Emergency Landing Field Identification Based on Artificial Neural Networks
Steffen Flämig, Wolfram H. Schiffmann · 2024
In the case of an aircraft's engines failure, it is necessary to approach a suitable landing field immediately. In some cases, the excess altitude over ground is not sufficient to glide to a regular airport with the given gliding characteristics of the aircraft. A suitable emergency landing field must then be flown to. Determining a suitable area in real time from the flying aircraft is often very difficult. Limited visibility conditions and a viewing angle that is too flat make it sometimes impossible to obtain sufficient information about the accessible surroundings. A solution is a database in which emergency landing fields are stored with data such as position, longitude, latitude, orien-tation, inclination and type of surfaces. These databases can be generated offline from a variety of geodata and accessed during the flight. For some areas of the world, high-resolution geospatial data such as digital surface models (DSM) and digital orthophotos (DOP) are available. In these cases, landable areas can be identified using conventional digital image processing methods. However, for large areas of the earth, only digital surface models with low resolution are available. In these cases, it is not possible to draw direct conclusions about their suitability as a landing field. We present a method that trains neural networks with data from areas for which high-resolution surface models are avail-able. These neural networks are then applied to lower resolution surface data. The results are compared with the results in the presence of high-resolution surface data. It is shown that with the presented method, emergency landing fields can be identified even using surface models with low resolution. This makes it possible to create a global emergency landing field database.