Content uploaded by Mangalraj Poobalasubramanian
Author content
All content in this area was uploaded by Mangalraj Poobalasubramanian on Oct 20, 2021
Content may be subject to copyright.
Post-earthquake Building Damage Detection Using Deep Learning
... The used method could be suitable for building damage classification. Mangalraj et al. [14] conducted damage assessment with deep learning techniques using pre-and post-earthquake images to accurately identify the buildings damaged in the earthquake. Ding et al. [15] used high-resolution images obtained from UAVs in their studies due to the advantages of time, flexibility, and high resolution in post-earthquake data collection. ...
... Constraint (13) force that each drone can depart from only the depot that it belongs to for the first sortie. Constraint (14) force that each drone can depart from only the depot that it arrived at the previous sortie for second and next sorties. Constraint (15) force that any departure time must be greater than previous arrival time plus expected time for refuel or recharge/ replace battery for the second and further sorties. ...
The aim of this research is to detect the post-disaster damage by drones as soon as possible so that decision makers can assign search and rescue teams effectively and efficiently. The main differences of this research from the others, which use drones in literature, are as: First, the regions are divided into grids and different importance values are assigned according to the number of buildings that are likely to be damaged and are vital for the response stage, such as hospitals, schools, and fire stations. Second, these importance levels are updated based on the day and time, which helps ordering the grids in a more realistic manner. Third, the depots are selected among the predetermined candidate locations in accordance with the purpose of objective function. Fourth, detection times at grids are considered as uncertain. Fifth, two versions of Ant Colony Optimization (ACO) are developed as alternatives to exact solution tools. Last, sensitivity analyzes are performed by reducing the number of sorties, reducing the number of drones, and comparing day and night importance values for each instance. According to the results, only for very small-scale instances, exact solution tool was able to reach the optimal while both versions of ACO reached to similar results within a very less CPU times. Additionally, these ACO algorithms also found good results for the larger scaled problems. Then the performance of these ACO algorithms and the exact solution method are compared based on the CPU time and solution quality.
... Among the existing DL structures used for damage assessment purposes, convolutional neural networks (CNNs) have been proven to be more effective and accurate in BDM generation due to their automatic feature extraction allowing for the effective extraction of multilevel features of buildings, such as colors, edges, and problemspecific features [20], [21]. Notably, amongst CNN-based structures, U-Net architecture [22] as an appropriate choice has attracted much attention in the context of building damage detection [23], [24], [25], [26], [27], [28]. Because the U-Net structure can work efficiently even with a small amount of labeled data [22], which is especially suited to building damage assessment where annotated data are challenging to be obtained. ...
Nowadays, unmanned aerial vehicle (UAV) remote sensing data are key operational sources used to produce a reliable building damage map (BDM), which is of great importance in instant response and rescue operations after earthquakes. The present study proposes a novel weighted ensemble transferred U-Net-based model (WETUM) consisting of two major steps to create a reliable binary BDM using UAV data. In the first step of the proposed approach, three individual initial BDMs are predicted by three pre-trained U-Net-based composite networks. In the second step, these three individual predictions are linearly integrated through a proposed grid search technique so that an optimized hybrid BDM (OHBDM) incorporating complementary damage information is made. The proposed WETUM was then compared with several conventional deep learning (DL) and machine learning (ML) models. The models were compared across two pivotal scenarios, addressing the impact of diverse feature sets on model performance and generalizability. Specifically, the first scenario focused solely on spectral features, while the second incorporated both spectral and geometrical features. To make the comparisons, this study conducted empirical analyses using UAV spectral and geometrical data acquired over Sarpol-e Zahab, Iran. The experimental findings showed that the synergic use of spectral and geometrical data boosted both DL- and ML-based approaches in damage detection. Moreover, the proposed WETUM with DDR values of 65.22 and 78.26 (%), respectively, for the first and second scenarios, outperformed all the compared methods. Notably, WETUM with only spectral data outperformed the random forest (RF) classifier equipped with many hand-crafted spectral and geometrical features, indicating the highest potential and generalizability of the proposed WETUM for building damage evaluation in a new unseen earthquake-affected area.
... Damage detection is a major priority in civil engineering, with the greatest focus on structures such as bridges, buildings, roads, and dams (Arya et al., 2021;Mangalraj et al., 2022;Rius et al., 2017;Zhang et al., 2017). This has led to the development of an integrated health monitoring system with sufficient damage detection technique to precisely obtain information about health, serviceability, integrity, and structural safety, thereby providing vital information on the operating state and structural reliability of important infrastructure . ...
The damage locating vector (DLV) is a popular second-level method for hazardous detection on truss structures, despite having limitations that can cause the misidentification of damaged parts. Some of these limitations may be attributed to sensors, damaged structural parts, and complex systems. Therefore, this study proposes a new damage detection technique known as EDLV (i.e., enhanced DLV) to resolve these truss structure problems. This technique is considered a third-level method due to its ability to obtain the stiffness/strength reduction of a damaged part. EDLV is a technique that combines DLV with an optimization algorithm to “enhance” the ability of DLV technique. To investigate the capability of EDLV, two plane truss structures with several damage scenarios were considered (i.e., single damage, double damages, triple damages and multiple damages). The results showed that EDLV was more robust for locating damaged parts than DLV. In addition to determining the proper damage locations on these parts, EDLV also precisely prioritized their corresponding stiffness reductions.
Evaluation of the severity of structural damage caused by earthquakes is time-consuming using labor-intensive approaches. Deep learning methods have been increasingly used to detect and classify image damages, whereas few studies have been applied to assess the level of damaged structural images such as no, light, moderate, or severe damages under earthquakes. The goal of this paper is to propose an approach to automatically categorize the structures using convolutional neural networks (CNNs). Two pre-trained CNNs, namely InceptionV3 and Xception were employed. Adaptive moment estimation algorithm (Adam) was adopted to optimize the parameters of deep learning models. A gradient-weighted class activation mapping (Grad-CAM) was applied for locating damages. The testing results highlighted that InceptionV3 and Xception algorithms showed a high performance with 86.67% and 88.33% accuracy, respectively. The Grad-CAM located successfully the actual damages to structures. It depicts the ability to use CNNs for the assessment of structural damages using images.
Nowadays, one of the most challenging issues in the field of remote sensing and satellite imagery is production of cadastral maps. Synthetic aperture radar (SAR) images are among the most widely used satellite images in the last decade. Due to radar nature of these images, the buildings in SAR images face with two problems: shadow and layover. Morphological mathematics are efficient tools for detection of buildings with providing a contextual profile containing shape and geometrical characteristics of the objects in radar images. By using the suggested method, two characteristics of shadow and brightness are detected separately. Then, the construction areas are extracted by using a fuzzy fusion approach. In this method, various parameters such as size and direction of the structural element and the weighting factor of the shadow, bright area, and the recursive parameter have to be determined independently. To this end, an iterative method using MSE is suggested. The experimental results show a detection rate of 94.3% achieved by the proposed method.
The PASCAL VOC Challenge performance has been significantly boosted by the prevalently CNN-based pipelines like Faster R-CNN. However, directly applying the Faster R-CNN to the small remote sensing objects usually renders poor performance. To address this issue, this paper investigates on how to modify Faster R-CNN for the task of small object detection in optical remote sensing images. First of all, we not only modify the RPN stage of Faster R-CNN by setting appropriate anchors but also leverage a single high-level feature map of a fine resolution by designing a similar architecture adopting top-down and skip connections. In addition, we incorporate context information to further boost small remote sensing object detection performance while we apply a simple sampling strategy to solve the issue about the imbalanced numbers of images between different classes. At last, we introduce a simple yet effective data augmentation method named 'random rotation' during training. Experimental results show that our modified Faster R-CNN algorithm improves the mean average precision by a large margin on detecting small remote sensing objects.
The main goal of this work was to identify the areas that are most susceptible to desertification in a part of the Algerian steppe, and to quantitatively assess the key factors that contribute to this desertification. In total, 139 desertified zones were mapped using field surveys and photo-interpretation. We selected 16 spectral and geomorphic predictive factors, which a priori play a significant role in desertification. They were mainly derived from Landsat 8 imagery and Shuttle Radar Topographic Mission digital elevation model (SRTM DEM). Some factors, such as the topographic position index (TPI) and curvature, were used for the first time in this kind of study. For this purpose, we adapted the logistic regression algorithm for desertification susceptibility mapping, which has been widely used for landslide susceptibility mapping. The logistic model was evaluated using the area under the receiver operating characteristic (ROC) curve. The model accuracy was 87.8%. We estimated the model uncertainties using a bootstrap method. Our analysis suggests that the predictive model is robust and stable. Our results indicate that land cover factors, including normalized difference vegetation index (NDVI) and rangeland classes, play a major role in determining desertification occurrence, while geomorphological factors have a limited impact. The predictive map shows that 44.57% of the area is classified as highly to very highly susceptible to desertification. The developed approach can be used to assess desertification in areas with similar characteristics and to guide possible actions to combat desertification.
In this paper, we focus on tackling the problem of automatic accurate localization of detected objects in high-resolution remote sensing images. The two major problems for object localization in remote sensing images caused by the complex context information such images contain are achieving generalizability of the features used to describe objects and achieving accurate object locations. To address these challenges, we propose a new object localization framework, which can be divided into three processes: region proposal, classification, and accurate object localization process. First, a region proposal method is used to generate candidate regions with the aim of detecting all objects of interest within these images. Then, generic image features from a local image corresponding to each region proposal are extracted by a combination model of 2-D reduction convolutional neural networks (CNNs). Finally, to improve the location accuracy, we propose an unsupervised score-based bounding box regression (USB-BBR) algorithm, combined with a nonmaximum suppression algorithm to optimize the bounding boxes of regions that detected as objects. Experiments show that the dimension-reduction model performs better than the retrained and fine-tuned models and the detection precision of the combined CNN model is much higher than that of any single model. Also our proposed USB-BBR algorithm can more accurately locate objects within an image. Compared with traditional features extraction methods, such as elliptic Fourier transform-based histogram of oriented gradients and local binary pattern histogram Fourier, our proposed localization framework shows robustness when dealing with different complex backgrounds.
In the recent years, remote sensing technologies become important tools in disaster management, especially in the aftermath of destructive natural disasters. In this paper, various remote sensing data acquired after the M w 7.0 Kumamoto earthquake, Japan are introduced. Airborne optical and Lidar data as well as satellite optical and SAR data were captured by various international space agencies and aerial survey companies. Preliminary research results using these remote sensing data are demonstrated, comparing with the data from field surveys and GNSS observation, and seismic records.
There is large consent that successful training of deep networks requires
many thousand annotated training samples. In this paper, we present a network
and training strategy that relies on the strong use of data augmentation to use
the available annotated samples more efficiently. The architecture consists of
a contracting path to capture context and a symmetric expanding path that
enables precise localization. We show that such a network can be trained
end-to-end from very few images and outperforms the prior best method (a
sliding-window convolutional network) on the ISBI challenge for segmentation of
neuronal structures in electron microscopic stacks. Using the same network
trained on transmitted light microscopy images (phase contrast and DIC) we won
the ISBI cell tracking challenge 2015 in these categories by a large margin.
Moreover, the network is fast. Segmentation of a 512x512 image takes less than
a second on a recent GPU. The full implementation (based on Caffe) and the
trained networks are available at
http://lmb.informatik.uni-freiburg.de/people/ronneber/u-net .
This book provides an essential appraisal of the recent advances in technologies, mathematical models and computational software used by those working with geodetic data. It explains the latest methods in processing and analyzing geodetic time series data from various space missions (i.e. GNSS, GRACE) and other technologies (i.e. tide gauges), using the most recent mathematical models. The book provides practical examples of how to apply these models to estimate seal level rise as well as rapid and evolving land motion changes due to gravity (ice sheet loss) and earthquakes respectively. It also provides a necessary overview of geodetic software and where to obtain them.
There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at
http://lmb.informatik.uni-freiburg.de/people/ronneber/u-net
.
Building damage detection after earthquake would help to rapid relief and response of disaster. In this study, an efficient method was proposed for building damage detection in urban area after earthquake using pre-event vector map and postevent pan-sharpened high spatial resolution image. At first, preprocessing was applied on the postevent satellite image. Second, results of pixel- and object-based classifications were integrated. In the following, geometric features of buildings were extracted including area, rectangular fit (rect_fit), and convexity. A decision-making system based on these features and an adaptive network-based fuzzy inference system (ANFIS) model was designed to attain building damage degree. A comprehensive sensitivity analysis was carried out to find proper parameters of the ANFIS model leading to accurate damage results. The proposed method was tested over earthquake data set of Bam city in Iran. The results of our method indicate that an overall accuracy of 76.36% and kappa coefficient of 0.63 were achieved to identify building damage degree. The obtained results indicate that the postevent geometrical features (relative change of different damage levels with respect to each other) along with the ANFIS model can help to reach better results in building damage detection.