Small-sample hyperspectral image classification method based on pseudo-label learning

Abstract

The invention provides a small sample hyperspectral image classification method based on pseudo label learning. The method comprises the following steps: sampling surrounding pixels by taking a hyperspectral pixel point as a center to generate a hyperspectral sample, constructing a small sample data set by using a small number of labeled samples, and allocating soft-pseudo labels to unlabeled samples by using the small sample data set so as to obtain an auxiliary data set; then, jointly training a two-branch deep neural network composed of a shared feature extractor and two different classifiers by using the small sample data set and the auxiliary data set; finally, using the trained network to predict labels for the test data, and realizing classification processing. According to the method, by effectively utilizing the potential information of the unmarked samples, the classification accuracy of the hyperspectral images of the small samples can be improved.

Description

technical field [0001] The invention belongs to the technical field of hyperspectral image processing, and in particular relates to a small-sample hyperspectral image classification method based on pseudo-label learning. Background technique [0002] Hyperspectral images can be represented as 3D data cubes, which contain rich spatial-spectral information, so they are widely used in remote sensing fields, such as environmental monitoring, military surveillance, etc. Hyperspectral image classification is one of the important tasks of hyperspectral image analysis, which aims to assign a predefined class label to each pixel. Many traditional machine learning methods have been applied to hyperspectral image classification, but its inherent shallow structure is difficult to efficiently learn discriminative features, so its classification performance needs to be improved. Different from traditional machine learning methods, deep neural networks have powerful feature representation...

AI Technical Summary

Problems solved by technology

However, the success of deep neural networks depends on a large number of labeled samples, which is difficult to obtain in practical applications, so it is necessary to use limited labeled samples to train high-precision deep neural network models

Training with only a small number of labeled samples can lead to overfitting of the model and the inability to generalize effectively on the test data

Images