Hyperspectral image semi-supervised classification method based on small sample learning

Abstract

The invention discloses a hyperspectral image semi-supervised classification method based on small sample learning. The method mainly solves the problem that in the prior art, in a small sample scene, a classification network is prone to overfitting, and unreliable pseudo-mark samples in the self-training process cause adverse effects on the training process of the classification network in the self-training process. According to the implementation scheme, the method comprises the following steps of: 1) obtaining five hyperspectral data sets from a hyperspectral database, and preprocessing the five hyperspectral data sets; 2) collecting a training set and a test set from the preprocessed data set; 3) constructing a hyperspectral image prototype classification network comprising two convolutional layers and a full connection layer; 4) using the training set to iteratively update each category prototype to complete the training of the classification network; and 5) inputting the test set into the trained network optimal model to obtain a test data classification result. The method solves the problem of overfitting in the existing small sample scene, improves the classification precision, and can be applied to geological exploration, urban remote sensing and ocean exploration.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a hyperspectral image classification method, which can be used in geological exploration, urban remote sensing and ocean detection. Background technique [0002] Hyperspectral image classification is the focus of research in the field of image processing. Hyperspectral images have the characteristics of large amount of data, many bands, and strong correlation between bands. Although these characteristics bring a lot of convenience to the classification process, due to the small number of labeled samples, the model is easy to overfit, making hyperspectral images Many challenges are faced in practical classification and recognition applications. [0003] The existing hyperspectral image classification methods are divided into unsupervised, semi-supervised and supervised classification methods according to whether unlabeled samples participate in training. Semi-supe...

AI Technical Summary

Problems solved by technology

[0012] The purpose of the present invention is to address the deficiencies of the above-mentioned existing self-training methods, to provide a hyperspectral image semi-supervised classification method based on small sample learning, so as to reduce the "false label" sample pair model with low confidence in the training process The impact of the model makes the model better represent the category distribution of the data, alleviates the over-fitting problem that is prone to occur in small sample scenarios, and improves the classification performance of the network

Images