AFA - DA - Departamento de Aeronáutica
Permanent URI for this community
Browse
Browsing AFA - DA - Departamento de Aeronáutica by Author "Almeida, João Miguel Brito de"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Application of artificial intelligence to the detection of foreign object debris at aerodromes’ movement areaPublication . Almeida, João Miguel Brito deThe present dissertation aims to develop a preliminary low-cost and passive system that detects Foreign Object Debris (FODs) at aerodromes based on computer vision with neural networks. FODs are a twofold problem involving safety risks and high associated costs. Although some systems already exist to detect FODs, these are based on radars, making them expensive. We build a dataset of images to test the viability of this solution, which other authors have already attempted, but the datasets are not publicly available. Moreover, we build a simplified system architecture to capture the images. In parallel, we develop a software pipeline that starts with image capturing scripts and ends in evaluating the models of neural networks we selected. The datasets created result from three different electro-optical sensors: visible, near-infrared and long-wave infrared. From the first, resulted a dataset of 9,260 images, 5,672 from the second and 10,388 from the third. Our approach to this problem is based on supervised learning with image classification and object detection, and we train the models in subsets of the datasets. We choose Xception as the neural network for image classification, achieving a 98.86% accuracy. In the case of object detection, we opt for a single-stage detector – YOLOv3 –, achieving an AP of 91.08%. Finally, we test the same models on new examples and verify a decrease in their performance to 77.92% accuracy for the classifier and 37.49% AP for the detector.