A lightweight object detection algorithm for aircraft skin defects based on deep learning

To solve the problem of redundancy in aircraft skin defect detection algorithms, this paper proposes a lightweight algorithm for target detection aircraft skin defects. We propose a new ELAN module, the ELAN-GDW module is used as the backbone module of the main network. A PCBL module is used to further reduce the computational cost and make the convolution more efficient and the memory utilization more sufficient. In addition, the SimAM attention module is introduced into the concat module to enhance the network’s multi-dimensional perception and multi-scale feature extraction capabilities. Experimental results on self-labeled dataset show that: compared with YOLOv7-tiny, GDWP-YOLO reduces the number of parameters by 69% with almost no loss of accuracy with only 4.0 MB of model size, illustrating it has better performance in environments with limited computing resources.