Scholarlink

This take a look at offers a comparative evaluation of deep gaining knowledge of models, which includes Convolutional Neural Networks (CNNs), AlexNet, and ResNet-50, for the detection and class of automobiles in satellite imagery, sourced from Google Earth. The research investigates the efficacy of those models in recognizing photos of motors, ships, and airplanes. Each model underwent schooling and assessment, related to preprocessing steps like standardization, normalization, and one-warm encoding. RMSprop, an optimizer used, yielded the highest accuracy. The fashions have been assessed based totally on their accuracy and loss values, with ResNet-50 demonstrating the highest accuracy of 99.39% on education records and ninety seven.51% on validation information. no matter being computationally intensive, ResNet-50's superior overall performance justified the fees. The findings underscore the capacity of these fashions for car recognition responsibilities in satellite tv for pc imagery, paving the manner for destiny studies on superior information augmentation, hyperparameter tuning, and complicated classification troubles.

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

Transportation Vehicle Recognition in Satellite Imagery via Convolutional Neural Networks and Transfer Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Valliappan Raman,
  • Putra Sumari,
  • M. Prabhavathy,
  • R. Akshaya

摘要

This take a look at offers a comparative evaluation of deep gaining knowledge of models, which includes Convolutional Neural Networks (CNNs), AlexNet, and ResNet-50, for the detection and class of automobiles in satellite imagery, sourced from Google Earth. The research investigates the efficacy of those models in recognizing photos of motors, ships, and airplanes. Each model underwent schooling and assessment, related to preprocessing steps like standardization, normalization, and one-warm encoding. RMSprop, an optimizer used, yielded the highest accuracy. The fashions have been assessed based totally on their accuracy and loss values, with ResNet-50 demonstrating the highest accuracy of 99.39% on education records and ninety seven.51% on validation information. no matter being computationally intensive, ResNet-50's superior overall performance justified the fees. The findings underscore the capacity of these fashions for car recognition responsibilities in satellite tv for pc imagery, paving the manner for destiny studies on superior information augmentation, hyperparameter tuning, and complicated classification troubles.