A Novel Deep Feature Fusion Network For Remote Sensing Scene Classification

Abstract

In this paper, we analyze the performance of a novel deep feature fusion network when applied to the problem of remote sensing scene classification. So far, many classical convolutional neural network models have shown remarkable performance in image classification. With the availability of the latest high resolution remote sensing image data, traditions CNN models‘ performance has been considerably reduced. In order to tackle this condition, the deep convolutional neural networks are used in recent studies. Their classification accuracy depends on the depth of the network, while a deeper network will bring about higher computational complexity. In this work, we employ a deep feature fusion model for remote sensing scene classification, which uses the features extracted from Deep ResNet50 and VGG16 which are pre-trained and fine-tuned. The analysis of the experimental data prove the feasibility of the feature fusion network in remote sensing scene classification.