Scholarlink

Since it provides a perceptive analysis of numerous environmental and geographic events, satellite image classification is a basic component of remote sensing. This work uses recent deep learning (DL) techniques, including Vision Transformers (ViT), to examine satellite imagery classification. Examining four models in great detail: Vision Transformers (ViT), VGG-16, CNN, and ResNet-50. CNN is here. There are 5631 high-resolution images in the dataset spanning four primary categories. The approach consists of data preparation, augmentation, and adding attention approaches to improve the model performance. With an incredible 99.62% accuracy in terms of testing, the experimental results reveal that the ViT model surpasses all the other models. It also reveals better memory, accuracy, and F-score. ResNet-50, VGG-16, and CNN models demonstrate a quite mediocre degree of performance with an accuracy range of 87–89%. The ViT method offers significant opportunities for use in remote sensing based on great accuracy and durability over earlier versions. The results show that ViT quite correctly picks out geographic correlations in satellite data. The exact and practical results of the research obtained from satellite images will help urban planning, disaster control, and environmental monitoring. Reliable picture categorization directs decisions in resource management, agriculture, and studies on climate change.

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

Enhancing Satellite Image Classification Using Vision Transformers and Deep Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • M. Sangeetha,
  • T. Harikala,
  • LNC. Prakash K,
  • V. Lavanya,
  • Penubaka Balaji,
  • Mary Teresa

摘要

Since it provides a perceptive analysis of numerous environmental and geographic events, satellite image classification is a basic component of remote sensing. This work uses recent deep learning (DL) techniques, including Vision Transformers (ViT), to examine satellite imagery classification. Examining four models in great detail: Vision Transformers (ViT), VGG-16, CNN, and ResNet-50. CNN is here. There are 5631 high-resolution images in the dataset spanning four primary categories. The approach consists of data preparation, augmentation, and adding attention approaches to improve the model performance. With an incredible 99.62% accuracy in terms of testing, the experimental results reveal that the ViT model surpasses all the other models. It also reveals better memory, accuracy, and F-score. ResNet-50, VGG-16, and CNN models demonstrate a quite mediocre degree of performance with an accuracy range of 87–89%. The ViT method offers significant opportunities for use in remote sensing based on great accuracy and durability over earlier versions. The results show that ViT quite correctly picks out geographic correlations in satellite data. The exact and practical results of the research obtained from satellite images will help urban planning, disaster control, and environmental monitoring. Reliable picture categorization directs decisions in resource management, agriculture, and studies on climate change.