Investigation and Extraction of Building Demolitions due to Earthquake using High Resolution Satellite Images

Document Type : Applied Article


1 Master of Remote Sensing and Geographic Information System, Faculty of Geography, University of Tehran

2 Assistant Professor, Faculty of Geography, University of Tehran

3 Associate Professor, Faculty of Geography, University of Tehran


Earthquake is one of the natural disasters that if occurs strongly in high population areas, will create great human catastrophe. Earthquake can provide considerable life and financial devastating effects, especially in urban regions. Observation of damaged buildings map is crucial for crisis management experts and helps them guide rescue teams to damaged locations in short period of time. Remote sensing and geographic information system is an efficient tool of rapid survey of condition of damaged buildings after the earthquake in urban regions. This research has been conducted with the aim of identification of demolished buildings due to earthquake by very high resolution satellite images and comparison of available efficient methods. To achieve these goals, very high resolution satellite images of Bam city, before and after the earthquake, and the observed damage map of the region were used. In this study, the best and the most appropriate textural indices were chosen after calculation of textural features of images by statistical analysis of logistic regression and correlation. Then, the condition of buildings demolition was classified by optimum obtained textural values and implementing Multilayer Perceptron (MLP) neural network systems, Adaptive Neuro-Fuzzy Inference System (ANFIS), and Support Vector Machines (SVM). Finally, the accuracy of all the presented techniques were compared with each other and the best proposed technique was selected and presented. According to the results, all the three MLP, SVM and ANFIS methods were good for classification of degrees of buildings demolition, but ANFIS method was better with 1% in overall accuracy, 4% in kappa coefficient, and 1.7% in RMSE.
Occurrence of natural disasters, especially in urban regions, causes abundant life and financial damages. Earthquake is one of the natural disasters that if occurs strongly in high population areas, will create great human catastrophe. The rescue of people from under debris and damaged regions after the earthquake leads to a reduction in life losses, but this subject can have the maximum efficiency only when the fast rescue operations have plan and goal. This is one of the most important concerns of crisis management managers in every country. Buildings are among regions with the highest destruction by earthquake.
To obtain buildings demolition map, one can perform through ground operations and identification groups with high accuracy, but this needs plenty of time and requirements. Recent advances about satellites from spatial, spectral, and temporal resolution point of view and even advance in image processing areas also has provided the observation possibility of changes in target regions through image analyses. Remote sensing (RS) and Geographic Information System (GIS) is an efficient tool for fast surveillance on damaged buildings in urban regions after the earthquake.
The goal of this research is identification and extraction of buildings demolition due to earthquake by investigation of textural features of terrains in image and comparison of building demolition classification techniques and presentation of the best technique considering the study area.
Methods and Materials
The case study in this research is Bam city, one of the cities of Kerman province located in Iran country. On Friday, January 05, 2004 at 5:26 (local time) a 6.6 Richter scale and a depth of 8.5 Kilometers earthquake occurred in Bam city and lasted for 12 seconds. QuickBird satellite images with 61 centimeters resolution were used in this research. Demolition map was also utilized as a helping data (ground) for determination of location of buildings and a basis for assessment of this study that Yamazaki et al. provided this map as intersections of location and condition of degrees of buildings demolition [16].
To perform more accurate calculations, initially a pre-processing step performs on raw data before processing on images. Considering the goal of this research, buildings are required to be extracted out of images before and after the earthquake. Then, the textural features of images were extracted and the best textural descriptors for determination and identification of buildings demolition were selected by statistical methods. To achieve demolition map, there is a need to models for classification and identification of degree of demolition. In this research, three Multilayer Perceptron (MLP) Neural Network, Adaptive Neuro-Fuzzy Inference System (ANFIS), and Support Vector Machines (SVM) techniques were employed and finally an accuracy evaluation and comparison of these techniques have been done.
Discussion and conclusion
After extraction of buildings from images and calculation of textural features, mean obtained numerical values for every constituent pixels of building has been calculated for results assessment and added to building descriptive table as a quantitative parameter and then all of the descriptors were normalized. In determination of the demolition due to earthquake by texture analysis, one can employ the assumption that the demolished areas have more irregular texture than the normal areas [8]. In this research, overall eight descriptors, including statistical first order and Haralick textural descriptors were implemented. Then, by logistic regression, the best texture was chosen that Variance, Dissimilarity, Homogeneity, and Contrast descriptors were used in identification and rehabilitation of demolished and normal buildings with the highest accuracy than the others. The models and techniques were also run by 206 educational samples and chosen textural images as the four input layers. The outputs are classification of degrees of buildings.
According to the results, by running the aforementioned models and comparison of overall accuracy, kappa coefficient, and RMSE, it has been determined that all the MLP, SVM, and ANFIS methods are similar for classification of degrees of buildings demolition, but have minor differences in accuracy, so that maybe by a look at demolition maps, the differences are not clear and there is a need to more precision. Totally, neuro-fuzzy method was better than the other two methods with 1% difference in overall accuracy, 4% in kappa coefficient and 1.7% in RMSE. However, the ANFIS method reached the first rank with minor difference. The superiority of this method can be interpreted because of combination of the neural network system and fuzzy logic concepts and simultaneously implementation of both of them.


[1].       باقری، میلاد؛ جلوخانی نیارکی، محمدرضا؛ و باقری، کیوان (1396). «بررسی پتانسیل اراضی استان کرمانشاه جهت کشت گندم دیم با استفاده از شبکۀ عصبی مصنوعی (کاربرد سنجش از دور و Gis در علوم منابع طبیعی)»، دانشگاه آزاد اسلامی واحد بوشهر، دورۀ 8، ، ش 4، ص 48-36.
[2].       جعفرآبادی، محمد؛ سلطانی، اکبر؛ و محمدی، سیده مؤمنه (1392). «سری آمار: همبستگی و رگرسیون»، مجلۀ دیابت و لیپید ایران، ش 6، ص 506-479.
[3].       شریفی، آرش؛ علیاری شوره‌دلی، مهدی؛ و تشنه‌لب، محمد (1389). «معرفی سیستم فازی شبه‌چندجمله‌ای تاکاگی-سوگنو-کانگ با کاربرد در شناسایی سیستم و کلاس‌بندی الگو». مجلۀ کنترل، دانشگاه خواجه نصیرالدین طوسی، ج4، ش 3، ص  28-15.
[4].       مختاری، محمدحسین؛ و نجفی، احمد (1394). «مقایسۀ روش‌های طبقه‌بندی ماشین بردار پشتیبان و شبکۀ عصبی مصنوعی در استخراج کاربری‌های اراضی از تصاویر ماهواره‌ای لندست TM»، مجلۀ علوم و فنون کشاورزی و منابع طبیعی، علوم آب و خاک، ج 19، ش 72، ص 45-35.
[5].              منهاج، محمدباقر (1397). مبانیشبکه‌هایعصبیمصنوعی، چ 12، تهران: انتشارات دانشگاه صنعتی امیرکبیر.
[6].       ویس، حیدر؛ و صمدزادگان، فرهاد (1392). «تعیین میزان تخریب ناشی از زلزله در ساختمان‌ها با استفاده از تصاویر ماهواره‌ای با قدرت تفکیک بالا»، علوم و فنون نقشهبرداری، سال 2، ش 2، ص 191-106.
[7]. Agency, F. E. M.; Security, U. D. O. H.; & America, U. S. O. (2001). Guide for All-Hazard Emergency Operations Planning.
[8]. Brunner, D.; Lemoine, G.; & Bruzzone, L. (2010). “Earthquake Damage Assessment of Buildings Using VHR Optical and SAR Imagery”, IEEE TRANSACTION ON GEOSCIENCE AND REMOTE SENSING.
[9]. Cotrufo, S.; Sandu, C.; Giulio Tonolo, F.; & Boccardo, P. (2018). “Building damage assessment scale tailored to remote sensing vertical imagery”, European Journal of Remote Sensing, 51(1), pp: 991-1005.
[10].            Dong, L.; & Shan, J. (2013). “A comprehensive review of earthquake-induced building damage detection with remote sensing techniques”, ISPRS Journal Of Photogrammetry and RemoteSensing, 84(0), 85-99. doi:
[13].            Menderes, A.; Erener, A.; & Sarp, G. (2015). “Automatic detection of damaged buildings after earthquake hazard by using remote sensing and information technologies”, Procedia Earth and Planetary Science, 15, pp: 257-262.
[14].            Mizutani, E.; Jang, J. S. R.; & Sun, C. T. (1997). Neuro-fuzzy and soft computing: a computational approach to learning and machine intelligence.
[15].            Yamazaki, F.; Miura, H.; & Matsuoka, M. (2007). “Identification of damaged areas due to the 2006 Central Java, Indonesia earthquake using satellite optical images”, Paper presented at the Urban Remote Sensing Joint Event.
[16].            Yamazaki, Fumio; & Kehiyama, Masayuki (2003). “Detection of Damage Due To The 2003 Bam, Iran Earthquake Using Terra- Aster imaging”, Workshop an Application of Remote Sensing Technologist for Disaster Response, University of California, Septemer 12Th,.