Estudo de viabilidade da análise automática através de imagens de pistas de pousos e decolagens quanto à resistência à derrapagem

Abstract

Air transportation is an activity of key importance for the economic development of regions connected by airfields. In view of this, the need to monitor the operational safety of runways (RWY) offered to aircraft travelling on them is vital for these activities. These include safety in terms of skid resistance. However, traditional methods can be costly in terms of time, financial and human resources. This paper studies the feasibility of developing a Convolutional Neural Network (CNN) model capable of classifying lane segments by images in terms of operational safety in terms of skid resistance using the coefficient of friction. This classification is based on the accumulation of rubber and its impact on the coefficient of friction. The aim is to obtain a tool to help aerodrome operators make decisions. The study uses grey-scale images collected from a Brazilian airport runway via Google Earth Pro and technical data obtained from official ANAC reports. In addition, data processing and analysis methods are investigated to generate a data set that contributes to the development of a reliable model with low computational costs. These include the K-means algorithm and Pearson’s correlation analysis. The model demonstrated the feasibility of developing a model capable of adequately classifying processed images, reinforcing the potential of this approach for the (RWY) studied. The K-means algorithm with K=3 using the complete dataset with the segmented images showing the area of rubber accumulation was the most efficient among the other experiments carried out. However, it reinforces the importance of creating a balanced and data-rich dataset, preventing overfitting and other limitations in the modelling.