[Home ] [Archive]   [ فارسی ]  
:: Main :: About :: Current Issue :: Archive :: Search :: Submit :: Contact ::
:: Volume 11, Issue 4 (6-2022) ::
JGST 2022, 11(4): 67-82 Back to browse issues page
Utilization of a deeply Refined Deep Residual Convolutional Neural Network to evaluate and compare the accuracy of Road detection from Sentinel 1 radar images (Case study: Tehran and Shiraz metropolises)
P. Zeaiean firouzabadi, S. H. Sheikhghaderi *, M. Kelarestaghi
Abstract:   (335 Views)
In recent years, Road detection and road extraction from satellite images with the advancement and development of deep learning algorithms in the field of semantic segmentation has received more and more attention of researchers. In this regard, most of the studies have been done in the field of Road detection and road extraction using optical images and in these studies, few studies have been performed using radar images worldwide. Therefore, the aim of this study was to use a deeply Refined Deep Residual Convolutional Neural Network (RDRCNN) to evaluate and compare the accuracy of road extraction from Sentinel 1 radar images in Tehran and Shiraz metropolitan areas in equal conditions in terms of number of educational samples, validation and architecture. It is the same. In this study, to extract the road using DNN, the VV-VH color combination of Sentinel 1 radar images from 8 different cities (Tehran, Mashhad, Isfahan, Shiraz, Tabriz, Urmia, Baghdad and Beijing) was used. Finally, the RDRCNN model with a residual connected unit (RCU) and a dilated perception unit (DPU) was used for road training and extraction. The research findings indicate that the RDRCNN model has performed almost the same in the process of identifying and extracting roads in the two cities of Tehran and Shiraz, and in general, the above model has performed slightly better in the city of Shiraz. In terms of accuracy evaluation metrics, for Tehran images, the criteria were Recall 57.66%, accuracy 51.29%, F1 score 54.43% and overall accuracy 92.78%, and for Shiraz images Recall criteria 60.77%, accuracy 54.71%, F1 score 57.40% and overall accuracy of 95.63% were obtained. The findings of this study show the low accuracy of road training and extraction from Sentinel 1 radar images for two metropolitan areas of Iran. In general, by comparing the results of this study with previous studies, it can be seen that one of the most important reasons for the low accuracy of the results is the low width of roads in Iranian cities; However, due to the lack of necessary studies in the field of road extraction with Sentinel 1 radar images, it is not possible to comment definitively on the results and it is suggested that more studies be done in this field.
Article number: 6
Keywords: Deep Learning, RDRCNN, Sentinel 1, Road Extraction, Tehran, Shiraz.
Full-Text [PDF 2076 kb]   (213 Downloads)    
Type of Study: Research | Subject: Photo&RS
Send email to the article author

Add your comments about this article
Your username or Email:

CAPTCHA


XML   Persian Abstract   Print


Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:

Zeaiean firouzabadi P, Sheikhghaderi S H, Kelarestaghi M. Utilization of a deeply Refined Deep Residual Convolutional Neural Network to evaluate and compare the accuracy of Road detection from Sentinel 1 radar images (Case study: Tehran and Shiraz metropolises). JGST 2022; 11 (4) :67-82
URL: http://jgst.issge.ir/article-1-1062-en.html


Rights and permissions
Creative Commons License This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Volume 11, Issue 4 (6-2022) Back to browse issues page
نشریه علمی علوم و فنون نقشه برداری Journal of Geomatics Science and Technology