Aircraft Part Recognition from Drones' Perspective Based on Semantic Segmentation

Utilizing drones for aircraft surface inspection is a convenient and efficient approach. However, drones may encounter interference en route to the predetermined starting point of flight, leading to positioning deviations. Therefore, to assist drones in achieving more accurate positioning, this pape...

Full description

Saved in:
Bibliographic Details
Published in2024 IEEE 7th International Conference on Electronic Information and Communication Technology (ICEICT) pp. 1288 - 1292
Main Authors Xiang, Boqing, Chang, Haiqing, Ye, Yiting
Format Conference Proceeding
LanguageEnglish
Published IEEE 31.07.2024
Subjects
Accuracy
Adaptation models
Aerospace electronics
Aircraft
Aircraft manufacture
Atmospheric modeling
Criss-Cross Attention
drones
Semantic segmentation
YOLOv5-seg
Online AccessGet full text

Cover

More Information
Summary:Utilizing drones for aircraft surface inspection is a convenient and efficient approach. However, drones may encounter interference en route to the predetermined starting point of flight, leading to positioning deviations. Therefore, to assist drones in achieving more accurate positioning, this paper proposes an aircraft part recognition method from drones' perspective based on semantic segmentation. This method aids in drone positioning by identifying parts of the aircraft such as the fuselage, wings, and engines. The YOLOv5-seg semantic segmentation model is employed for target segmentation and recognition. The Criss-Cross Attention module is added to enhance the recognition effect of aircraft parts, and Inner-IoU loss is applied to replace the CIoU loss to accelerate convergence. The results indicate that the method employed in this study achieves a mAP@0.5 of 92%, demonstrating a high level of accuracy in the recognition of aircraft parts, thus possessing significant practical application value.
ISSN:2836-7782
DOI:10.1109/ICEICT61637.2024.10671067