Deep Hash Assisted Network for Object Detection in Remote Sensing Images

Hongbin Ma,Wei Wang,Min Wang,Zepei Sun,Shuyuan Yang,Guangying Xu

doi:10.1109/access.2020.3024720

Abstract

Remote Sensing Images (RSIs) often have extremely wide width and abundant terrain. In order to achieve rapid object detection in large RSIs, in this paper, a Deep Hash Assisted Network (DHAN) is constructed by introducing a hashing encoding of images in a two-stage deep neural network. Different with the available detection networks, DHAN first locates candidate object regions and then transfers the learned features to another Region Proposal Network (RPN) for detection. On the one hand, it can avoid the calculations on the background irrelevant to objects. On the other hand, the built hash encoding layer in DHAN can accelerate the detection via binary hash features. Moreover, a self attention layer is designed and combined with the convolution layer, to distinguish relatively small objects regions from a very large scene. The proposed method is tested on several public data sets, and the comparison results show that DHAN can remarkably improve the detection efficiency on large RSIs and simultaneously achieve high detection accuracy.

Highlights

Due to the increasing spatial resolution of optical imaging devices, larger and larger Remote Sensing Images (RSIs) have been collected from spaceborne and airborne sensors
In this paper we propose a Deep Hash Assisted Network (DHAN) by introducing deep hashing encoding into Faster-Regions with Convolutional Neural Network (RCNN), aiming at improving the speed of objects detection in large RSIs
The proposed method is tested on several public data sets, and the results show that DHAN can remarkably improve the detection efficiency on large RSIs and simultaneously achieve high detection accuracy

Summary

Introduction

Due to the increasing spatial resolution of optical imaging devices, larger and larger Remote Sensing Images (RSIs) have been collected from spaceborne and airborne sensors. Locating and identifying the interested objects from these images rapidly, has been an important topic in the field of remote sensing data processing. Many Deep Neural Networks (DNNs) have been. Developed for objects detection, which can be categorized as one-stage and two-stage DNNs. One-stage DNNs regard the object detection as an end-to-end regression task, such as Regions with Convolutional Neural Network (RCNN) proposed by Ross Girshick et al [8]. Two-stage DNNs first make a candidate target regions proposal and perform the classification on them. One-stage DNNs are often faster than two-stage DNNs by combining the objects detection and classification together, such as YOLO [16], [17] and SSD [18]. The detection accuracies of one-stage DNNs are often lower than that of two-stage DNNs

Objectives

Methods

Results

Conclusion