Small sample hyperspectral classification method based on data enhancement

Abstract

The invention discloses a small sample hyperspectral classification method based on data enhancement. The method comprises the following steps: inputting hyperspectral image data to obtain a sample set; dividing N neighborhood regions with different sizes for each sample point in the sample set; enabling each sample point to obtain N neighborhood sample sets, processing the neighborhood sample sets to obtain corresponding newly-added sample points, combining all the corresponding newly-added sample points into an amplified data set of the sample points, and traversing the sample sets to obtain enhanced image data; using the original hyperspectral image data and the enhanced image data for training a classifier; and carrying out classification identification on the to-be-identified sample points enhanced by using the data in the above steps in the hyperspectral image by using a voting method or an optimization method by using the trained classifier. According to the invention, the data enhancement of the hyperspectral image data is realized, the problem of small samples is solved to a certain extent, a better classifier is trained through the amplified training samples, and the classification recognition rate of the hyperspectral data is obviously improved under similar conditions.

Description

technical field [0001] The invention belongs to the field of small-sample hyperspectral classification and identification, and relates to a small-sample hyperspectral classification method based on data enhancement. Background technique [0002] Hyperspectral images combine image information and spectral information of samples. Image information can reflect the external quality characteristics of the sample such as size, shape, defect, etc. Since different components have different spectral absorption, the image will reflect a certain defect more significantly at a specific wavelength, and the spectral information can fully reflect the quality of the sample. Differences in internal physical structure and chemical composition. It is often difficult to collect a large number of training samples from hyperspectral remote sensing images, and it is difficult to obtain high-precision classification accuracy with only a small number of training samples. Therefore, data enhancement...

AI Technical Summary

Problems solved by technology

However, unlike image data and text data, people's understanding of hyperspectral data is not perfect, and the enhancement methods used for image data and text data are not fully applicable to hyperspectral data.

In the early days, some scholars imitated the practice of the image field, and performed operations such as random perturbation, noise addition, and rotation on the hyperspectral spectrum to obtain enhanced samples. However, the enhanced data only achieved a small improvement in the classification accuracy, which can be improved by about 2 to 3 %

Images