Scholarlink

<p>The efficient recognition of container number in a complex natural environment is of great significance in container transportation process. However, the number of parameters and calculation amount of the algorithms based on deep learning are huge, making it difficult to apply them to the low-cost equipment. In response to this challenge, we propose an improved lightweight algorithm ACCR-YOLOv7 based on YOLOv7-Tiny, which is suitable for container number recognition in various complex environments. First, we propose a lightweight extended efficient layer aggregation networks (ELAN), namely G-ELAN, to enhance the feature extraction capability. Secondly, an efficient Spatial Pyramid Pooling Module (ESPPM) is designed to increase the receptive field of the network for detecting targets. In addition, we redesign the original neck structure by reducing the output layer that is insensitive to large targets and replacing all <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13198_2025_2719_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="39" /> </InlineMediaObject> <EquationSource Format="TEX">\(3\times 3\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>3</mn> <mo>×</mo> <mn>3</mn> </mrow> </math></EquationSource> </InlineEquation> convolutions with lightweight GSconv convolutions, which reduces the number of parameters while improving the accuracy. Finally, a Simple, Parameter-Free Attention Module (SimAM) is embedded into the backbone network and the neck network to eliminate the background noise and conflicting information. In the experimental results, our improved network achieved impressive mAP scores of 99.2% and 97.33% on the container number number dataset and letter dataset, respectively. This is an improvement of 1.73% and 0.15% compared to the original YOLOv7-Tiny network. More importantly, the Params, GFlops, and Time are reduced by 27.48%, 41.70%, and 37.17%, respectively.</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

Lightweight container number recognition based on deep learning

  • 期刊论文

  • 发表日期：

    2025年1月25日

查看全文
  • Tao Liu,
  • Xianqing Wu,
  • Fang Li

摘要

The efficient recognition of container number in a complex natural environment is of great significance in container transportation process. However, the number of parameters and calculation amount of the algorithms based on deep learning are huge, making it difficult to apply them to the low-cost equipment. In response to this challenge, we propose an improved lightweight algorithm ACCR-YOLOv7 based on YOLOv7-Tiny, which is suitable for container number recognition in various complex environments. First, we propose a lightweight extended efficient layer aggregation networks (ELAN), namely G-ELAN, to enhance the feature extraction capability. Secondly, an efficient Spatial Pyramid Pooling Module (ESPPM) is designed to increase the receptive field of the network for detecting targets. In addition, we redesign the original neck structure by reducing the output layer that is insensitive to large targets and replacing all \(3\times 3\) 3 × 3 convolutions with lightweight GSconv convolutions, which reduces the number of parameters while improving the accuracy. Finally, a Simple, Parameter-Free Attention Module (SimAM) is embedded into the backbone network and the neck network to eliminate the background noise and conflicting information. In the experimental results, our improved network achieved impressive mAP scores of 99.2% and 97.33% on the container number number dataset and letter dataset, respectively. This is an improvement of 1.73% and 0.15% compared to the original YOLOv7-Tiny network. More importantly, the Params, GFlops, and Time are reduced by 27.48%, 41.70%, and 37.17%, respectively.