VW-SAE:一种改进的光谱数据特征表示方法

VW-SAE:an Improved Method of Spectral Data Representation

针对高光谱图像维度高、目标特征提取不准确的问题,提出了一种可变加权堆叠式自编码器(variable-wise weighted stacked autoencoder,VW-SAE)的光谱数据特征表示方法。VW-SAE方法在堆叠式自编码器(SAE)的基础上,从每个AE的输入层中识别出重要的变量,通过对输出变量的相关性分析,将输出信息映射在AE目标函数的不同变量,引入不同的权值进行训练,逐层提取获得与输出相关的特征,并将其堆叠形成深网络。通过对每层网络权重的调控,在降低光谱数据维度的过程中,更好地提取光谱数据中的特征信息,进而提高了预测模型的精度。为验证方法的有效性,使用已采集的10248张水稻图像,在堆叠式自编码器结合全连接神经网络(SAE-FNN)的基础上,搭建了VW-SAE-FNN模型对水稻氮元素进行检测,实验结果表明该方法与SAE方法相比准确率明显提升。

Aiming at the problems of the high dimensionality of hyperspectral images and inaccurate target feature extraction,a variable-Wise Weighted Stacked Autoencoder(VW-SAE)spectral data feature representation method was proposed.Based on the stacked autoencoder(SAE),the VW-SAE method identified important variables from the input layer of each AE,and maped the output information in the AE objective function through the correlation analysis of the output variables.Different variables introduced different weights for training,extracted the features related to the output layer by layer,and stacked them to form a deep network.By adjusting the weight of each layer of the network,in the process of reducing the dimensionality of the spectral data,the feature information in the spectral data was better extracted at the same time,thereby improving the accuracy of the prediction model.To verify the effectiveness of the method,using more than 10248 rice images that had been collected was used.A VW-SAE-FNN model was built based on the stacked autoencoder combined with the fully connected gods network(SAE-FNN).The test results showed that the accuracy of this method was significantly improved compared to the SAE method.