Superpixel-level feature extraction method based on spatial pyramid pooling

Abstract

The invention relates to a radar image processing and interpretation technology, and in particular relates to a feature extraction method used for extracting superpixel-level features of a polarization SAR (Synthetic Aperture Radar) image. According to the method in the invention, the superpixel-level features of the polarization SAR image are extracted by utilization of a spatial pyramid poolingmodel; all scales of features of various feature layers in a superpixel block can be connected in series, so that overall features of the superpixel block are obtained; the superpixel-level feature extraction time is greatly reduced; requirements on a classifier are reduced; and a polarization SAR image classification result is improved.

Description

technical field [0001] The invention belongs to radar image processing and interpretation technology, and in particular relates to a superpixel-level feature extraction method based on spatial pyramid pooling. Background technique [0002] Today in the 21st century, remote sensing and earth observation technology is playing an increasingly important role in the military and civilian fields. Synthetic Aperture Radar (SAR) is an important research direction in the field of remote sensing. SAR is an active ground sensor that analyzes ground surface scattering information by emitting and receiving specific electromagnetic waves. Compared with traditional optical sensors, SAR is not affected by light and weather, and can realize all-weather and all-weather earth observation. In today's increasingly complex and changeable climate environment and battlefield environment, SAR has more and more important significance. Compared with traditional SAR, polarimetric SAR provides richer...

AI Technical Summary

Problems solved by technology

[0005] (1) When the selected dictionary is too large, the words lack generality, are sensitive to noise, and the amount of calculation is large; if the dictionary is too small, the word discrimination performance is poor, and similar features cannot be represented

[0006] (2) The similarity measure function is used to classify the image features to the corresponding words in the word book, but the similarity measure suitable for polarized SAR features is difficult to express

[0007] (3) Whether the center point selected by the selection method of the clustering center can represent the overall polarization SAR data, the existing clustering methods such as K-means tend to converge to the local optimum and are sensitive to the outliers of the features

Images