Foreign object debris detection based on improved YOLOv5 algorithm

Abstract

Aiming at the small target characteristics of Foreign Object Debris (FOD) on airport runways, a FOD target detection algorithm based on improved YOLOv5 is proposed. Based on the YOLOv5 network, this paper refers to the lightweight and efficient Convolutional Block Attention Module (CBAM), so that the model can focus on important features when detecting objects of different sizes and improve feature extraction ability. The PANet network is improved into BiFPN weighted bidirectional feature pyramid network, which integrates features and strengthens the extraction of deep features of images. Replace the border regression loss function EIoU to improve the regression speed and accuracy. The experimental results show that the improved YOLOv5 target detection algorithm achieves 98.2 % accuracy and 50.2 fps speed while meeting the real-time requirements. Compared with YOLOv5, it has higher detection accuracy and effectively solves the problems of low positioning accuracy and missed detection in the original YOLOv5. Thus, the effectiveness and engineering practicability of the improved YOLOv5 algorithm for the detection of foreign objects on the airport runway at night are verified.