Scholarlink

The paper presents a study of complex systems modeling using neural modeling on the PlantVillage dataset. Using complex data of agricultural systems, deep neural networks of the ResNet50 and DenseNet121 models were built and the efficiency of models built on the basis of Adam and RMSProp optimizers was shown. A comparative analysis of the accuracy metrics and errors of models with the Adam and RMSProp optimizers was carried out. A full forecasting report was built for the task of classifying and identifying data for the corresponding classes of plant diseases. Estimates of a single metric F1 were obtained, which combines completeness and accuracy using the harmonic mean. The error matrix of the obtained forecasts was analyzed in quantitative and percentage terms for accurately and approximately classified classes. Using ROC curves for the created models, the efficiency and quality of the ResNet50 and DenseNet121 models were shown. To improve the quality of neural network optimization models, modern deep learning optimizers based on the Adam, RMSProp and SGD + Momentum optimizers were used. The problems of optimization of nonlinear modeling for agricultural systems are considered and the EfficientNetB4 model with Adam and RMSProp optimizers with preliminary data processing is built. The advantage of the EfficientNetB4 model built with the Adam optimizer is shown in comparison with other models using the RMSProp and SGD + Momentum optimizers.

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

Optimization of Transfer Learning Models for Forecasting Complex Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Chyngyz Sabitov,
  • Zhamin Sheishenov,
  • Bakyt Omuralieva,
  • Ainura Dyikanova,
  • Baratbek Sabitov

摘要

The paper presents a study of complex systems modeling using neural modeling on the PlantVillage dataset. Using complex data of agricultural systems, deep neural networks of the ResNet50 and DenseNet121 models were built and the efficiency of models built on the basis of Adam and RMSProp optimizers was shown. A comparative analysis of the accuracy metrics and errors of models with the Adam and RMSProp optimizers was carried out. A full forecasting report was built for the task of classifying and identifying data for the corresponding classes of plant diseases. Estimates of a single metric F1 were obtained, which combines completeness and accuracy using the harmonic mean. The error matrix of the obtained forecasts was analyzed in quantitative and percentage terms for accurately and approximately classified classes. Using ROC curves for the created models, the efficiency and quality of the ResNet50 and DenseNet121 models were shown. To improve the quality of neural network optimization models, modern deep learning optimizers based on the Adam, RMSProp and SGD + Momentum optimizers were used. The problems of optimization of nonlinear modeling for agricultural systems are considered and the EfficientNetB4 model with Adam and RMSProp optimizers with preliminary data processing is built. The advantage of the EfficientNetB4 model built with the Adam optimizer is shown in comparison with other models using the RMSProp and SGD + Momentum optimizers.