AirNet – X: Object Detection From Aerial Images for Enhanced Disaster Detection

ABSTRACT Natural disasters present significant challenges to emergency responders, as they must quickly assess damage, locate survivors, and allocate resources effectively, often in chaotic and inaccessible environments. Traditional methods of ground‑based assessments are slow, limited in scope, and hindered by obstructions, leading to delays in critical decision‑making. This chapter addresses these limitations by developing an deep learning model of “AirNet–X:Aerial Object Detection System for Enhanced Disaster Detection” that leverages advanced machine learning techniques. By introducing AI‑powered aerial object detection, it offers a transformative solution to one of disaster management’s most pressing problems: the need for rapid and reliable situational awareness. It underscores the potential of technology to enhance societal resilience, reduce response times, and save lives during critical emergencies. The solution integrates the modified ResNet50 architecture, ahigh-performing convolutional neural network, with the DOTA(Dataset for Object Detection in Aerial Images), a comprehensive dataset of annotated aerial imagery. By training the system on16 distinct object categories, including vehicles, buildings, and ships, the model learns to detect and classify objects from aerial drone images accurately. Using oriented bounding boxes, it identifies disaster indicators like collapsed structures, blocked roads, and stranded individuals, providing actionable intelligence in real time. The results demonstrate the system’s effectiveness, with classification accuracy ranging from 80–90% for most categories and slightly lower accuracy for smaller or obscured objects. This system empowers emergency responders to make faster, more informed decisions, prioritizing rescue efforts and optimizing resource deployment. Beyond the immediate response, it aids recovery efforts by mapping damaged areas and monitoring infrastructure repair progress.