Scholarlink

<p>Weather image classification plays a crucial role in weather forecasting, disaster management, and transportation optimization. In this study, we propose a systematic architectural enhancement of multiple pretrained deep learning models—namely ResNet50, EfficientNetB0, EfficientNetB4, and EfficientNetB7—for weather image classification. Our key innovation lies in replacing the Flatten layer with a Global Average Pooling (GAP) layer, followed by a Dense layer with 128 neurons and a Dropout layer (rate = 0.5), aiming to reduce overfitting and improve generalization across models. This unified modification framework allows us to effectively compare the performance impact of architectural tuning across different backbone networks. Evaluated on the WEAPD (Weather Phenomenon Database), our method achieves a classification accuracy of 96%, outperforming existing baselines. These results demonstrate the effectiveness of architecture-level optimization in enhancing the robustness and accuracy of weather classification models, paving the way for more reliable forecasting and environmental decision-making systems.</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

Enhanced weather image classification via architectural optimization of pretrained deep models

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Sekineh Asadi Amiri,
  • Zeynab Davoudi

摘要

Weather image classification plays a crucial role in weather forecasting, disaster management, and transportation optimization. In this study, we propose a systematic architectural enhancement of multiple pretrained deep learning models—namely ResNet50, EfficientNetB0, EfficientNetB4, and EfficientNetB7—for weather image classification. Our key innovation lies in replacing the Flatten layer with a Global Average Pooling (GAP) layer, followed by a Dense layer with 128 neurons and a Dropout layer (rate = 0.5), aiming to reduce overfitting and improve generalization across models. This unified modification framework allows us to effectively compare the performance impact of architectural tuning across different backbone networks. Evaluated on the WEAPD (Weather Phenomenon Database), our method achieves a classification accuracy of 96%, outperforming existing baselines. These results demonstrate the effectiveness of architecture-level optimization in enhancing the robustness and accuracy of weather classification models, paving the way for more reliable forecasting and environmental decision-making systems.