Hyperspectral image classification method based on attention mechanism and generative adversarial network

Abstract

The invention provides a hyperspectral image classification method based on an attention mechanism and a generative adversarial network, and relates to the technical field of hyperspectral image processing and application. The method comprises the following steps: performing normalization and principal component analysis on a hyperspectral image data set to obtain a real sample; after transpose convolution operation is carried out on the feature matrix, inputting generated features into a combined space-spectrum combined attention mechanism module, and obtaining a generated sample; performing four times of convolution on the real sample and the generated sample to obtain a hierarchical feature of the input sample; adding the real sample features and corresponding equal-size features in the generator to generate new fusion features, and inputting the new fusion features to a combined space-spectrum combined attention mechanism module; and inputting the hierarchical features sequentially into ConvLSTM along a spectrum channel, and realizing classification through a softmax function in a recognizer. According to the invention, the quality of the generated sample can be effectively improved, and the classification capability of a discriminator is improved by using the generated high-quality sample.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image processing and application, in particular to a hyperspectral image classification method based on an attention mechanism and a generative confrontation network. Background technique [0002] Acquisition and collection of hyperspectral data has become more accessible and inexpensive over the past few decades. A hyperspectral image (HSI) is a three-dimensional (3D) data cube with hundreds of spectral bands per pixel, and each spectral band corresponds to a 2D image. HSI contains rich spectral information and spatial information, and HSI processing has been widely used in many practical applications, such as military affairs, agriculture and astronomy. The basis of these applications is HSI classification, which is achieved by assigning a specific class to each pixel. It mainly includes two tasks: the representation of effective features and the design of high-level classifiers. [000...

AI Technical Summary

Problems solved by technology

The bootstrap information provided by the discriminator is limited, and the generator does not have direct access to the true sample distribution

Therefore, it is difficult to ensure that the generator is always updated according to the true sample distribution

When it comes to HSI data with high-dimensional features, the generated samples are more difficult to approximate the real samples with complex spatial-spectral distribution, which may further degrade the classification performance of the discriminator

Images