Scholarlink

<p>To address issues such as rot, disease, and low detection accuracy in eggplant production, this paper proposes an improved detection model based on YOLOv8n, named YOLOv8n-EggplantDiseases (YOLOv8n-ED). The model enhances target detection by redesigning the feature extraction network and optimizing the detection architecture. Specifically, it integrates C2f_CGA and Swin Transformer into the backbone to improve semantic representation and small-object recognition in complex backgrounds. Additionally, the SimAM attention-free mechanism is incorporated to further enhance contextual understanding. The original convolution layers are replaced with the ADown module, and BoTNet is introduced to strengthen target identification. In the Neck, ODConv (omni-dimensional dynamic convolution) and C2f_SimAM modules are employed to improve feature expressiveness. Moreover, the WIoU loss function is adopted to address class imbalance and scale variation, accelerate convergence, and enhance regression accuracy. The proposed YOLOv8n-ED achieves&#xa0;73.2% Precision, 66.5% Recall, 70.4% mAP@.5, and 39.5% mAP@.5-.95, outperforming baseline models by + 2.5% Recall, + 1.9% mAP@.5, and + 1% mAP@.5-.95.</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

YOLOv8n-EggplantDiseases: an enhanced model for accurate disease detection in eggplants

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • Qingan Yao,
  • Congmin Zhang,
  • Yuncong Feng,
  • Xuexiao Wang

摘要

To address issues such as rot, disease, and low detection accuracy in eggplant production, this paper proposes an improved detection model based on YOLOv8n, named YOLOv8n-EggplantDiseases (YOLOv8n-ED). The model enhances target detection by redesigning the feature extraction network and optimizing the detection architecture. Specifically, it integrates C2f_CGA and Swin Transformer into the backbone to improve semantic representation and small-object recognition in complex backgrounds. Additionally, the SimAM attention-free mechanism is incorporated to further enhance contextual understanding. The original convolution layers are replaced with the ADown module, and BoTNet is introduced to strengthen target identification. In the Neck, ODConv (omni-dimensional dynamic convolution) and C2f_SimAM modules are employed to improve feature expressiveness. Moreover, the WIoU loss function is adopted to address class imbalance and scale variation, accelerate convergence, and enhance regression accuracy. The proposed YOLOv8n-ED achieves 73.2% Precision, 66.5% Recall, 70.4% mAP@.5, and 39.5% mAP@.5-.95, outperforming baseline models by + 2.5% Recall, + 1.9% mAP@.5, and + 1% mAP@.5-.95.