Scholarlink

This paper presents RicNet, an optimized convolutional neural network (CNN) architecture designed specifically for detection of diseases on leafs of rice. RicNet’s efficacy was evaluated against traditional CNN models such as InceptionV3, DenseNet121 and VGG16, and as well as the specialized SE_SPnet, through comprehensive performance metrics including specificity, f1-score, recall, accuracy, precision, F1-score. Our unique approach focuses on varying dropout configurations within RicNet to determine their influence on model performance in identifying diseases like bacterial blight, blast, brown spot, and tungro. Experimental results demonstrate that RicNet, with tailored dropout rates, significantly outperforms conventional models, achieving up to 99.41% accuracy and maintaining high levels of precision and specificity. In particular, the configuration with dropout rates of 0.2 at both evaluated layers showcased superior performance, affirming the impact of fine-tuned dropout on enhancing the robustness and reliability of disease detection networks. The findings suggest that the specialized RicNet architecture not only advances the state of agricultural disease detection but also offers potential applications in other domains requiring precise and reliable image-based classification. This study underscores the importance of architectural optimizations in CNNs and proposes future lines of inquiry into further enhancements for disease detection systems.

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

RicNet: Enhancing Rice Leaf Disease Detection Through Optimized Convolutional Neural Network Architectures and Dropout Configurations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vijay Katkar,
  • Shrihari Khatawkar,
  • Suyash Jadhav,
  • Rohit Malame,
  • Manoj Rathod,
  • Abhijeet Urunkar

摘要

This paper presents RicNet, an optimized convolutional neural network (CNN) architecture designed specifically for detection of diseases on leafs of rice. RicNet’s efficacy was evaluated against traditional CNN models such as InceptionV3, DenseNet121 and VGG16, and as well as the specialized SE_SPnet, through comprehensive performance metrics including specificity, f1-score, recall, accuracy, precision, F1-score. Our unique approach focuses on varying dropout configurations within RicNet to determine their influence on model performance in identifying diseases like bacterial blight, blast, brown spot, and tungro. Experimental results demonstrate that RicNet, with tailored dropout rates, significantly outperforms conventional models, achieving up to 99.41% accuracy and maintaining high levels of precision and specificity. In particular, the configuration with dropout rates of 0.2 at both evaluated layers showcased superior performance, affirming the impact of fine-tuned dropout on enhancing the robustness and reliability of disease detection networks. The findings suggest that the specialized RicNet architecture not only advances the state of agricultural disease detection but also offers potential applications in other domains requiring precise and reliable image-based classification. This study underscores the importance of architectural optimizations in CNNs and proposes future lines of inquiry into further enhancements for disease detection systems.