Hyperspectral image classification method based on node pyramid

Abstract

The invention relates to a hyperspectral image classification method based on a node pyramid, and the method comprises the steps: taking each foreground pixel in a hyperspectral image as a node in a graph structure, taking a spectral feature as a node feature, and constructing a connection relation between nodes according to the similarity of the nodes, thereby obtaining a bottommost layer graph structure; clustering the nodes by using a classic hierarchical clustering method to obtain N graph structures with different fine granularities; performing node classification on the graph structures of the N hierarchies by adopting a classic graph convolutional neural network; and performing voting correction on a high-fine-granularity result based on a low-fine-granularity classification result. According to the method, the problem that scattered points in a hyperspectral image distributed in a strip mode are prone to being wrongly classified is solved, and the problem of training time consumption caused by too many nodes in classification based on a graph network is solved.

Description

Technical field [0001] The present invention relates to a hyperspectral image classification method based on a node pyramid, belonging to the field of hyperspectral image classification technology. Background [0002] Diagrams are an intuitive, abstract mathematical representation of objects and their interrelationships. Data with interrelationships—Graph structure data is ubiquitous in many fields and is widely used. With the emergence of large amounts of data, traditional graph algorithms have great limitations in solving some deep-seated important problems, such as node classification and link prediction. The graph neural network model considers the scale, heterogeneity and deep topological information of the input data, etc., and shows convincing reliable performance in mining deep and effective topological information, extracting key complex characteristics of data and realizing rapid processing of massive data, such as predicting the characteristics of chemical molecules, ...

AI Technical Summary

Problems solved by technology

[0004] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art, to propose a hyperspectral image classification method based on node pyramid, to classify based on layered graph nodes, to obtain hierarchical classification results, combined with multi-level classification results, To maximize the accuracy of image classification, solve the problem of easy misclassification of scattered points in hyperspectral images distributed in strips, and the problem of time-consuming training caused by too many nodes in graph network-based classification

Images