Unsupervised hyperspectral image classification method based on maximum-minimum distance embedding

Abstract

The invention discloses an unsupervised hyperspectral image classification method based on maximum-minimum distance embedding, and the method comprises the following specific steps: S1, employing multi-scale spatial features to replace original features, and obtaining a multi-scale feature vector; S2, inputting the multi-scale feature vector into a maximum-minimum distance embedded depth auto-encoder model, and performing feature extraction and dimension reduction; S3, obtaining an initial clustering result through K-means clustering; S4, optimizing the initial clustering result by utilizing guided filtering. According to the method, spatial context information is fully utilized, and multi-scale spatial features are adopted so as to enhance the discrimination of data. In order to overcomethe problem of high dimension in clustering, a maximum-minimum distance embedded depth auto-encoder is adopted to realize feature representation and dimension reduction processing, the clustering timeis reduced, a kmeans clustering result is optimized by adopting guided filtering, and the classification performance is further improved.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image classification, in particular to an unsupervised hyperspectral image classification method with maximum-minimum distance embedding. Background technique [0002] The rapid development of remote sensing technology makes it easier to collect hyperspectral images. Hyperspectral images contain hundreds of continuous bands, which can more accurately express the characteristics of substances. Therefore, hyperspectral images are widely used in various fields, such as surface observation, gas leakage, etc. Hyperspectral image classification, as a key issue in the application of hyperspectral images, has received extensive attention and research. [0003] At present, hyperspectral image classification is mainly based on supervised methods, such as SVM, KNN, deep learning, etc. In the early days, it mainly focused on pixel-level classification, and the classification results were obtained by ...

AI Technical Summary

Problems solved by technology

Hyperspectral images not only have a large amount of data, but also have high dimensions, and the difference between categories is not obvious, which brings challenges to the application of clustering algorithms

Images