Hyperspectral image classification method and system based on center-domain interactive learning

Abstract

The invention discloses a hyperspectral image classification method and system based on center-domain interactive learning, and the method comprises the steps: firstly introducing the concept of center observation pixels and neighborhood auxiliary pixels; a central region, a neighbor region and a surrounding region are generated from the near to the distant by taking each central observation pixel as a center by using a hierarchical region sampling strategy based on a region segmentation idea, and data support is provided for subsequent processing. Then, a central area containing a central observation pixel is sent to a central Transform branch, and fine-grained feature expression is obtained; next, a neighbor region and a surrounding region containing neighborhood auxiliary pixels are sent to a neighborhood Transform branch, and coarse-grained feature expression is obtained; and finally, fusing the features output by the two branches, and sending the fused features into a multi-layer perceptron to complete classification. According to the method, low-level detail features of ground features are considered, global high-level semantic information is reserved, and the hyperspectral image classification performance can be enhanced.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to an image classification method and system, in particular to a hyperspectral image classification method and system based on center-domain interactive learning. Background technique [0002] Hyperspectral images have the characteristics of unified atlas, numerous bands, and high spectral resolution. They contain rich spectral features and spatial information, and provide an effective solution for the fine classification of ground objects. With the vigorous development of deep learning techniques, convolutional neural networks, fully convolutional neural networks and recurrent neural networks have achieved good performance in hyperspectral image classification tasks. Specifically, convolutional neural networks have the properties of local connection and weight sharing, which can automatically extract deep robust features without requiring too many parameters. However, most ...

AI Technical Summary

Problems solved by technology

However, this way of feature extraction is static, which greatly ignores the interaction between the central observation pixel and neighboring auxiliary pixels in the image, which brings new challenges to the fine classification of hyperspectral images.

Images