基于卷积神经网络的高光谱图像重建方法

Hyperspectral image reconstruction based on convolutional neural network

为了提高高光谱图像的成像质量,满足应用需求,提出基于卷积神经网络的高光谱图像重建方法研究。通过预处理方式消除高光谱图像中的暗电流信息与噪声数据,以此为基础,获取高光谱图像SIFT特征,选择高光谱图像最优波段,最大限度地保留高光谱图像的波段特征,有效融合卷积神经网络构建高光谱图像重建模型,将最优波段高光谱图像数据输入至构建模型中,输出结果即为高光谱图像重建结果。实验数据显示:应用提出方法获得的评价指标MSE最小值为6,评价指标Q最大值为6.8,PSNR与SSIM最大值分别为40.349 dB与0.964 4,充分证实提出方法高光谱图像重建质量较佳。

In order to improve the imaging quality of hyperspectral images and meet the application requirements,a hyperspectral image reconstruction method based on convolutional neural network is proposed.The dark current in-formation and noise data in the hyperspectral image are eliminated by preprocessing.Based on this,the SIFT features of the hyperspectral image are obtained,the optimal band of the hyperspectral image is selected,the band features of the hyperspectral image are retained to the maximum extent,the convolution neural network is effectively fused to con-struct the hyperspectral image reconstruction model,and the hyperspectral image data of the optimal band is input into the construction model,The output result is the hyperspectral image reconstruction result.The experimental data show that the minimum value of MSE,the maximum value of the evaluation index Q,the maximum value of PSNR and SSIM obtained by the proposed method are 6,6.8,40.349 dB and 0.9644 respectively,which fully proves that the quality of the hyperspectral image reconstruction by the proposed method is better.