Scholarlink

<p>Accurate detection of pomegranate fruits is crucial for advancing smart harvesting and promoting orchard automation. Current fruit detection faces challenges such as small fruit size, color similarity to the background, and occlusion. To overcome these issues, we have designed the Pomegranate Lightweight You Only Look Once (PL-YOLO) model for precise pomegranate fruit detection in complex environments. The PL-YOLO model features four key innovations. First, the Image Edge Feature Extraction (IEFE) module combines Sobel and standard convolution to extract spatial information, enabling the model to learn rich, deep features. Second, the Feature Pyramid Shared Fusion (FPSF) module addresses information loss during Spatial Pyramid Pooling Fast (SPPF) pooling. Third, the Context Guide Attention Feature Pyramid Network (CGAFPN) efficiently guides the model to focus on critical feature information, enhancing recognition accuracy. Finally, the Detail Enhanced Shared Head (DESH) strengthens the detection head’s detail-capturing ability while reducing parameters. Experimental results show that PL-YOLO achieves a mean Average Precision (mAP) of 93.3% and a Frames Per Second (FPS) of 149.5. It outperforms the baseline model, with a 10.8% increase in FPS, along with respective reductions of 18.6% in parameters, 17.5% in Giga Floating Point Operations (GFLOPs), and 14.8% in model size. Overall, PL-YOLO provides robust technical support for intelligent pomegranate harvesting and effectively advances orchard automation.</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

PL-YOLO: a lightweight method for real-time detection of pomegranates

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Xi Chen,
  • Guohui Wang

摘要

Accurate detection of pomegranate fruits is crucial for advancing smart harvesting and promoting orchard automation. Current fruit detection faces challenges such as small fruit size, color similarity to the background, and occlusion. To overcome these issues, we have designed the Pomegranate Lightweight You Only Look Once (PL-YOLO) model for precise pomegranate fruit detection in complex environments. The PL-YOLO model features four key innovations. First, the Image Edge Feature Extraction (IEFE) module combines Sobel and standard convolution to extract spatial information, enabling the model to learn rich, deep features. Second, the Feature Pyramid Shared Fusion (FPSF) module addresses information loss during Spatial Pyramid Pooling Fast (SPPF) pooling. Third, the Context Guide Attention Feature Pyramid Network (CGAFPN) efficiently guides the model to focus on critical feature information, enhancing recognition accuracy. Finally, the Detail Enhanced Shared Head (DESH) strengthens the detection head’s detail-capturing ability while reducing parameters. Experimental results show that PL-YOLO achieves a mean Average Precision (mAP) of 93.3% and a Frames Per Second (FPS) of 149.5. It outperforms the baseline model, with a 10.8% increase in FPS, along with respective reductions of 18.6% in parameters, 17.5% in Giga Floating Point Operations (GFLOPs), and 14.8% in model size. Overall, PL-YOLO provides robust technical support for intelligent pomegranate harvesting and effectively advances orchard automation.