Scholarlink

<p>Crop pest identification is critical for ensuring agricultural productivity and food security. While Convolutional Neural Networks (CNNs) have shown great promise in this domain, their performance is highly dependent on the tedious and often suboptimal process of manual hyperparameter tuning. To overcome this limitation, this paper proposes a novel CNN hyperparameter optimization method based on an enhanced differential evolution algorithm. The core of our approach is a new Memory-based Strategy Selection Successful-History Adaptive Differential Evolution (ML-SHADE) algorithm. The key innovations of ML-SHADE include: (1) a dynamic fuzzy weight system that adaptively adjusts the influence of elite individuals during the mutation phase based on their fitness and the optimization stage; (2) a memory selection mechanism (MSM) that records fitness improvements rather than simple success counts, enabling more informed and probabilistic strategy selection between the classical approach and our new fuzzy-weighted method. We applied the proposed ML-SHADE to optimize a CNN model for identifying ten species of pentatomidae stinkbug pests. Experimental results demonstrate that our method significantly outperforms seven other competing optimization algorithms. The ML-SHADE-optimized CNN achieved a remarkable recognition accuracy of 97.46% on the pest dataset, highlighting its superiority and potential as a robust tool for automated agricultural pest management.</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

Convolutional neural network optimized by successful history-based adaptive differential evolution with memory strategy selection to identify crop pests

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Xiaobing Yu,
  • Yifei Li,
  • Yangchen Lu

摘要

Crop pest identification is critical for ensuring agricultural productivity and food security. While Convolutional Neural Networks (CNNs) have shown great promise in this domain, their performance is highly dependent on the tedious and often suboptimal process of manual hyperparameter tuning. To overcome this limitation, this paper proposes a novel CNN hyperparameter optimization method based on an enhanced differential evolution algorithm. The core of our approach is a new Memory-based Strategy Selection Successful-History Adaptive Differential Evolution (ML-SHADE) algorithm. The key innovations of ML-SHADE include: (1) a dynamic fuzzy weight system that adaptively adjusts the influence of elite individuals during the mutation phase based on their fitness and the optimization stage; (2) a memory selection mechanism (MSM) that records fitness improvements rather than simple success counts, enabling more informed and probabilistic strategy selection between the classical approach and our new fuzzy-weighted method. We applied the proposed ML-SHADE to optimize a CNN model for identifying ten species of pentatomidae stinkbug pests. Experimental results demonstrate that our method significantly outperforms seven other competing optimization algorithms. The ML-SHADE-optimized CNN achieved a remarkable recognition accuracy of 97.46% on the pest dataset, highlighting its superiority and potential as a robust tool for automated agricultural pest management.