Scholarlink

<p>Smart harvesting of persimmon fruits is a critical component in advancing its production chain, with the primary challenge being the real-time and accurate detection of fruit. However, existing computer vision methods still struggle to detect persimmon fruits in complex scenes, such as those with complex backgrounds, small target fruits, leaf occlusion and overlapping fruits. In this paper, an improved YOLOv8n model PerD-YOLOv8 (persimmon fruit detection-YOLOv8n) was developed to exploit the detection accuracy of persimmon fruit. Firstly, FasterNet was selected as the backbone feature extraction network of YOLOv8n to improve the feature extraction ability of the model in complex background situations. Secondly, the P2 detection layer was added and fused with the bidirectional feature pyramid network (BiFPN) to replace the path aggregation network-feature pyramid networks (PAN-FPN) structure of YOLOv8n, to improve detection accuracy of small targets and reduce complexity. Lastly, the Wise-Intersection over Union (WIoU) loss function was introduced to optimise the training process, improving fruit localisation accuracy in case of leaf occlusion and fruit overlap. The experimental results show that the precision (P), recall (R), mAP@0.5, and mAP@0.5:0.95 of PerD-YOLOv8 reached 95.2%, 90.4%, 96.3%, and 84.0%, respectively, which displays noticeable advantage compared with Faster R-CNN, SSD, YOLOv3-Tiny, YOLOv4-Tiny, YOLOv5n, YOLOv6, YOLOv7, YOLOv8n, and RT-DETR. The model performs well in detecting persimmon fruits under complex scenarios, which could provide technical support for the development of persimmon picking robots.</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

Persimmon fruit detection in complex scenes based on PerD-YOLOv8

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Haozhuang Liu,
  • Wenjuan Gu,
  • Wenbo Wang,
  • Yang Zou,
  • Hang Yang,
  • Tiangui Li

摘要

Smart harvesting of persimmon fruits is a critical component in advancing its production chain, with the primary challenge being the real-time and accurate detection of fruit. However, existing computer vision methods still struggle to detect persimmon fruits in complex scenes, such as those with complex backgrounds, small target fruits, leaf occlusion and overlapping fruits. In this paper, an improved YOLOv8n model PerD-YOLOv8 (persimmon fruit detection-YOLOv8n) was developed to exploit the detection accuracy of persimmon fruit. Firstly, FasterNet was selected as the backbone feature extraction network of YOLOv8n to improve the feature extraction ability of the model in complex background situations. Secondly, the P2 detection layer was added and fused with the bidirectional feature pyramid network (BiFPN) to replace the path aggregation network-feature pyramid networks (PAN-FPN) structure of YOLOv8n, to improve detection accuracy of small targets and reduce complexity. Lastly, the Wise-Intersection over Union (WIoU) loss function was introduced to optimise the training process, improving fruit localisation accuracy in case of leaf occlusion and fruit overlap. The experimental results show that the precision (P), recall (R), mAP@0.5, and mAP@0.5:0.95 of PerD-YOLOv8 reached 95.2%, 90.4%, 96.3%, and 84.0%, respectively, which displays noticeable advantage compared with Faster R-CNN, SSD, YOLOv3-Tiny, YOLOv4-Tiny, YOLOv5n, YOLOv6, YOLOv7, YOLOv8n, and RT-DETR. The model performs well in detecting persimmon fruits under complex scenarios, which could provide technical support for the development of persimmon picking robots.