Content Based Image Retrieval of Remote Sensing Images Based on Deep Features

Abstract

This paper presents the results of applying deep features to the problem of content based image retieval of remote sensing images. Extraction of deep features from the last layers of a trained convolutional neural network from deep learning approaches demonstrates a higher performance than feature extraction using shallow methods. In this paper we used deep features obtained from a fine tuned convolutional neural network and we also focused on experiments of dimension reduction methods of these deep features. We test these methods using UCM Merced and RSSCN7 datasets. Despite their shorter length deep features obtained as a result of dimension reduction methods, are shown to achieve higher performance of content-based retrieval.