基于快速去噪和深度信念网络的高光谱图像分类方法

A hyperspectral image classification method based on fast denoising and deep belief network

为了改善传统分类方法在高光谱遥感图像去噪和特征提取方面的不足,提出了一种基于快速去噪和深度信念网络的高光谱图像分类方法。该方法利用图像的二阶偏导数和梯度共同控制扩散速度,采用改进的自适应扩散系数对不同区域进行扩散,并利用深度信念网络对去噪后的图像进行地物分类。实验结果表明,与传统的分类方法相比,该方法提高了高光谱图像地物分类的精度。

In order to improve the shortcomings of the traditional classification methods in hyperspectral remote sensing image denoising and feature extraction,a hyperspectral image classification method based on fast denoising and deep belief network is proposed.The method uses the second partial derivative of the image and the gradient to control the diffusion velocity,and uses an improved adaptive diffusion coefficient of diffusion in different regions.Finally the depth of belief network is used for denoising images to classify these terrain features.The experimental results show that compared with the traditional classification methods,the proposed method obviously improves the classification accuracy of the remote sensing image.