Research on airport runway foreign object debris detection technology based on improved YOLOv7

Abstract

To address the low accuracy of existing algorithms for detecting foreign object debris (FOD) on airport runways, a FOD detection algorithm is proposed using improved YOLOv7. Firstly, the SimAM attention mechanism is integrated into the ELAN module of YOLOv7, which can extract the feature information of small targets more effectively without adding the network parameters while inferring the 3D attention weight for the feature map. Secondly, the PANet structure in YOLOv7 feature fusion network is replaced by BiFPN structure to realize multi-scale feature fusion and cross-scale connection between different layers. Eventually, the bounding box Loss function of YOLOv7 is replaced with SIOU Loss for the purpose of improving the accuracy and speed of bounding box regression. The improved algorithm is tested on the selfmade FOD dataset, as well as the experimental results depict that the average accuracy rate is 93.5%, which is 4.8% higher than that before the improvement, and satisfactory experimental results are obtained, which can meet the FOD detection tasks of a large number of small targets.