Scholarlink

<p>This study aims to develop a real-time object detection algorithm to improve the accuracy and robustness of strawberry ripeness detection for elevated strawberry-picking robots operating in complex environments. To this end, we propose CCEK-YOLOv5s, an improved YOLOv5s-based model. Specifically, an attention mechanism is incorporated into the backbone network, and a bidirectional feature pyramid structure is integrated into the neck network, thereby enhancing the model’s ability to extract discriminative features of strawberries under complex conditions. The dataset was collected from the elevated strawberry planting base of the College of Agriculture at the Henan Institute of Science and Technology (Xinxiang, Henan Province) and augmented to include strawberries at different ripeness levels. Experimental results demonstrate that CCEK-YOLOv5s achieves a mean Average Precision (mAP50) of 97.9%, outperforming YOLOv5s (97.5%), YOLOv4 (94%), YOLOv3 (92%), and Faster R-CNN (97%). In terms of computational complexity, CCEK-YOLOv5s requires only 22.4 GFLOPs, which is substantially lower than Faster R-CNN (180 GFLOPs) and Mask R-CNN (200 GFLOPs), indicating a favorable balance between detection accuracy and efficiency. Comparative detection experiments further confirm the superiority of CCEK-YOLOv5s under varying illumination conditions, occlusion scenarios, and small target sizes. Heatmap visualization analysis using the EigenCAM method reveals that it most accurately reflects the model’s focus on strawberry target regions, validating the efficiency of CCEK-YOLOv5s in feature extraction. These findings highlight the practical value of CCEK-YOLOv5s and provide effective technical support for the intelligent application of elevated strawberry-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

Research on real-time target detection algorithm for strawberry-picking robot based on improved YOLOv5

  • 期刊论文

  • 发表日期：

    2025年10月31日

查看全文
  • Lilu Huang,
  • Haibo Li,
  • Hui Yang,
  • Chengguo Fu,
  • Yaqi Zhang,
  • Fengwei Chang

摘要

This study aims to develop a real-time object detection algorithm to improve the accuracy and robustness of strawberry ripeness detection for elevated strawberry-picking robots operating in complex environments. To this end, we propose CCEK-YOLOv5s, an improved YOLOv5s-based model. Specifically, an attention mechanism is incorporated into the backbone network, and a bidirectional feature pyramid structure is integrated into the neck network, thereby enhancing the model’s ability to extract discriminative features of strawberries under complex conditions. The dataset was collected from the elevated strawberry planting base of the College of Agriculture at the Henan Institute of Science and Technology (Xinxiang, Henan Province) and augmented to include strawberries at different ripeness levels. Experimental results demonstrate that CCEK-YOLOv5s achieves a mean Average Precision (mAP50) of 97.9%, outperforming YOLOv5s (97.5%), YOLOv4 (94%), YOLOv3 (92%), and Faster R-CNN (97%). In terms of computational complexity, CCEK-YOLOv5s requires only 22.4 GFLOPs, which is substantially lower than Faster R-CNN (180 GFLOPs) and Mask R-CNN (200 GFLOPs), indicating a favorable balance between detection accuracy and efficiency. Comparative detection experiments further confirm the superiority of CCEK-YOLOv5s under varying illumination conditions, occlusion scenarios, and small target sizes. Heatmap visualization analysis using the EigenCAM method reveals that it most accurately reflects the model’s focus on strawberry target regions, validating the efficiency of CCEK-YOLOv5s in feature extraction. These findings highlight the practical value of CCEK-YOLOv5s and provide effective technical support for the intelligent application of elevated strawberry-picking robots.