A pixel-wise labelled dataset of Moroccan aircraft emergency landing sites for semantic segmentation applications

Abstract

Detecting emergency aircraft landing sites is crucial for ensuring passenger and crew safety during unexpected forced landings caused by factors like engine malfunctions, adverse weather, or other aviation emergencies. In this article, we present a dataset consisting of Google Maps images with their corresponding masks, specifically crafted with manual annotations of emergency aircraft landing sites, distinguishing between safe areas with suitable conditions for emergency landings and unsafe areas presenting hazardous conditions. Drawing on detailed guidelines from the Federal Aviation Administration, the annotations focus on key features such as slope, surface type, and obstacle presence, with the goal of pinpointing appropriate landing areas. The proposed dataset has 4180 images, with 2090 raw images accompanied by their corresponding annotation instances. This dataset employs a semantic segmentation approach, categorizing the image pixels into two ''Safe'' and ''Unsafe'' classes based on authenticated terrain-specific attributes, thereby offering a nuanced understanding of the viability of various landing sites in emergency scenarios.

Keywords

• Semantic segmentation
• Object detection
• Emergency landing
• Image dataset
• Classification