Automatic Building Change Detection on Aerial Images using Convolutional Neural Networks and Handcrafted Features

Diego Alonso Javier Quispe,Jose Sulla-Torres

doi:10.14569/ijacsa.2020.0110683

Diego Alonso Javier Quispe, Jose Sulla-Torres

Open Access

https://doi.org/10.14569/ijacsa.2020.0110683

Copy DOI

Abstract

In this article, we present a new framework to solve the task of building change detection, making use of a convolutional neural network (CNN) for the building detection step, and a set of handcrafted features extraction for the change detection. The buildings are extracted using the method called Mask R-CNN which is a neural network used for object- based instance segmentation and has been tested in different case studies to segment different types of objects obtaining good results. The buildings are detected in bitemporal images, where three different comparison metrics MSE, PSNR and SSIM are used to differentiate if there are changes in buildings, we used this metrics in the Hue, Saturation and Brightness representation of the image. Finally the characteristics are classified by two algorithms, Support Vector Machine and Random Forest, so that both results can be compared. The experiments were performed in a large dataset called WHU building dataset, which contains very high-resolution (VHR) aerial images. The results obtained are comparable to those of the state of the art.

Highlights

Building detection and change detection is a field that has been studied for a long time and attempts to solve different problems such as urban planning, cadastral updating, damage detection by natural disasters, among many others
In this article we propose a new framework for detecting changes in buildings using very high resolution aerial images (VHR)
The third section was used as a test to building detection and all this section was used for the building change detection step

Summary

Introduction

Building detection and change detection is a field that has been studied for a long time and attempts to solve different problems such as urban planning, cadastral updating, damage detection by natural disasters, among many others. Different researchers use different types of data to carry out their experiments. In their most basic form, aerial images or RGB satellite images are used, but these images have different drawbacks, starting with low resolution, perspective, lighting changes, shadows, and various variations that a building can have depends on the country, the city and the area where the images were captured. Other authors use different sensors to obtain more information such as multispectral sensors, synthetic aperture radar (SAR), light detection and ranging (LiDAR), digital surface models (DSM) and so on [1]. Using DSM allows us to obtain a 3D model of buildings, the advantage is that having altitude information allows us to better analyze changes in buildings, the drawback is that it is difficult to obtain such information

Methods

Results

Conclusion