小样本的多模态遥感影像高层特征融合分类

Multimodal Remote Sensing Image Classification with Small Sample Size Based on High-Level Feature Fusion

在使用深度学习模型研究遥感影像地物分类问题时,某些地物的遥感影像可用于训练的样本很少。同时,多样化的遥感影像获取方式产生了大量不同空间分辨率的多模态遥感影像。融合这些多模态遥感影像,弥补样本量少导致分类精度低的缺陷,是小样本的遥感影像高精度分类领域中亟待解决的问题。针对上述问题,提出了考虑两种空间分辨率遥感影像相关关系的融合分类方法。首先,使用两个并行的深度学习网络分别提取两种空间分辨率影像的高层特征;其次,将提取到的高层特征通过融合方法进行融合;最后,得到融合后的高层特征作为输入,训练整个融合分类模型。实验表明,不同融合策略的分类精度不同,本文提出的基于高层特征级别的融合策略可以有效提高分类精度。

The training sample size for some objects on the ground is quite small when applying a deep learning model to study the classification of remote sensing images.Meanwhile,diversified remote sensing image acquisition methods generate numerous multimodal remote sensing images with different spatial resolutions.Fusing these multi-modal remote sensing images to remedy the small sample size defect and achieve a highly precise classification of remote sensing images is an urgent problem to be solved.To this end,the present study proposes a fusion method for image classification based on the correlation of two spatial resolutions.A deep learning network is utilized to extract the high-level features of the remote sensing images in two spatial resolutions.Two types of high-level features are integrated via the proposed fusion strategy and further used as the input to train the whole network model.The experimental results demonstrate that the proposed fusion algorithm can achieve high classification accuracy.Further,because different fusion rules have different classification accuracies,a suitable selection can improve the classification accuracy.