Classification method for hyperspectral remote sensing image based on full convolutional network

Abstract

The invention discloses a classification method for a hyperspectral remote sensing image based on a full convolutional network. The method comprises the following three steps of data preprocessing, feature extraction and classification; the hyperspectral remote sensing image is input into the full convolutional network to be processed, the characteristics of the full convolutional network are usedfor extracting and classifying the features of the hyperspectral remote sensing image. firstly, the hyperspectral remote sensing data is lowered to low-dimensional space, then, a low-dimensional hyperspectral image is input into the full convolutional network model to be processed, and a convolutional layer obtained in a processing process is taken as features; then, the obtained feature image isclassified into a training set and a testing set, and a sparse representation dictionary is trained; and finally, the testing set is subjected to sparse reconstruction, and a classification result isobtained. The spectrum and the spatial features of the hyperspectral data can be fully combined, the advantage that the full convolutional network does not restrict input sizes is used to fully extract feature information from the hyperspectral remote sensing image data of different sizes and different sample distribution situations, and an accurate classification result is obtained.

Description

technical field [0001] The invention relates to the technical field of hyperspectral remote sensing image processing, in particular to the classification of hyperspectral remote sensing images, in particular, a method for classifying hyperspectral remote sensing images based on a fully convolutional network. Background technique [0002] Remote sensing image refers to the film that records the size of electromagnetic waves of various surface objects, and is a form of expression of remote sensing information obtained through remote sensing detection. Since the absorption characteristics of many surface substances are only in the spectral resolution range of 20-40nm, hyperspectral images can identify substances that cannot be detected in broadband remote sensing. [0003] Compared with traditional visible light and multispectral data, the improvement of spectral resolution and spectral range of hyperspectral remote sensing data has enhanced the spectral detection ability of gr...

AI Technical Summary

Problems solved by technology

However, a single spectral information or spatial information cannot fully represent the characteristics of hyperspectral remote sensing images.

Images