Scholarlink

<p>Precision agriculture requires early and precise fruit disease detection for excellent yields and quality. Examining things together takes time and is simple to make errors, thus smart, automated solutions are needed. Using visual image data, this study proposes a Hybrid Deep Stacking Ensemble (HDSE) architecture for fruit disease classification. The suggested model uses the best aspects of many deep learning architectures, such as EfficientNet, DenseNet, and ResNet, which are all combined through a stacked ensemble technique. Each base learner independently extracts deep spatial and texture features from fruit images. A meta-learner that has been improved via gradient boosting, then combines these features. A fine-grained attention mechanism is added to make it easier to discern the difference between subtle disease symptoms. The model is trained and tested on standard fruit disease Kaggle dataset that include a variety of fruit varieties and disease categories. The HDSE outperforms single-model and bagging-based ensemble techniques in accuracy (98.4%), precision (96.8%), and recall (96.5%). In recognizing comparable illness signs, the recommended strategy is more generalizable and robust. Stacking and fine-grained feature attention modules function well in ablation studies. Modern, mobile-based precision agriculture systems might employ the HDSE architecture to help farmers decrease crop loss in real time.</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

Hybrid Deep Stacking Ensemble for Accurate Fruit Disease Classification in Precision Agriculture

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • A. Kavitha,
  • V. Ravichandran,
  • A. N. Duraivel,
  • S. Mohan

摘要

Precision agriculture requires early and precise fruit disease detection for excellent yields and quality. Examining things together takes time and is simple to make errors, thus smart, automated solutions are needed. Using visual image data, this study proposes a Hybrid Deep Stacking Ensemble (HDSE) architecture for fruit disease classification. The suggested model uses the best aspects of many deep learning architectures, such as EfficientNet, DenseNet, and ResNet, which are all combined through a stacked ensemble technique. Each base learner independently extracts deep spatial and texture features from fruit images. A meta-learner that has been improved via gradient boosting, then combines these features. A fine-grained attention mechanism is added to make it easier to discern the difference between subtle disease symptoms. The model is trained and tested on standard fruit disease Kaggle dataset that include a variety of fruit varieties and disease categories. The HDSE outperforms single-model and bagging-based ensemble techniques in accuracy (98.4%), precision (96.8%), and recall (96.5%). In recognizing comparable illness signs, the recommended strategy is more generalizable and robust. Stacking and fine-grained feature attention modules function well in ablation studies. Modern, mobile-based precision agriculture systems might employ the HDSE architecture to help farmers decrease crop loss in real time.