Scholarlink

<p>Solidago canadensis is considered a highly invasive weed, posing a significant threat to socio-economic interests, natural ecosystems, and biodiversity. Timely detection of this species is crucial for maintaining biodiversity and ecological balance. Despite advances in deep learning methods for object detection, challenges remain in handling complex weed backgrounds, diverse morphological appearances across different growth stages, and limited sample availability. This study introduces a novel object detection method named Advanced Convolutional Block Attention YOLO (ACBA-YOLO), specifically tailored for Solidago canadensis detection. Built upon the YOLO11 architecture and inspired by attention mechanism methodologies, we have incorporated an improved CBA module and substituted the original SiLU activation function with the AReLU function to enhance its performance. By strategically fusing limited attention mechanism modules, the proposed approach substantially improves localization and classification accuracy, thereby yielding more precise detection results for Solidago canadensis. A custom Solidago canadensis dataset was built, and extensive data augmentation was performed to ensure the model’s generalization capability and robustness across diverse scenarios representing different growth cycles and geographical regions. The experimental results demonstrate that the proposed ACBA-YOLO achieves an improvement of 0.9% in <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\text {AP}_{50}^{val}\)</EquationSource> </InlineEquation> and a 1.5% enhancement in <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\text {AP}_{50:95}^{val}\)</EquationSource> </InlineEquation> compared to YOLO11n, with only a marginal increase in computational cost. Compared to current mainstream object detection algorithms, ACBA-YOLO exhibits superior performance, confirming its effectiveness in Solidago canadensis detection. This paper not only proposes an efficient and accurate identification method for Solidago canadensis but also provides novel insights and directions for the field of weed recognition.</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

Automatically detect Solidago canadensis using an improved attention mechanism network

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Hao Du,
  • Hejia Gao,
  • Changyin Sun

摘要

Solidago canadensis is considered a highly invasive weed, posing a significant threat to socio-economic interests, natural ecosystems, and biodiversity. Timely detection of this species is crucial for maintaining biodiversity and ecological balance. Despite advances in deep learning methods for object detection, challenges remain in handling complex weed backgrounds, diverse morphological appearances across different growth stages, and limited sample availability. This study introduces a novel object detection method named Advanced Convolutional Block Attention YOLO (ACBA-YOLO), specifically tailored for Solidago canadensis detection. Built upon the YOLO11 architecture and inspired by attention mechanism methodologies, we have incorporated an improved CBA module and substituted the original SiLU activation function with the AReLU function to enhance its performance. By strategically fusing limited attention mechanism modules, the proposed approach substantially improves localization and classification accuracy, thereby yielding more precise detection results for Solidago canadensis. A custom Solidago canadensis dataset was built, and extensive data augmentation was performed to ensure the model’s generalization capability and robustness across diverse scenarios representing different growth cycles and geographical regions. The experimental results demonstrate that the proposed ACBA-YOLO achieves an improvement of 0.9% in \(\text {AP}_{50}^{val}\) and a 1.5% enhancement in \(\text {AP}_{50:95}^{val}\) compared to YOLO11n, with only a marginal increase in computational cost. Compared to current mainstream object detection algorithms, ACBA-YOLO exhibits superior performance, confirming its effectiveness in Solidago canadensis detection. This paper not only proposes an efficient and accurate identification method for Solidago canadensis but also provides novel insights and directions for the field of weed recognition.