Abstract:Very High Resolution (VHR) geospatial image analysis is crucial for humanitarian assistance in both natural and anthropogenic crises, as it allows to rapidly identify the most critical areas that need support. Nonetheless, manually inspecting large areas is time-consuming and requires domain expertise. Thanks to their accuracy, generalization capabilities, and highly parallelizable workload, Deep Neural Networks (DNNs) provide an excellent way to automate this task. Nevertheless, there is a scarcity of VHR data pertaining to conflict situations, and consequently, of studies on the effectiveness of DNNs in those scenarios. Motivated by this, our work extensively studies the applicability of a collection of state-of-the-art Convolutional Neural Networks (CNNs) originally developed for natural disasters damage assessment in a war scenario. To this end, we build an annotated dataset with pre- and post-conflict images of the Ukrainian city of Mariupol. We then explore the transferability of the CNN models in both zero-shot and learning scenarios, demonstrating their potential and limitations. To the best of our knowledge, this is the first study to use sub-meter resolution imagery to assess building damage in combat zones.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: using deep - learning methods to assess building damage in conflict areas (such as war scenarios). Specifically, the author focuses on how to apply the deep neural network models originally used in natural - disaster scenarios to the assessment of building damage caused by war. ### Problem Background 1. **The Importance of High - Resolution Geospatial Images** - In humanitarian aid and crisis response, very - high - resolution (VHR) geospatial image analysis is crucial because it can quickly identify key areas in greatest need of support. - Traditional methods rely on manual inspection and reporting, which are often time - consuming and require professional knowledge. 2. **Advantages of Deep Learning** - Deep neural networks (DNNs) provide an excellent method for automating this task because of their accuracy, generalization ability, and highly parallel workload. - DNNs perform well in handling large - volume geospatial data and are able to learn and extract relevant features from raw images. 3. **Data Scarcity** - Although there is a large amount of VHR data on natural disasters, the VHR data related to war scenarios is relatively scarce, which limits the development of DNN models for conflict areas. ### Research Objectives - **Verify Transferability**: Research whether existing deep convolutional neural networks (CNNs) can be transferred from natural - disaster scenarios to war scenarios for building - damage assessment. - **Construct a New Dataset**: Collect and label a high - resolution satellite - image dataset of Mariupol, Ukraine, before and after the war. - **Evaluate Model Performance**: Evaluate the performance of these models in war scenarios through zero - shot and fine - tuning experiments, and explore their potential and limitations. ### Main Contributions 1. **Extensive Testing of SotA CNN Models**: Conducted extensive testing on four state - of - the - art CNN models and compared their performance in natural - disaster and war scenarios. 2. **Propose a Data - Augmentation Pipeline**: To improve model accuracy on small datasets, proposed a custom - made data - augmentation pipeline. 3. **Ablation Study**: Through ablation study, emphasized the importance of data - augmentation and dilated - filtering pre - processing. 4. **Performance Metrics**: The models achieved a maximum F1 - score of 69% and 59% on 2 - class and 3 - class damage - assessment problems respectively, as well as a maximum balanced - accuracy rate of 86% and 79%. ### Formula Representation When evaluating model performance, the following formulas were used: - **F1 - Score** \[ F1_{Ci}=\frac{2TP_{Ci}}{2TP_{Ci}+FP_{Ci}+FN_{Ci}} \] where \( TP_{Ci} \), \( FP_{Ci} \) and \( FN_{Ci} \) represent the number of true positives, false positives and false negatives of the \( i \) - th class respectively. - **Balanced Accuracy Score (BAS)** \[ BAS = \frac{1}{C}\sum_{i = 1}^{C}\frac{TP_i}{TP_i+FN_i} \] where \( C \) is the total number of classes. ### Summary This paper aims to fill the research gap in building - damage assessment in war scenarios. By constructing a new dataset and evaluating the transferability of existing models, it provides valuable insights for future research and practical applications.

Building Damage Assessment in Conflict Zones: A Deep Learning Approach Using Geospatial Sub-Meter Resolution Data

On the Deployment of Post-Disaster Building Damage Assessment Tools Using Satellite Imagery: A Deep Learning Approach.

DEEP LEARNING FOR AUTOMATIC BUILDING DAMAGE ASSESSMENT: APPLICATION IN POST-DISASTER SCENARIOS USING UAV DATA

Building Damage Detection in Satellite Imagery Using Convolutional Neural Networks

DeepDamageNet: A two-step deep-learning model for multi-disaster building damage segmentation and classification using satellite imagery

Superpixel-Based Building Damage Detection from Post-earthquake Very High Resolution Imagery Using Deep Neural Networks

RescueNet: Joint Building Segmentation and Damage Assessment from Satellite Imagery

AB2CD: AI for Building Climate Damage Classification and Detection

Evaluation of Deep Learning Models for Building Damage Mapping in Emergency Response Settings

Automated building damage assessment and large‐scale mapping by integrating satellite imagery, GIS, and deep learning

Multi-view deep learning for reliable post-disaster damage classification

A Deep Learning-Based Method for the Semi-Automatic Identification of Built-Up Areas within Risk Zones Using Aerial Imagery and Multi-Source GIS Data: An Application for Landslide Risk

Assessment of Convolutional Neural Network Architectures for Earthquake-Induced Building Damage Detection based on Pre- and Post-Event Orthophoto Images

Deep learning‐based multi‐class damage detection for autonomous post‐disaster reconnaissance

Learning from multimodal and multitemporal earth observation data for building damage mapping

Technical Solution Discussion for Key Challenges of Operational Convolutional Neural Network-Based Building-Damage Assessment from Satellite Imagery: Perspective from Benchmark xBD Dataset

Efficient building damage assessment from post-disaster aerial video using lightweight deep learning models

Building Damage Detection from Post-Event Aerial Imagery Using Single Shot Multibox Detector

Interpretability in Convolutional Neural Networks for Building Damage Classification in Satellite Imagery

Deep learning for post‐hurricane aerial damage assessment of buildings

Enhanced satellite imagery analysis for post-disaster building damage assessment using integrated ResNet-U-Net model