MTUSI: Neural network for automatic landing of an aircraft

2023: Implementation of a neural network for automatic aircraft landing

Scientists from MTUSI have introduced a neural network for automatic landing of an aircraft. The university announced this on December 13, 2023.

The development of unmanned aerial vehicles allows you to solve a large number of tasks: search, environmental monitoring, control of traffic flows, and performing the functions of a rescuer. Scientists believe that the use of a neural network in the control of unmanned aerial vehicles (UAVs) in particular during landing is relevant, since it is this mode that is the most complex and responsible and has a large degree of aircraft accident.

As part of the work on the master's thesis at the Department of Intelligent Systems in Control and Automation, MTUSI analyzed the current situation and proposed its own development in this direction - a neural network model for the automatic landing of an unmanned aerial vehicle (UAV), the purpose of which is to minimize accidents during landing.

During the study, work was carried out with the architecture of a direct propagation neural network (perceptron) of two hidden layers. To train the neural network, a set of data was produced from 22 satellite images depicting various terrain, which simulate photographs taken from the onboard UAV camera. During the further expansion of the data set, augmentation was carried out, which made it possible to scale satellite images and bring them to 100 images.

The results confirmed that the neural network model has high accuracy when automatically landing an unmanned aerial vehicle.

For more fundamental research results and error minimization in the design of the neural network model, we will include more training materials and consider all possible biases and variances for the developed network. Soon we will start working with convolutional neural networks, with expanded libraries, which will improve the results of the automatic landing of an unmanned aerial vehicle (UAV), - commented Lilia Voronova, Head of the Department of Intelligent Systems in Control and Automation, MTUSI, Professor, Doctor of Medical Sciences

According to scientists, the use of a neural network model in an automatic landing of an aircraft will increase the efficiency, safety and quality of flights. This is due to the fact that the neural network adapts the control system and adjusts it to changing flight conditions and ensures the stability and maneuverability of the aircraft. The wider introduction of a neural network in the operation of UAVs will allow taking into account various factors such as speed, height, temperature, pressure, etc.