Post-Disaster Classification of Building Damage Using Transfer Learning

Abstract. In the wake of a volcanic eruption, the rapid assessment of building damage is paramount for effective response and recovery planning. Uncrewed aerial vehicles, UAVs, offer a unique opportunity for assessing damage after a volcanic eruption, with the ability to collect on-demand imagery safely and rapidly from multiple perspectives at high resolutions. In this work, we established a UAV-appropriate tephra fall building damage state framework and used it to label ∼50 000 building bounding boxes around ∼2000 individual buildings in 2811 optical images collected during surveys conducted after the 2021 eruption of La Soufrière volcano, St Vincent and the Grenadines. We used these labelled data to train convolutional neural networks (CNNs) for (1) building localisation (average precision equals 0.728) and (2) damage classification into two levels of granularity: no damage vs. damage (F1 score = 0.809) and moderate damage vs. major damage (F1 score = 0.838) (1 is the maximum obtainable for both metrics). The trained models were incorporated into a pipeline along with all the necessary image processing steps to generate spatial data (a georeferenced vector with damage state attributes) for rapid tephra fall building damage mapping. Using our pipeline, we assessed tephra fall building damage for the town of Owia, finding that 22 % of buildings that received 50–90 mm of tephra accumulation experienced at least moderate damage. The pipeline is expected to perform well across other volcanic islands in the Caribbean where building types are similar, though it would benefit from additional testing. Through cross-validation, we found that the UAV look angle had a minor effect on the performance of damage classification models, while for the building localisation model, the performance was affected by both the look angle and the size of the buildings in images. These observations were used to develop a set of recommendations for data collection during future UAV tephra fall building damage surveys. This is the first attempt to automate tephra fall building damage assessment solely using post-event data. We expect that incorporating additional training data from future eruptions will further refine our model and improve its applicability worldwide. To facilitate continued development and collaboration all trained models and the pipeline code can be downloaded from GitHub.

Scalable and rapid building damage detection after hurricane Ian using causal Bayesian networks and InSAR imagery

A novel attention-based deep learning method for post-disaster building damage classification

Timely and accurate assessment of hurricane-induced building damage is crucial for effective post-hurricane response and recovery efforts. Recently, remote sensing technologies provide large-scale optical or Interferometric Synthetic Aperture Radar (InSAR) imagery data immediately after a disastrous event, which can be readily used to conduct rapid building damage assessment. Compared to optical satellite imageries, the Synthetic Aperture Radar can penetrate cloud cover and provide more complete spatial cover-age of damaged zones in various weather conditions. However, these InSAR imageries often contain highly noisy and mixed signals induced by co-occurring or co-located building damage, flood, flood/wind-induced vegetation changes, as well as anthropogenic activities, making it challenging to extract accurate building damage information. In this paper, we introduced a causality-informed Bayesian network inference approach for rapid post-hurricane building damage detection from InSAR imagery. This approach encoded complex causal dependencies among wind, flood, building damage, and InSAR imagery using a holistic causal Bayesian network. Based on the causal Bayesian network, we further jointly inferred the large-scale unobserved building damage by fusing the information from InSAR imagery with prior physical models of flood and wind, without the need for ground truth labels. Furthermore, we validated our estimation results in a real-world devastating hurricane---the 2022 Hurricane Ian. We gathered and annotated building damage ground truth data in Lee County, Florida, and compared the introduced method's estimation results with the ground truth and benchmarked it against state-of-the-art models to assess the effectiveness of our proposed method. Results show that our method achieves rapid and accurate detection of building damage, with significantly reduced processing time compared to traditional manual inspection methods.

The ability to evaluate the damage to buildings both accurately and precisely is essential for disaster recovery, planning, and rescue services. This paper proposes a new approach based on integrating machine learning algorithms in building damage classification. To achieve higher precision in classifying the level of building damage, this research proposes a new technique that employs machine learning strategies. The researchers were able to train the model to be able to differentiate the different levels of building damage and the feature extraction was performed through machine learning. The model effectively extracts and learns multiple complex signals which represent different degrees of damage from a well picked database which include several degrees of damage. In a single pass, the Siamese U-Net can perform feature extraction and similarity measurement between two different images. The efficiency and effectiveness of the Siamese U-net model can be increased by reducing inference time, thus increasing its ability to deliver faster predictions while also improving its accuracy. The suggested Enhanced U-Net (EU-Net) could greatly increase the accuracy of building-level classification. As it turned out, the results are very promising and reach beyond traditional approaches with bringing more sample opportunities of machine learning integration in the building damage assessment context. Additionally, this study believes that the accuracy of building damage classification can be further enhanced demonstrating the usefulness of machine learning in disaster management.

Sudden-onset natural disasters, such as destructive earthquakes, pose significant threats to human life and property. The use of high-resolution remote sensing (HRRS) images for automated assessment of building damage can rapidly and accurately provide spatial distribution information and statistical data on building damage, assisting in disaster response and relief efforts. However, the task is exceedingly challenging due to the diverse and intricate appearance of damaged buildings in HRRS images, coupled with interference from surrounding areas that exhibit certain damage characteristics as a result of the disaster. To overcome these issues, this article proposes a weakly supervised building damage assessment method based on scene change detection in pre- and post-disaster bitemporal images. This method fully leverages visual information of building boundaries and deeper semantic information of building scenes from pre-disaster images to guide the identification of building damage in post-disaster images. Specifically, the method first generates fine-grained subbuilding objects with detailed boundaries from pre-disaster images by combining semantic segmentation of buildings with superpixel segmentation. Then, bitemporal image scene blocks obtained using sub-building objects as clues are input into our proposed Siamese local-global visual transformer (SLgViT) network, enabling scene change detection guided by deep semantic information from pre-disaster images. Finally, the change detection results serve as the basis to depict pixel-level building damage in post-disaster images. The proposed SLgViT network is primarily composed of a specially designed local-global visual transformer (LgViT) module and a cross-Siamese interaction fusion (CSIF) module, both of which play a crucial role in the deep mining and integrated interaction of local and global semantic features from pre- and post-disaster images. It is noteworthy that our method operates in a weakly supervised manner. The training of the SLgViT network requires only scene patches centered around building objects from pre- and post-disaster bitemporal images, along with image-level annotations. Experiments conducted with satellite images from the 2010 Port-au-Prince, Haiti earthquake and unmanned aerial vehicle (UAV) images from the 2019 Changning, China earthquake have demonstrated the effectiveness and superior performance of the proposed method.

Assessing building damage in coastal communities after a hurricane event is crucial for reducing both immediate and long-term disaster impacts, as well as for enhancing resilience planning and disaster preparedness. Despite the extensive data collection efforts of the post-hurricane reconnaissance teams, some information on the structural features of damaged buildings is often missing due to various reasons, like the absence of relevant documents or severe building damage, thereby limiting our comprehensive understanding of building resilience to natural disasters. This study introduces a machine learning approach based on extreme gradient boosting (XGBoost) to reconstruct missing structural features of the damaged buildings from four types of data (known structural, geospatial, hazard, and damage level information). XGBoost models were trained based on the reconnaissance datasets collected from four regions affected by Hurricanes. For each region, we analyzed the model’s performance depending on the missing structural features. We also demonstrated the importance of including geospatial, hazard, and damage level data by showing improved performance of XGBoost models compared to those trained only on known structural data. Furthermore, we examined how the accuracy of the XGBoost approach changes if multiple structural features are missing. This XGBoost approach has the potential to support post-hurricane building damage assessments by providing missing building details, enabling comprehensive post-disaster analysis.

Building damage assessment based on satellite imagery is important to aid disaster response and resource allocation in the affected areas very quickly. Recent developments in deep learning approaches have led to an improvement in automated damage detection, but most existing approaches have had a difficult time combining pre-and post-disaster imagery for effective classification of nuanced types of damage. This gap restricts the precision of the identification of minor and major damage which are crucial for prioritizing recovery efforts. The goal of this research is to propose a novel improved damage assessment framework, CADSNet (Cross-Aligned Dual-Stream Network) with dual stream ResNet34 backbone and deformable cross-alignment for robust feature fusion to overcome these limitations. The methodology includes preprocessing on the xBD dataset to aligned 512x512 tiles using joint percentile stretching and generating building, damage, ignore and change mask for high-quality training samples. Experimental evaluation proves that CADSNet has state-of-the-art performance with a Macro-F1 score of 0.772, Building IoU of 0.686 and Damage AP of 0.880, outperforming several recent benchmarks. These improvements improve both spatial localization and categorical accuracy and allow more reliable and timely assessment of building damages, which can have a very positive effect on disaster management agencies in their decision-making and resource allocation process.

The collapse of buildings caused by destructive earthquakes often caused severe casualties and economic losses. After an earthquake, the assessment of building damage is one of the most important tasks in earthquake emergency response. Accurate assessment of building damage will be essential in making plans of emergency responses. Four-Polarimetric Synthetic Aperture Radar (PolSAR) data has the advantages of Synthetic Aperture Radar (SAR) imaging that is not occluded by sunlight and clouds, it also contains the most abundant information of four polarimetric channels. Due to the large amount of information in PolSAR data, only a single post-earthquake PolSAR image can be used to identify building damage of post-earthquake. It is easy to overestimate the number of collapsed buildings and the damage degree of earthquakes only using a traditional polarimetric decomposition method for PolSAR data. The layout of urban buildings can be diverse. Buildings can stand in parallel in typical SAR imaging with strong scattering features, there are also some oriented standing buildings with lower scattering intensity and with similar scattering characteristics of collapsed buildings, thus these oriented buildings are often misconstrued as collapsed buildings. The spatial frequency of SAR images can be clearly rendered in the frequency domain. In this study, we propose a new texture feature based on Fourier transform, namely the sector texture feature of the Fourier amplitude spectrum (STFFAS), to solve the overestimate of damage of buildings, which are caused by earthquakes. STFFAS can well describe the difference in texture between oriented buildings and collapsed buildings and accurately recognize the two types of buildings. The STFFAS index can be defined as follows:             (1)where ‘FFT’, ‘std’, ‘mean’ and ‘lg’ represent the function of 2D fast Fourier transform, standard deviation, mean values and logarithm to the base 10, respectively; ‘real’ and ‘imag’ represent the real parts and imaginary parts of complex numbers, respectively. Meanwhile, based on the Yamaguchi four-component decomposition method and the STFFAS texture feature parameter, we develop a solution to identify the damage of buildings only using a single post-earthquake PolSAR image. The Ms7.1 Yushu earthquake, which occurred in Yushu County of China on 14th April, 2010, is used as a study case to carry out the experiment with 75000 undamaged and damaged building samples. With the proposed method, the overall accuracy of correct building damage recognition with STFFAS is 81.30%. The Producer‘s Accuracy (PA) of damaged buildings, which is the correct recognition rate of collapsed buildings, is 81.06%; and the PA of undamaged buildings, which is the correct recognition rate of undamaged buildings, reaches 81.42%. Compared with the traditional polarimetric decomposition method, 70.18% standing buildings are successfully isolated from the mixture of collapsed buildings. Therefore this new method has greatly improved the accuracy and reliability of extracting damage information of buildings. This result well confirmed that the texture feature in frequency domain is effective for building damage recognition.

In recent years, remote-sensing (RS) technologies have been used together with image processing and traditional techniques in various disaster-related works. Among these is detecting building damage from orthophoto imagery that was inflicted by earthquakes. Automatic and visual techniques are considered as typical methods to produce building damage maps using RS images. The visual technique, however, is time-consuming due to manual sampling. The automatic method is able to detect the damaged building by extracting the defect features. However, various design methods and widely changing real-world conditions, such as shadow and light changes, cause challenges to the extensive appointing of automatic methods. As a potential solution for such challenges, this research proposes the adaption of deep learning (DL), specifically convolutional neural networks (CNN), which has a high ability to learn features automatically, to identify damaged buildings from pre- and post-event RS imageries. Since RS data revolves around imagery, CNNs can arguably be most effective at automatically discovering relevant features, avoiding the need for feature engineering based on expert knowledge. In this work, we focus on RS imageries from orthophoto imageries for damaged-building detection, specifically for (i) background, (ii) no damage, (iii) minor damage, and (iv) debris classifications. The gist is to uncover the CNN architecture that will work best for this purpose. To this end, three CNN models, namely the twin model, fusion model, and composite model, are applied to the pre- and post-orthophoto imageries collected from the 2016 Kumamoto earthquake, Japan. The robustness of the models was evaluated using four evaluation metrics, namely overall accuracy (OA), producer accuracy (PA), user accuracy (UA), and F1 score. According to the obtained results, the twin model achieved higher accuracy (OA = 76.86%; F1 score = 0.761) compare to the fusion model (OA = 72.27%; F1 score = 0.714) and composite (OA = 69.24%; F1 score = 0.682) models.

A deep transfer learning-based damage assessment on post-event very high-resolution orthophotos

Nowadays, the Structural Building Health Damage Monitoring System (SBHDMS) is a crucial technology for predicting the civil building structures' health. SBHDMS contains abnormal changes in the buildings in terms of damage levels. Natural Disasters like Earthquakes, Floods, and cyclones affect the unusual changes in the buildings. If the building undergoes any natural disaster, the sensors capture the vibration data or change the buildings' structure. Due to the vibration data, these unusual changes can be analyzed. Here sensors or Machine Learning based Building Damage Prediction (MLBDP) are used for capturing and collecting the vibration data. This paper proposes a Novel Rough Set based Artificial Neural Network with Support Vector Machine (RAS) metaheuristic method. RAS method is used to predict the damaged building's vibration data levels captured by the sensors. For the feature reduction subset, we use one of the essential pre-processing method called the Rough set theory (RST) strategy. RAS has two contributions. The first one is the Support Vector Machine (SVM) classification method used for identifying the structures of the buildings. The artificial Neural Network (ANN) method used to predict the buildings' damage levels is the second contribution. The proposed method (RAS) is accurately predicting the conditions of the construction building structure and predicting the damage levels, without human intervention. Comparing the results states that the proposed method accuracy is better than SVM's classification methods, ANN. The prediction analysis depicts that the RAS method can effectively detect the damage levels.

Based on the classification provided by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT), the damage level of buildings impacted by the 2011 Great East Japan tsunami can be separated into six levels (from minor damage to washed away). The objective of this paper is to identify the significant predictor variables and the direction of their potential relationship to the damage level in order to create a predicting formula for damage level. This study used the detailed data of damaged buildings in Ishinomaki city, Miyagi prefecture, Japan, collected by MLIT. The explanatory variables tested included the inundation depth, number of floors, structural material, and function of the building. Ordinal regression was applied to model the relationship between the ordinal outcome variable (damage level) and the predictors. The findings indicated that inundation depth, structural material, and function of building were significantly associated with the damage level. In addition to this new type of model, this research provides a valuable insight into the relative influence of different factors on building damage and suggestions that may help to revise the classification of current standards. This study can contribute to academic tsunami research by assessing the contribution of different variables to the observed damage using new approaches based on statistical analysis and regression. Moreover, practical applications of these results include understanding of the predominant factors driving tsunami damage to structures, implementation of the relevant variables into the proposed, or alternative model in order to improve current damage predictions by taking into account not only inundation depth, but also variables such as structural material and function of building.

Image-based preliminary emergency assessment of damaged buildings after earthquake: Taiwan case studies

Earthquakes are one of the most destructive natural disasters. Efficiently and quickly acquiring building earthquake damage information can help to reduce the casualties after an earthquake. In this paper, building damage information is extracted using a single post-earthquake PolSAR image. In PolSAR images, since the undamaged oriented buildings characterized by volume scattering with weaker scattering power are very similar to collapsed buildings, the collapsed buildings are difficult to extract accurately. In this paper, the difference in the relative contribution change rate of scattering components before and after polarization orientation angle (POA) compensation is proposed to enhance the difference between collapsed buildings and oriented buildings, in order that the collapsed buildings can be extracted more accurately. The “4.14” Yushu earthquake, which occurred in Yushu County, Qinghai province of China, is used as the case study to test the proposed method, and an airborne high-resolution PolSAR image of the urban region of Yushu County is used in the experiment. The experimental results show that the accuracy of building damage information extraction can be improved by the use of the proposed method, compared with the traditional polarimetric classification.