassify and process the small graph spot of an edge in the classification graph after fusion processing so that the classification precision of the graph edge is increased and the training speed is accelerated.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a polarization synthetic aperture radar SAR (Synthetic Aperture Radar) image classification method based on a clustering and refinement residual model in the technical field of polarization synthetic aperture radar image ground object classification. The invention can be used to classify ground objects in polarimetric SAR images. Background technique [0002] Polarization synthetic aperture radar has many outstanding advantages, such as not being affected by time, and can image 24 hours a day. Polarization SAR images have unique advantages and broad application prospects, and have been successfully used in land use classification, change detection, surface parameter inversion, soil moisture and soil moisture inversion, artificial target classification, building extraction, etc. [0003] With the further development of full-polarization SAR remote sensing technolog...

AI Technical Summary

Problems solved by technology

Although this method uses multi-objective decomposition to obtain comprehensive polarization characteristics, the shortcomings of this method are that the feature information is not comprehensive, resulting in low classification accuracy, and the training speed of SVM is faster than that of residual neural network. training is much slower

However, the shortcomings of this method are that the full convolutional network is too deep, which leads to too long network training time, and the full convolutional neural network often has misclassifications, missing points, and separation of images when classifying images. The spots of the image are relatively messy, so there are many small spots on the edge of the image, and the classification effect of the edge pixels of the image is poor

Images