Scholarlink

<p>Foreign object debris (FOD) on runways may cause irreparable damage and detecting FOD in intellectual technology has attracted more and more attention. Due to the limitation of the equipment, the main goal of the FOD detection methods is to obtain as high accuracy as possible by employing the model with as few parameters as possible. In this article, we propose a Lightweight FOD detection model named LF-YOLO, which can improve the detection accuracy of small FOD items and simultaneously decrease the parameters of the implemented model. The proposed LF-YOLO follows the overall framework of YOLOv8 and can be viewed as one of its improved variants. Firstly, to compensate for the loss of information regarding small objects during feature extraction, high-resolution feature maps were incorporated into the detection layer to fuse multiscale features, and the large object detection layer was removed from the model. Secondly, a lightweight backbone with strong feature extraction ability was developed by introducing Lightweight Downsampling Convolution (LDConv) modules. Subsequently, the lightweight Slim-Head was proposed by introducing slim-neck and Group-RepConv with Efficient Channel Attention Mechanism Head (GREHead) modules. Ultimately, to validate the effectiveness of the proposed model, comparative experiments were conducted using the small target FOD dataset. The results demonstrate that the proposed LF-YOLO can achieve better accuracy, over the other state-of-the-art (SOTA) methods with small parameters. (Our code is available at <a href="https://github.com/zflyy/LF-YOLO.git">https://github.com/zflyy/LF-YOLO.git</a>).</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An improved and lightweight small-scale foreign object debris detection model

  • 期刊论文

  • 发表日期：

    2025年4月28日

查看全文
  • Heng Zhang,
  • Wei Fu,
  • Xiaoming Wang,
  • Dong Li,
  • Danchen Zhu,
  • Xingwang Su

摘要

Foreign object debris (FOD) on runways may cause irreparable damage and detecting FOD in intellectual technology has attracted more and more attention. Due to the limitation of the equipment, the main goal of the FOD detection methods is to obtain as high accuracy as possible by employing the model with as few parameters as possible. In this article, we propose a Lightweight FOD detection model named LF-YOLO, which can improve the detection accuracy of small FOD items and simultaneously decrease the parameters of the implemented model. The proposed LF-YOLO follows the overall framework of YOLOv8 and can be viewed as one of its improved variants. Firstly, to compensate for the loss of information regarding small objects during feature extraction, high-resolution feature maps were incorporated into the detection layer to fuse multiscale features, and the large object detection layer was removed from the model. Secondly, a lightweight backbone with strong feature extraction ability was developed by introducing Lightweight Downsampling Convolution (LDConv) modules. Subsequently, the lightweight Slim-Head was proposed by introducing slim-neck and Group-RepConv with Efficient Channel Attention Mechanism Head (GREHead) modules. Ultimately, to validate the effectiveness of the proposed model, comparative experiments were conducted using the small target FOD dataset. The results demonstrate that the proposed LF-YOLO can achieve better accuracy, over the other state-of-the-art (SOTA) methods with small parameters. (Our code is available at https://github.com/zflyy/LF-YOLO.git).