摘要
针对传统遥感图像分类方法难以取得更高精度的问题,提出一种根据深度学习思想的基于栈式去噪自编码器的遥感图象分类方法。首先,将多个去噪自编码器栈式叠加构成深度网络模型,用无监督的layer-wise方法由下至上训练每一层网络并在训练数据中加入噪声以得到更为稳健的特征表达;然后,通过反向传播(BP)神经网络对特征进行有监督学习并利用误差反向传播对整个网络参数进行进一步优化得到最终的模型;最后,利用国产高分一号遥感数据进行实验验证。基于栈式去噪自编码器的遥感图像分类方法的总体分类精度和kappa精度分别达到95.7%和95.5%,均高于传统的支持向量机(SVM)和BP神经网络的分类精度。实验结果表明,所提出的方法能有效提高遥感图像的分类精度。
Focusing on the issue that conventional remote sensing image classification methods can＇ t reach better effect, a new remote sensing image classification method based on Stacked Denoising AutoEncoder （SDAE） inspired by deep learning was pmpesed. Firstly, the deep network model was built through the stacked layers of Denoising AutoEncoder （DAE）, then the unsupervised Greedy layer-wise training algorithm was used to train each layer in turn with noised input for more robust expression, characteristics were learnt supervised by Back Propagation （BP） neural network and the whole net was optimized by using error back propagation. Finally, GF-1 remote sensing data were used for evaluation and the total accuracy and kappa accuracy which were higher than those of Support Vector Machine （SVM） and BP neural network reached 95.7% and 95.5% respectively. The experiment results show that the proposed method can effectively improve the accuracy of remote sensing image classification.
出处
《计算机应用》
CSCD
北大核心
2016年第A02期171-174,188,共5页
journal of Computer Applications
关键词
深度学习
栈式去噪自编码器
反向传播神经网络
遥感图像
地物分类
deep learning
Stacked Denoising AutoEncoder （SDAE）
Back Propagation （BP） neural network
remote sensing image
land cover classification
