Scholarlink

<p>A growing number of Internet of Things (IoT) devices are utilizing mobile CPUs to perform 1D and 2D barcode scanning functions. While the YOLO11 algorithm has achieved promising results in object detection and barcode detection, achieving real-time performance on computation constrained mobile CPU platforms without decreasing barcode detection accuracy remains a challenge. According to the characteristics of barcode detection, this paper proposes a lightweight object detection network named LBC-YOLO11 (Lightweight Barcode YOLO11). By significantly reducing the number of channels in layers of the network, replacing original convolutions in the backbone with depthwise separable convolutions (DSConv), and designing a lightweight detection head, the proposed method substantially decreases the number of parameters and computational complexity without considerable sacrificing barcode detection accuracy. Experimental results conducted on a 2.0GHz MT8788 ARM CPU platform show that, compared to the baseline original YOLO11n model, the proposed method achieves approximately 10<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> and 6<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> improvements running speed in single-core and multi-core scenarios, with memory usage maintained around 95–99 MB, power consumption reduced by 10–17%, and CPU utilization optimized, increasing single-core efficiency while reducing multi-core load. respectively, across multiple public and self-collected datasets, with only minimal declines in recall and precision for barcode detection. The average inference frame rates reach about 45 fps and 56 fps, respectively. These results demonstrate that the proposed algorithm can achieve real-time performance on common ARM CPU-based IoT devices while maintaining effective code detection performance.</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

Real-time barcode detection for embedded IoT systems using lightweight YOLO architecture

  • 期刊论文

  • 发表日期：

    2026年3月28日

查看全文
  • Jiazhen Zhu,
  • Yufan Ye,
  • Changcai Lai,
  • Jie Jiang

摘要

A growing number of Internet of Things (IoT) devices are utilizing mobile CPUs to perform 1D and 2D barcode scanning functions. While the YOLO11 algorithm has achieved promising results in object detection and barcode detection, achieving real-time performance on computation constrained mobile CPU platforms without decreasing barcode detection accuracy remains a challenge. According to the characteristics of barcode detection, this paper proposes a lightweight object detection network named LBC-YOLO11 (Lightweight Barcode YOLO11). By significantly reducing the number of channels in layers of the network, replacing original convolutions in the backbone with depthwise separable convolutions (DSConv), and designing a lightweight detection head, the proposed method substantially decreases the number of parameters and computational complexity without considerable sacrificing barcode detection accuracy. Experimental results conducted on a 2.0GHz MT8788 ARM CPU platform show that, compared to the baseline original YOLO11n model, the proposed method achieves approximately 10 \(\times \) × and 6 \(\times \) × improvements running speed in single-core and multi-core scenarios, with memory usage maintained around 95–99 MB, power consumption reduced by 10–17%, and CPU utilization optimized, increasing single-core efficiency while reducing multi-core load. respectively, across multiple public and self-collected datasets, with only minimal declines in recall and precision for barcode detection. The average inference frame rates reach about 45 fps and 56 fps, respectively. These results demonstrate that the proposed algorithm can achieve real-time performance on common ARM CPU-based IoT devices while maintaining effective code detection performance.