面向农业的实用人工智能:用于检测菜叶病的自我监督的注意力框架
Nilavro Das Kabya1, Md Shaifullah Sharafat1, Rahimul Islam Emu1
1Electrical and Computer Engineering, North South University, Dhaka, Bangladesh.
PloS one
|January 16, 2026
概括
一个新的深度学习框架准确地分类了马拉巴菜叶病如Alternaria和草虫. 高效的SimSiam-CBAM-ResNet-50模型为农业应用提供了一个实用的解决方案.
科学领域:
- 农业科学 农业科学
- 计算机科学 计算机科学
- 植物病理学 植物病理学
背景情况:
- 马拉巴尔菜在孟加拉国至关重要,但易受Alternaria叶斑和草虫的感染.
- 准确的疾病鉴定对于保持作物产量和质量至关重要.
研究的目的:
- 开发一个高效和可解释的深度学习框架,用于马拉巴菜叶病的自动分类.
- 在有限的数据条件下评估各种深度学习架构的性能,包括CNN和变压器.
- 用自我监督的预训来解决注释短缺的问题.
主要方法:
- 创建了马拉巴菜图像的精心策划的数据集,分为健康,Alternaria和草类别.
- 像SpinachCNN,Spinach-ResSENet,SpinachViT和SwinV2-Base这样的模型被训练和评估.
- 在未标记的数据上使用SimSiam进行自我监督的预训,然后进行监督的微调.
- 域优化的模型包含了注意力机制 (挤压和刺激,卷积块注意力模块).
主要成果:
- 该SimSiam-CBAM-ResNet-50模型实现了97.31%的测试准确度和高ROC-AUC,证明了对噪声的稳定性.
- 像SwinV2-Base这样的基于变压器的模型显示出略高的精度,但需要大量的预训练和更多的参数.
- 可解释性方法 (Grad-CAM,LayerCAM) 证实了模型的重点是相关的病理区域.
结论:
- 拟议的SimSiam-CBAM-ResNet-50为马拉巴菜疾病检测提供了一个参数高效和可部署的解决方案.
- 自主监督学习有效地缓解了农业数据集中的注释稀缺性.
- 深度学习为精准农业提供了强大的工具,有助于疾病管理和作物保护.
相关概念视频
Light Acquisition
9.4K
In order to produce glucose, plants need to capture sufficient light energy. Many modern plants have evolved leaves specialized for light acquisition. Leaves can be only millimeters in width or tens of meters wide, depending on the environment. Due to competition for sunlight, evolution has driven the evolution of increasingly larger leaves and taller plants, to avoid shading by their neighbors with contaminant elaboration of root architecture and mechanisms to transport water and nutrients.
9.4K
Key Elements for Plant Nutrition
24.0K
Like all living organisms, plants require organic and inorganic nutrients to survive, reproduce, grow and maintain homeostasis. To identify nutrients that are essential for plant functioning, researchers have leveraged a technique called hydroponics. In hydroponic culture systems, plants are grown—without soil—in water-based solutions containing nutrients. At least 17 nutrients have been identified as essential elements required by plants. Plants acquire these elements from the...
24.0K


