Scholarlink

In this paper we present an image classification model for high-resolution remote sensing images. It combines deep transfer learning with discriminant features. We propose the use of MobileNetV3 and VGG-16 transfer learning to improve the accuracy of scene recognition. We also analyse the feature discrimination capability of both the models and identify the one that is suitable for high resolution remote sensing images. We then carry out experimental analysis of both network and our findings show accuracy of 93.55%. The training and testing sets accuracy has been observed as 98.53% and 93.5%, and the loss value stabilizes at around 17.41% for testing set. Along with the improvement in the accuracy we also tried to provide the explainability of the proposed model that justifies the result. Experimental analysis shows improvement in accuracy, discrimination in features and explanations for reliable high resolution image classification.

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

Analysis of Feature Subspace and Explainability of CNNs for Classification of High Resolution Remote Sensing Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Krishna Gupta,
  • Manisha Parlewar

摘要

In this paper we present an image classification model for high-resolution remote sensing images. It combines deep transfer learning with discriminant features. We propose the use of MobileNetV3 and VGG-16 transfer learning to improve the accuracy of scene recognition. We also analyse the feature discrimination capability of both the models and identify the one that is suitable for high resolution remote sensing images. We then carry out experimental analysis of both network and our findings show accuracy of 93.55%. The training and testing sets accuracy has been observed as 98.53% and 93.5%, and the loss value stabilizes at around 17.41% for testing set. Along with the improvement in the accuracy we also tried to provide the explainability of the proposed model that justifies the result. Experimental analysis shows improvement in accuracy, discrimination in features and explanations for reliable high resolution image classification.