Scholarlink

<p>Plant diseases pose a critical challenge to global agricultural productivity, affecting crop yield, food safety, and economic sustainability. Early detection of diseases enables timely intervention and reduces dependence on pesticides, lowering production costs and minimizing crop losses. Among the earliest signs of disease are small lesions that form on plant leaves. These fine-grained symptoms are difficult to detect due to weak visual cues and high sensitivity to environmental factors such as noise, lighting variation, and occlusion in uncontrolled field conditions. This paper presents ACAM-EfficientDet, a deep learning-based detection framework designed to address these challenges. The model introduces three core components: (1) an Adaptive Contextual Attention Module (ACAM) for fusing global and local spatial features; (2) a Cross-Scale Context-Aware BiFPN (CSCA-BiFPN) for preserving detail and enhancing multi-scale lesion detection; and (3) a multi-resolution enhancement module to improve robustness under environmental variability. Experiments were conducted on the PlantVillage dataset and a curated Small Lesion Dataset (SLD). The model achieved a mean Average Precision (mAP) of 93.21% at IoU 0.75, outperforming Faster R-CNN, YOLOv7, and EfficientDet-D4 by 2.9–8.7%. Small lesion recall improved by 4.3%, with an average inference speed of 31 ms. Results confirm that ACAM-EfficientDet delivers accurate, efficient, and robust detection suitable for time-sensitive and resource-constrained agricultural 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

ACAM-EfficientDet: a robust deep learning framework for detecting small lesions on plant leaves

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Azeddine Aissaoui,
  • Imene Aloui,
  • Samir Boudibi,
  • Zine-Eddine Khomri,
  • Abdelhamid Foughalia

摘要

Plant diseases pose a critical challenge to global agricultural productivity, affecting crop yield, food safety, and economic sustainability. Early detection of diseases enables timely intervention and reduces dependence on pesticides, lowering production costs and minimizing crop losses. Among the earliest signs of disease are small lesions that form on plant leaves. These fine-grained symptoms are difficult to detect due to weak visual cues and high sensitivity to environmental factors such as noise, lighting variation, and occlusion in uncontrolled field conditions. This paper presents ACAM-EfficientDet, a deep learning-based detection framework designed to address these challenges. The model introduces three core components: (1) an Adaptive Contextual Attention Module (ACAM) for fusing global and local spatial features; (2) a Cross-Scale Context-Aware BiFPN (CSCA-BiFPN) for preserving detail and enhancing multi-scale lesion detection; and (3) a multi-resolution enhancement module to improve robustness under environmental variability. Experiments were conducted on the PlantVillage dataset and a curated Small Lesion Dataset (SLD). The model achieved a mean Average Precision (mAP) of 93.21% at IoU 0.75, outperforming Faster R-CNN, YOLOv7, and EfficientDet-D4 by 2.9–8.7%. Small lesion recall improved by 4.3%, with an average inference speed of 31 ms. Results confirm that ACAM-EfficientDet delivers accurate, efficient, and robust detection suitable for time-sensitive and resource-constrained agricultural environments.