Scholarlink

<p>Maize disease detection in natural environments presents multiple challenges, including weak target features, background noise interference, and high model complexity. To address these issues, we propose a spatial multi-scale efficient real-time detector (SMSERT-DETR). The network integrates a SwConv-ResNet backbone to reduce redundant computation, along with an efficient attention module that enhances the detection of weak feature regions and improves recognition of subtle disease symptoms. Furthermore, we introduce a novel spatial multi-scale feature fusion architecture that suppresses irrelevant background noise and captures multi-scale target features, thereby further improving the model’s feature extraction capabilities. Experimental results on the PlantVillage maize disease dataset demonstrate that our proposed method achieves a precision of 93.6%, recall of 85.7%, mAP50 of 92%, and mAP50:95 of 77%, significantly outperforming other common object detection algorithms on the same dataset. Additionally, SMSERT-DETR achieves an 18.4% reduction in GFLOPS and a 33% reduction in parameter count, enabling efficient deployment in resource-constrained environments.</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

SMSERT-DETR: spatial multi-scale efficient real-time detector for maize disease detection

  • 期刊论文

  • 发表日期：

    2025年5月24日

查看全文
  • Yanan Liu,
  • Fei Yan,
  • Siyu Li,
  • Lan Liu,
  • Yunqing Liu

摘要

Maize disease detection in natural environments presents multiple challenges, including weak target features, background noise interference, and high model complexity. To address these issues, we propose a spatial multi-scale efficient real-time detector (SMSERT-DETR). The network integrates a SwConv-ResNet backbone to reduce redundant computation, along with an efficient attention module that enhances the detection of weak feature regions and improves recognition of subtle disease symptoms. Furthermore, we introduce a novel spatial multi-scale feature fusion architecture that suppresses irrelevant background noise and captures multi-scale target features, thereby further improving the model’s feature extraction capabilities. Experimental results on the PlantVillage maize disease dataset demonstrate that our proposed method achieves a precision of 93.6%, recall of 85.7%, mAP50 of 92%, and mAP50:95 of 77%, significantly outperforming other common object detection algorithms on the same dataset. Additionally, SMSERT-DETR achieves an 18.4% reduction in GFLOPS and a 33% reduction in parameter count, enabling efficient deployment in resource-constrained environments.