FloodNet: A High Resolution Aerial Imagery Dataset for Post Flood Scene Understanding (2012.02951v1)

Abstract: Visual scene understanding is the core task in making any crucial decision in any computer vision system. Although popular computer vision datasets like Cityscapes, MS-COCO, PASCAL provide good benchmarks for several tasks (e.g. image classification, segmentation, object detection), these datasets are hardly suitable for post disaster damage assessments. On the other hand, existing natural disaster datasets include mainly satellite imagery which have low spatial resolution and a high revisit period. Therefore, they do not have a scope to provide quick and efficient damage assessment tasks. Unmanned Aerial Vehicle(UAV) can effortlessly access difficult places during any disaster and collect high resolution imagery that is required for aforementioned tasks of computer vision. To address these issues we present a high resolution UAV imagery, FloodNet, captured after the hurricane Harvey. This dataset demonstrates the post flooded damages of the affected areas. The images are labeled pixel-wise for semantic segmentation task and questions are produced for the task of visual question answering. FloodNet poses several challenges including detection of flooded roads and buildings and distinguishing between natural water and flooded water. With the advancement of deep learning algorithms, we can analyze the impact of any disaster which can make a precise understanding of the affected areas. In this paper, we compare and contrast the performances of baseline methods for image classification, semantic segmentation, and visual question answering on our dataset.

• The paper introduces FloodNet, a UAV-acquired high-resolution imagery dataset for detailed post-disaster scene analysis.
• It evaluates computer vision methods like semantic segmentation and VQA, with models such as ResNet50 and DeepLabv3+ achieving noteworthy performance.
• FloodNet’s detailed imagery provides actionable insights for effective disaster damage assessment and inspires future multi-task research.

FloodNet: A High Resolution Aerial Imagery Dataset for Post Flood Scene Understanding

FloodNet is positioned as a specialized dataset designed to enhance post-disaster scene understanding through high-resolution aerial imagery acquired via Unmanned Aerial Vehicles (UAVs). This dataset, primarily captured during the aftermath of Hurricane Harvey, serves as a novel resource for tackling the intricate challenges associated with natural disaster damage assessment using computer vision techniques. The paper introducing FloodNet explores its creation, characteristics, and implications for future research in the domain of disaster management and response.

Motivation and Dataset Overview

The impetus for FloodNet’s development lies in the limitations of existing datasets such as Cityscapes, MS-COCO, and PASCAL, which are adept in various computer vision tasks but fall short in the field of post-disaster assessment due to their contextual and resolution constraints. Traditional datasets relying on satellite imagery suffer from low spatial resolution and extended revisit periods, rendering them inefficient for rapid damage assessment in the critical phases following a disaster. The utilization of UAV-based imagery in FloodNet presents a significant advancement by providing detailed, high-resolution captures from challenging terrains, crucial for precise analysis and decision-making in crisis situations.

FloodNet comprises UAV images meticulously labeled on a pixel level for semantic segmentation and equipped with structured visual question answering components. This dataset presents the unique situations encountered post-flood, such as distinguishing between different water bodies and identifying submerged infrastructures. The provision of high-detail images aligns with the precise classification of flooded versus non-flooded buildings and roads, a task pivotal for effective disaster response.

Methodological Assessment and Comparative Analysis

The paper evaluates the baseline performance of established computer vision methodologies, namely image classification, semantic segmentation, and visual question answering (VQA), using FloodNet. Various convolutional neural networks (CNNs), such as ResNet50, InceptionNetv3, and Xception, demonstrated varying effectiveness in the classification task, with ResNet50 yielding the highest test accuracy. Semantic segmentation through state-of-the-art models like PSPNet, ENet, and DeepLabv3+ successfully highlights the advantages of high-resolution imagery in capturing fine details and contextual nuances critical in post-disaster environments.

Alongside image processing pipelines, FloodNet’s integration with VQA tasks presents an interdisciplinary challenge intersecting natural language processing with computer vision. Models such as the Multimodal Factorized Bilinear (MFB) network showed superior performance in VQA tasks, a testament to the dataset’s capability to support complex scene understanding through question-based frameworks.

Implications and Future Directions

FloodNet marks a significant contribution to disaster management research through its multi-faceted exploration of high-resolution disaster imagery. Practically, it offers a promising tool for enhancing disaster response operations by providing accurate, real-time insights into crisis-affected areas. Theoretically, this dataset lays the groundwork for advanced neural network architectures specifically tailored to address post-disaster scenario complexities, with potential iterative improvements to ensure adaptability across various disaster types beyond flooding.

Future endeavors, as suggested by the authors, involve the expansion of such datasets to incorporate diverse natural disasters, fostering a more comprehensive machine learning ecosystem for emergency response. Additionally, more sophisticated algorithms can be developed, leveraging FloodNet as a benchmark for enhancing multi-task learning capabilities in disaster-stricken environments. FloodNet thus stands as a pivotal resource, inspiring further research in the integration of UAV imagery with complex machine learning frameworks, aimed at elevating the precision and efficacy of disaster management solutions.