Deep learning for aircraft emergency landing identification: A New Moroccan Terrain Dataset Case Study

In the aviation industry, forced landings are undesirable occurrences that may occur to an aircraft while it is in flight. They may result from unanticipated circumstances, or engine failures. One approach that could be used to find prospective emergency landing locations for airplanes is image segmentation. Accurate segmentation should enhance aviation safety procedures generally by improving the efficient identification of safe landing zones. The accurate identification of safe landing places is crucial to aviation safety in emergency situations. Using a pixel-wise labeled dataset, this paper introduces a deep learning framework for semantic segmentation with the goal of identifying emergency landing zones in Morocco [1]. With an emphasis on terrain features like surface type, slope, and obstructions, the study assesses cutting-edge architectures such as DeepLabV3+, SegFormer, Faster R-CNN, and U-Net. To improve model generalization, a strong preparation pipeline that uses data augmentation approaches is used. Metrics like F1-score, recall, precision, and Dice coefficients show how effective the model is, and experimental findings show notable gains in segmentation accuracy. Critical gaps in the literature have prompted our investigation, especially with regard to aircraft and emergency runways, which have received relatively little attention. Since the urgent landing objective is a point rather than a piste, the majority of the work that has been published thus far is specifically for UAVs, ([2]-[7]). The results highlight the possibility of incorporating these techniques for safer emergency landings into autonomous aircraft systems. Hybrid models will be investigated in future studies, and the application will be expanded to include more varied datasets.