A deep residual neural network for low altitude remote sensing image classification

In the recent years the applications of deep neural networks are increasing rapidly. There are two important factors determining the efficiency of training a computer vision system using deep neural networks. The first factor is the difficulty of training a very deep neural network with large number of parameters. The second factor is the efficiency of the trained network for decreasing the computational cost. In this paper an efficient deep neural network which uses the grid size reduction, factorization and hyper parameter tuning is proposed. In order to deal with large number of layers the residual units are used. A series of experimental simulations are performed on the application of the proposed deep neural network for classification of aerial images. The experimental results show that the proposed architecture has acceptable accuracy for aerial scene classification.

I. Introduction

In the last decade many advances in both deep learning and remote sensing fields has been achieved. These achievements created a variety of research opportunities, namely scene classification, object detection, surveillance and structural inspection. A significant body of research in remote sensing is focused on image descriptors. These descriptors use low-level feature extractors such as Scale Invariant Feature Transform (SIFT) or local binary patterns (LBP). In recent years many studies has been devoted to the same problem using high-level feature extractors like Convolutional Neural Networks (CNN) [1] [2]. There are number of problems in training a neural network and one of these problems is that as the depth of the neural network increases the time and computational cost for the training process is also increases. This is one of the limiting factors for using the deep neural networks in various applications. In order to solve this problem an identity mapping named residual unit is proposed [3].