Hyperspectral image classification method based on multi-scale spatial-spectral feature joint learning

Abstract

The invention provides a hyperspectral image classification method based on multi-scale spatial-spectral feature joint learning, and the method comprises the steps: effectively extracting spectral features and spatial features of different scales through a multi-scale spectral feature extraction module and a multi-scale spatial feature extraction module respectively; the spectral-spatial feature fusion module is used for performing combined extraction of spectral features and spatial features, so that combined learning of the spectral-spatial features is realized, rich spectral and spatial information in the hyperspectral image is fully utilized, and the classification precision is improved; meanwhile, spectrum-space feature extraction is achieved through three sets of convolutional neural networks, the first two-dimensional size of a convolution kernel in a first set of three-dimensional convolutional layers I is 1 * 1, the second set of three-dimensional convolutional layers is two-dimensional convolutional layers, and compared with a multi-layer three-dimensional convolutional neural network of which the three dimensions are all not 1, the extraction efficiency is improved. On the premise of ensuring the classification performance, model lightweight can be realized, and the training speed during model acquisition and the classification speed during model use are accelerated.

Description

technical field

[0001] The invention belongs to the field of remote sensing image processing, in particular to a hyperspectral image classification method based on joint learning of multi-scale spatial spectral features. Background technique

[0002] Hyperspectral images are three-dimensional image data composed of hundreds of spectral bands, containing rich spectral and spatial information, and have been widely used in urban development, geological exploration, environmental supervision, precision agriculture and other fields. The purpose of hyperspectral image classification is to classify each pixel in the image into a specific object category according to the obtained sample features, and generate a corresponding classification result map according to the different object categories.

[0003] With the rapid development of deep learning theory in recent years, deep learning methods based on convolutional neural networks (CNN) have been widely used in the field of hyperspe...

Images