Scholarlink

<p>The banana (Musa spp.) is an important fruit crop that is grown all over the world, particularly in tropical and subtropical countries. It plays a significant role in ensuring food security and promoting economic growth in these regions. However, banana plants are vulnerable to nutrient deficiencies, which can adversely affect their growth, productivity, and the quality of the fruit produced, leading to diseases. This study introduces an Input-Reliant Depth-Wise Convolution with Multi-Scale Network (IDWC-MSN), a novel deep-learning architecture for detecting eight major banana nutrient deficiencies. A public dataset comprising 7214 labeled banana-leaf images captured from farms across Karnataka, India, was employed. Data augmentation mitigated class imbalance. The model integrates adaptive multi-scale pooling, attention mechanisms, and input-reliant depth-wise convolutions, enabling robust feature extraction under varying field conditions, and also increases the accuracy and interpretability of nutrient insufficiency detection. The proposed approach provides multi-scale summarization using adaptive pooling with the behavior of maintained convolutional feature maps. This allows for focused and effective nutrient management techniques for sustainable banana cultivation. The proposed IDWC-MSN achieved 91.34% accuracy, and 0.91 weighted F1-score, outperforming ResNet-50 (71.64%), VGG-16 (62.83%), MobileNet V3 (88.27%), EfficientNet-B0 (89.42%), and SqueezeNet (90.12%). Misclassification analysis revealed manganese and zinc deficiencies as the most confusable due to symptom similarity.</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

A multi-scale deep learning approach for nutrient deficiency detection in banana plants using adaptive feature pooling

  • 期刊论文

  • 发表日期：

    2026年2月21日

查看全文
  • C. K. Sunil,
  • R. Anusha,
  • C. K. Raghavendra,
  • Guowei Dai

摘要

The banana (Musa spp.) is an important fruit crop that is grown all over the world, particularly in tropical and subtropical countries. It plays a significant role in ensuring food security and promoting economic growth in these regions. However, banana plants are vulnerable to nutrient deficiencies, which can adversely affect their growth, productivity, and the quality of the fruit produced, leading to diseases. This study introduces an Input-Reliant Depth-Wise Convolution with Multi-Scale Network (IDWC-MSN), a novel deep-learning architecture for detecting eight major banana nutrient deficiencies. A public dataset comprising 7214 labeled banana-leaf images captured from farms across Karnataka, India, was employed. Data augmentation mitigated class imbalance. The model integrates adaptive multi-scale pooling, attention mechanisms, and input-reliant depth-wise convolutions, enabling robust feature extraction under varying field conditions, and also increases the accuracy and interpretability of nutrient insufficiency detection. The proposed approach provides multi-scale summarization using adaptive pooling with the behavior of maintained convolutional feature maps. This allows for focused and effective nutrient management techniques for sustainable banana cultivation. The proposed IDWC-MSN achieved 91.34% accuracy, and 0.91 weighted F1-score, outperforming ResNet-50 (71.64%), VGG-16 (62.83%), MobileNet V3 (88.27%), EfficientNet-B0 (89.42%), and SqueezeNet (90.12%). Misclassification analysis revealed manganese and zinc deficiencies as the most confusable due to symptom similarity.