Scholarlink

<p>Aircraft skin damage detection is crucial for ensuring flight safety. This article presents an enhanced object detection algorithm tailored for scenarios with low image complexity. Initially, considering the low image complexity inherent in aircraft skin damage data, an additive feature fusion attention mechanism is proposed to enhance the YOLOv7 neck feature fusion approach, aiming at diminishing the computational and parametric loads of the model. Secondly, the Inner-CIoU loss function is enhanced and the dynamic Inner-CIoU loss function is introduced to substitute the original CIoU loss function in YOLOv7. Subsequently, to validate the proposed method, an Aircraft Skin Damage Dataset encompassing five types of damage, with image backgrounds solely comprising the aircraft skin, is generated. Lastly, experimental results demonstrate that the proposed additive feature fusion attention mechanism, tailored for scenarios like Aircraft Skin Damage Dataset with low image complexity, significantly reduces the model parameters without compromising model accuracy. Compared to the YOLOv7, the proposed method improves the detection accuracy by 1.6% and reduces the number of parameters by 19.4%; the enhanced YOLOv7 model is compared with mainstream object detection models to illustrate the superiority of the improved algorithm. The Pascal VOC2007 Database and Aluminum Profile Surface Detection Database are selected as control groups, which are used to further validate the correctness of the proposed theory, namely the additive feature fusion attention mechanism.</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 additive feature fusion attention based on YOLO network for aircraft skin damage detection

  • 期刊论文

  • 发表日期：

    2025年3月15日

查看全文
  • Jun Wu,
  • Yajing Zhang,
  • Tengfei Shan,
  • Zhiwei Xing,
  • Jiusheng Chen,
  • Runxia Guo

摘要

Aircraft skin damage detection is crucial for ensuring flight safety. This article presents an enhanced object detection algorithm tailored for scenarios with low image complexity. Initially, considering the low image complexity inherent in aircraft skin damage data, an additive feature fusion attention mechanism is proposed to enhance the YOLOv7 neck feature fusion approach, aiming at diminishing the computational and parametric loads of the model. Secondly, the Inner-CIoU loss function is enhanced and the dynamic Inner-CIoU loss function is introduced to substitute the original CIoU loss function in YOLOv7. Subsequently, to validate the proposed method, an Aircraft Skin Damage Dataset encompassing five types of damage, with image backgrounds solely comprising the aircraft skin, is generated. Lastly, experimental results demonstrate that the proposed additive feature fusion attention mechanism, tailored for scenarios like Aircraft Skin Damage Dataset with low image complexity, significantly reduces the model parameters without compromising model accuracy. Compared to the YOLOv7, the proposed method improves the detection accuracy by 1.6% and reduces the number of parameters by 19.4%; the enhanced YOLOv7 model is compared with mainstream object detection models to illustrate the superiority of the improved algorithm. The Pascal VOC2007 Database and Aluminum Profile Surface Detection Database are selected as control groups, which are used to further validate the correctness of the proposed theory, namely the additive feature fusion attention mechanism.