Foreign Object Debris Detection in UAS Based Thermal and RGB Images using AI Assisted Models
Abstract (300 words maximum)
Foreign Object Debris (FOD) is a major concern for aircraft operations. FOD present on an airplane runway can pose a special hazard. FOD presents many dangers because it can cause harm to an airplane. Damages and delays caused by FOD cost the aerospace industry billions of dollars every year . Lots of research has been done on effective ways to identify and locate FOD so it can be removed. This research aims to introduce a new facet of FOD detection using thermal imaging. In this work, we present the use of thermal imaging sensors mounted on Unmanned Aerial Systems (UAS) to quickly identify FOD. Thermal imaging and the You Only Look Once (YOLO) Artificial Intelligence (AI) based object detector are developed and utilized. To train the AI model, raw images are captured. Then these images are manually annotated, to identify visible FOD. The annotated images are used to train and build a custom YOLO model. We then comprehensively tested the pre-defined and custom-built YOLO models to compare the accuracy and speed of FOD detection between image types, as well as the model efficacy. Initial results suggest that thermal imaging in general was more effective for most types of debris compared to RGB imaging. Custom built YOLO models were also found more accurate than the previously existing canned models. Future work in this field will focus on monitoring different variables, such as environmental factors like time of day and temperature. More tests will also be run to implement this architecture in real time to better test the system in an airport setting.
Use of AI Disclaimer
no
Academic department under which the project should be listed
SPCEET – Electrical and Computer Engineering
Primary Investigator (PI) Name
Adeel Khalid
Foreign Object Debris Detection in UAS Based Thermal and RGB Images using AI Assisted Models
Foreign Object Debris (FOD) is a major concern for aircraft operations. FOD present on an airplane runway can pose a special hazard. FOD presents many dangers because it can cause harm to an airplane. Damages and delays caused by FOD cost the aerospace industry billions of dollars every year . Lots of research has been done on effective ways to identify and locate FOD so it can be removed. This research aims to introduce a new facet of FOD detection using thermal imaging. In this work, we present the use of thermal imaging sensors mounted on Unmanned Aerial Systems (UAS) to quickly identify FOD. Thermal imaging and the You Only Look Once (YOLO) Artificial Intelligence (AI) based object detector are developed and utilized. To train the AI model, raw images are captured. Then these images are manually annotated, to identify visible FOD. The annotated images are used to train and build a custom YOLO model. We then comprehensively tested the pre-defined and custom-built YOLO models to compare the accuracy and speed of FOD detection between image types, as well as the model efficacy. Initial results suggest that thermal imaging in general was more effective for most types of debris compared to RGB imaging. Custom built YOLO models were also found more accurate than the previously existing canned models. Future work in this field will focus on monitoring different variables, such as environmental factors like time of day and temperature. More tests will also be run to implement this architecture in real time to better test the system in an airport setting.