Semi-supervised classification method for hyperspectral remote sensing images based on local manifold learning composition

Abstract

The invention discloses a hyperspectral remote sensing image semi-supervised classification method based on local manifold learning composition. The realization steps are: (1) preparing a training sample set, including a small amount of labeled data and a large amount of unlabeled data; (2) Based on the distance measurement method of spectral angle mapping, select k nearest neighbor points for each sample point in the training sample set; use the local manifold learning algorithm to obtain the weights between the connection points in the graph structure, and calculate the graph adjacency matrix , to get the corresponding graph structure; based on the graph adjacency matrix, based on the GFHF algorithm to classify the unlabeled data; (3) use the generalization algorithm of GFHF to classify other data points in the image. The invention connects two widely used algorithms, the local manifold learning dimensionality reduction algorithm and the semi-supervised classification algorithm, through a "graph", and shows good applicability to the classification of various hyperspectral remote sensing data, and can significantly improve the Classification Accuracy of Spectral Remote Sensing Images.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and in particular relates to a semi-supervised classification method based on local manifold learning composition, which is used for hyperspectral remote sensing image classification. Background technique [0002] Classification of ground objects using remote sensing images is to identify the physical properties of ground objects based on the acquired remote sensing information. The classification results can clearly reflect the spatial distribution of ground objects for the production of thematic maps, and it is convenient for people to understand and discover laws from them. decision making. Land feature classification has important applications in agricultural monitoring, soil survey, mineral mapping, and urban environmental monitoring. [0003] Hyperspectral remote sensing is a remote sensing science and technology with high spectral resolution. It has the characteris...

AI Technical Summary

Problems solved by technology

[0005] In hyperspectral remote sensing image classification, the heat kernel function (Heat Kernel, HK) is currently used to calculate the image weight, but this weight can only reflect the relationship between two points, and does not make full use of the local area of ​​the neighborhood where the data point is located. structure information

In the field of machine learning, in 2008, some scholars proposed a graph weight calculation method based on non-negative local linear reconstruction (Local Liner Reconstruction, LLR) (F. Wang and C. Zhang, "Label propagation through linear neighborhoods," IEEE Transactions on Knowledge and DataEngineering, vol.20, no.1, pp.55-67, Jan.2008.), considering the local structure information of the data, but the non-negative constraints on the weights may make the weight of the points on the manifold boundary structure effect is not good

In 2009, Yan proposed a graph weight calculation method based on sparse representation (Sparse Representation, SR), but due to the need to solve the reconstruction weight globally, the amount of calculation is large (S.Yan and H.Wang, "Semi-supervised learning by sparse representation,"in Proceedings of SIAM International Conference on DataMining,Nevada,USA,Apr.2009,pp.792-801.)

Images