Hyperspectral image classification method based on three-dimensional non-local mean filtering

Abstract

The invention discloses a hyperspectral image classification method based on three-dimensional non-local mean filtering, and mainly solves the problems that neighborhood information of a hyperspectral image cannot be effectively utilized, and spatial-spectral domain characteristics of the hyperspectral image cannot be effectively combined in the prior art. The hyperspectral image classification method comprises the following steps: 1) reading in hyperspectral image data having labels; 2) setting a size of a neighborhood; 3) carrying out three-dimensional non-local mean filtering on the input hyperspectral image data to obtain the spatial-spectral domain characteristics; 4) carrying out normalization on the filtered characteristics; 5) determining a training sample set and a testing sample set by employing the normalized characteristics; 6) carrying out training on an SVM classifier by employing the testing sample set and the corresponding labels; and 7) carrying out classification on the testing sample set by employing the trained SVM classifier to obtain a classification result. The hyperspectral image classification method has the advantages of being high in classification precision, small in cost, and simple and easy in operation, and can be used in terrain classification of three-dimensional hyperspectral image data.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and relates to a method for classifying hyperspectral images, which can be used for classifying ground features of hyperspectral images. Background technique [0002] In recent years, hyperspectral image classification has become a research hotspot in the field of hyperspectral remote sensing. The goal of hyperspectral image classification is to assign a specific class to each point pixel. The study found that the spectral characteristics and spatial characteristics of pixels of the same category are consistent, and the spectral characteristics and spatial characteristics of pixels of different categories are significantly different, so this characteristic can be used for hyperspectral image classification. Hyperspectral image data has the characteristics of high dimensionality and few samples. It is a typical three-dimensional image data. Each pixel contains hundreds of ...

AI Technical Summary

Problems solved by technology

However, this method ignores the connection between the spectral information and the spatial information in the hyperspectral image data. The spatial information is jointly reflected by the spectral band and the neighborhood, so the separate processing of the spatial information and the spectral information is prone to errors. Affects the accuracy of points, and extracting features separately will also increase the time to process data and increase complexity

[0004] The traditional non-local mean filtering is mainly proposed for two-dimensional images. Since the hyperspectral image data has a spectral dimension in addition to the spatial dimension, that is to say, the hyperspectral image data is a typical three-dimensional data, so it cannot be directly Non-local mean filtering is applied to hyperspectral data; in addition, in non-local mean filtering, the Euclidean distance is used to calculate the similarity of two pixels in the field, and it is used as a weight for filtering, because the hyperspectral image The features in the data are mainly reflected by the third-dimensional spectral channel, that is, the pixels in the hyperspectral image are represented by vectors. At this time, the Euclidean distance cannot fully reflect the similarity between the two pixels, resulting in hyperspectral image The classification accuracy of the data is reduced

Images