Hyperspectral image classification method based on deep learning and multi-instance learning

Abstract

The invention discloses a hyperspectral image classification method based on deep learning and multi-instance learning, and the method comprises the following steps: S1, carrying out the preprocessingof a hyperspectral image, and deleting a part of redundant wavebands; S2, performing feature extraction on the preprocessed hyperspectral image obtained in the step S1 to obtain example features; S3,mapping the example features obtained in the step S2 into example features of an image packet; and S4, classifying the example features of the image packet obtained in the step S3. The feature extraction efficiency and the classification precision can be improved.

Description

technical field [0001] The invention relates to a hyperspectral image classification method, in particular to a hyperspectral image classification method based on deep learning and multi-instance learning. Background technique [0002] Generally speaking, the hyperspectral data classification process includes two processes of feature extraction and classification. The features obtained after feature extraction are more accurate in describing the characteristics of ground objects, and the measurement is more detailed, which plays a vital role in the performance of the classifier. [0003] At present, the commonly used supervised classification methods include minimum distance classification, support vector machines and neural networks. For a long period of time, SVMs have dominated the processing of hyperspectral image classification tasks. Joachims et al. proposed the SVM method based on transductive reasoning. During the training process, the hyperplane of the SVM and the...

AI Technical Summary

Problems solved by technology

In the existing application of multi-instance learning, most of them are three-channel algorithms in medicine. This algorithm has fewer channels and data calculations, so the conversion efficiency is acceptable, and the effect of using SVM classification is also acceptable. However, for hyperspectral images For data of up to hundreds of channels, the simple packet conversion function cannot achieve the original effect, and the selection of the classifier also has new requirements

The feature extraction efficiency and classification accuracy of hyperspectral image classification methods in the prior art are not enough

Images