基于迁移学习和多尺度融合的遥感影像场景分类方法研究

Research on Remote Sensing Image Scene Classification Method Based on Transfer Learning and Multi-scale Fusion

随着计算机运算能力的提升以及深度学习技术的发展,无须人工参与的深度学习方法已成为遥感影像分类的主流方法。因此,提出一种基于深度学习并嵌入注意力机制和融合多尺度特征的神经网络对遥感影像进行场景分类。该模型使用迁移学习减少训练样本不足带来的负面影响;在网络中嵌入注意机制、融合多尺度特征来提高对小尺寸地物目标分类的能力,并验证了模型的有效性。通过实验分析得出所提模型对遥感影像场景分类是可行且有效的。

With the improvement of computer computing power and the development of Deep Learning technology,Deep Learning methods that do not require human intervention have become the mainstream method for remote sensing image classification.Therefore,this paper proposes a neural network based on Deep Learning,embedding Attention Mechanism and blending multi-scale features for scene classification of remote sensing images.This model uses Transfer Learning to reduce the negative impact from insufficient training samples.It embeds Attention Mechanisms and blends multi-scale features in the network to improve the ability to classify small-sized terrain targets,and verifying the effectiveness of the model.Through experimental analysis,it is concluded that the proposed model is feasible and effective for remote sensing image scene classification.