Deep Fusion Feature Based Object Detection Method for High Resolution Optical Remote Sensing Images

Eric Wang,Zhe Nie,Zuodong Liang,Xun Zhang,Yueping Li,Siu Yiu,Juntao Yu

doi:10.3390/app9061130

Abstract

With the rapid growth of high-resolution remote sensing image-based applications, one of the fundamental problems in managing the increasing number of remote sensing images is automatic object detection. In this paper, we present a fusion feature-based deep learning approach to detect objects in high-resolution remote sensing images. It employs fine-tuning from ImageNet as a pre-training model to address the challenge of it lacking a large amount of training datasets in remote sensing. Besides, we improve the binarized normed gradients algorithm by multiple weak feature scoring models for candidate window selection and design a deep fusion feature extraction method with the context feature and object feature. Experiments are performed on different sizes of high-resolution optical remote sensing images. The results show that our model is better than regular models, and the average detection accuracy is 8.86% higher than objNet.

Highlights

Object detection for remote sensing images is an important research field
The experiments were divided into two parts: (1) we briefly analyzed the applicability of the selective search algorithm for candidate region proposal in remote sensing images; (2) we evaluated the result of the binarized normed gradients (BING) algorithm and pBING algorithm
We present a deep fusion feature approach to detect objects in high-resolution remote sensing images

Summary

Introduction

Object detection for remote sensing images is an important research field. The applications of object detection in remote sensing images are more and more popular, such as city planning and environmental exploration. Object detection for remote sensing images is a more difficult job since remote sensing images are quite different from regular images. One high-resolution optical remote sensing image contains more objects with more shapes and texture information than a regular image, and the objects may be scattered in the whole image. The object to be detected is relatively small and close to the background. If we zoom out from a remote sensing image to a small size for a global view, we would lose many details, and the objects may almost be invisible. Object detection for remote sensing images is harder work than for regular images to some extent

Methods

Results

Conclusion