Scholarlink

<p>X-ray security inspection is critical for maintaining public safety and transportation security. However, traditional manual inspection methods are often ineffective due to the challenges posed by complex backgrounds and severe occlusions in X-ray images, resulting in false positives and negatives. This study proposes an enhanced object detection framework based on the YOLOv8 model to address these challenges. Key improvements include the integration of the ADown downsampling module to reduce computational complexity while enhancing detection accuracy and the incorporation of Deformable Convolutional Networks v2 (DCNv2) to improve deformable feature extraction. To strengthen feature representation, the Spatial Pyramid Pooling-Fast with ReLU and Efficient Local Attention (SPPF_RE) module is introduced to effectively integrate global and local features. Additionally, the Dynamic Head (DyHead) module is employed to enhance detection in complex backgrounds, while the Pixels-IOU (PIoU) loss function improves the detection accuracy of rotated objects. Experimental results on the OPIXray and HIXray datasets demonstrate that the proposed framework significantly outperforms the baseline model, achieving notable improvements in detection accuracy. The code can be accessed via the following link: <a href="https://github.com/Guanfj2024/x-ray-detection.git">https://github.com/Guanfj2024/x-ray-detection.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 YOLOv8 model for prohibited item detection with deformable convolution and dynamic head

  • 期刊论文

  • 发表日期：

    2025年3月17日

查看全文
  • Fangjing Guan,
  • Heng Zhang,
  • Xiaoming Wang

摘要

X-ray security inspection is critical for maintaining public safety and transportation security. However, traditional manual inspection methods are often ineffective due to the challenges posed by complex backgrounds and severe occlusions in X-ray images, resulting in false positives and negatives. This study proposes an enhanced object detection framework based on the YOLOv8 model to address these challenges. Key improvements include the integration of the ADown downsampling module to reduce computational complexity while enhancing detection accuracy and the incorporation of Deformable Convolutional Networks v2 (DCNv2) to improve deformable feature extraction. To strengthen feature representation, the Spatial Pyramid Pooling-Fast with ReLU and Efficient Local Attention (SPPF_RE) module is introduced to effectively integrate global and local features. Additionally, the Dynamic Head (DyHead) module is employed to enhance detection in complex backgrounds, while the Pixels-IOU (PIoU) loss function improves the detection accuracy of rotated objects. Experimental results on the OPIXray and HIXray datasets demonstrate that the proposed framework significantly outperforms the baseline model, achieving notable improvements in detection accuracy. The code can be accessed via the following link: https://github.com/Guanfj2024/x-ray-detection.git