Image classification method based on multi-scale dense convolutional neural network and spectral attention mechanism

Abstract

The invention relates to an image classification method based on a multi-scale dense convolutional neural network and a spectral attention mechanism. By using a dense connection mechanism to construct a multi-scale dense convolutional neural network, the dense connection mechanism can effectively alleviate the problem of gradient disappearance and strengthen Feature propagation encourages feature reuse and greatly reduces the number of parameters, reducing the demand for training samples in the network training process; in addition, the network is combined with the spectral attention mechanism to make more effective use of features in the spectral direction. The invention realizes autonomous extraction of depth features of hyperspectral images and high-precision classification under the condition of small samples. Compared with the existing hyperspectral image classification method based on deep learning, the present invention has smaller sample requirements and higher precision.

Description

technical field [0001] The invention relates to a small-sample hyperspectral image classification method based on a multi-scale dense convolutional neural network and a spectral attention mechanism, which belongs to the field of image processing. Background technique [0002] Hyperspectral remote sensing images have high spectral resolution, multiple imaging bands, and large amounts of information, and are widely used in remote sensing applications. Hyperspectral image classification technology is a very important content in hyperspectral image processing technology. It mainly includes two parts: feature extraction and classification. The classification features extracted from the original hyperspectral image have a great influence on the subsequent classification accuracy: the robustness of classification features Strong robustness can greatly improve the classification accuracy; on the contrary, classification features with poor robustness will significantly reduce the cla...

AI Technical Summary

Problems solved by technology

[0004] However, the existing methods for extracting spatial-spectral features of hyperspectral images using deep models are very complicated. It is often necessary to compress and reduce the dimension of the original hyperspectral image in the spectral space, and then extract the spatial and spectral features of the reduced data. Combined to get spatial spectral features

Dimensionality reduction is computationally intensive, and certain spectral information is lost, affecting accuracy

In addition, the existing methods for extracting spatial spectral features of hyperspectral images using deep models require a large number of training samples to train the network, but it is very difficult to mark the collected hyperspectral images in practice. plenty of time

Images