Scholarlink

<p>Rice is a staple food crop for more than half of the world’s population, and its yield is highly responsive to well-balanced mineral nutrition, especially nitrogen, phosphorus, and potassium. Their deficiencies impair plant growth and yield, but traditional diagnosis through visual inspection is subjective and impractical for large-scale applications. To address this, we propose a lightweight deep learning model based on MobileNetV2, enhanced with a convolutional block attention module (CBAM), to automatically classify rice leaf nutrient deficiencies. Leveraging transfer learning with pretrained weights enables high accuracy even with a relatively small dataset of 1679 images. The CBAM further improves discriminative feature extraction by focusing on relevant spatial and channel-wise information. Our model achieves a test accuracy of 98.81% and a macro-average AUC of 0.9999, with minimal computational overhead suitable for deployment on resource-constrained devices such as smartphones and drones. This approach offers an accurate, scalable, and real-time solution for precision fertilizer management and rapid diagnosis in modern agriculture.</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

Transfer learning-based lightweight MobileNetV2 with CBAM attention for rice leaf nutrient deficiency detection

  • 期刊论文

  • 发表日期：

    2025年9月20日

查看全文
  • Nitya Ranjan Manihira,
  • Prabira Kumar Sethy

摘要

Rice is a staple food crop for more than half of the world’s population, and its yield is highly responsive to well-balanced mineral nutrition, especially nitrogen, phosphorus, and potassium. Their deficiencies impair plant growth and yield, but traditional diagnosis through visual inspection is subjective and impractical for large-scale applications. To address this, we propose a lightweight deep learning model based on MobileNetV2, enhanced with a convolutional block attention module (CBAM), to automatically classify rice leaf nutrient deficiencies. Leveraging transfer learning with pretrained weights enables high accuracy even with a relatively small dataset of 1679 images. The CBAM further improves discriminative feature extraction by focusing on relevant spatial and channel-wise information. Our model achieves a test accuracy of 98.81% and a macro-average AUC of 0.9999, with minimal computational overhead suitable for deployment on resource-constrained devices such as smartphones and drones. This approach offers an accurate, scalable, and real-time solution for precision fertilizer management and rapid diagnosis in modern agriculture.