Scholarlink

<p>The field of image interpretations from remotely sensed data has grown considerably in recent years, and remote-sensing image scene categorisation(RSISC) has gained substantial importance. The recent development in Convolutional Neural Networks (CNNs) has resulted in a significant improvement in the classification of remotely sensed scenes. However, because of the intricacy of small objects in high-resolution images, CNNs are unable to effectively achieve the scene classification objectives. To address this challenge, we propose a novel framework, termed Adaptive Inter-Intra Level Convolutional Block Feature-Fusion with Level-Based Attention (AIICBFF-LBA), designed to extract discriminative feature representations from remote-sensing images. The model leverages feature maps extracted from the middle and final convolutional blocks of ResNet50. A channel-spatial attention method is employed at the middle layers, while channel attention is applied at the deeper layers to enhance feature representation. The framework integrates a SoftMax classifier with Stochastic Gradient Descent optimization and Label Smoothing Regularization to achieve robust classification. The AIICBFF-LBA framework was evaluated on three openly accessible benchmark datasets, demonstrating superior or comparable performance in classification accuracy and reported metrics compared to state-of-the-art methods.</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

Remote Sensing Image Scene Classification Using Level-Based Attention of Inter-Intra Convolutional Features with Label Smoothing Regularization

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • G Akila,
  • R Gayathri

摘要

The field of image interpretations from remotely sensed data has grown considerably in recent years, and remote-sensing image scene categorisation(RSISC) has gained substantial importance. The recent development in Convolutional Neural Networks (CNNs) has resulted in a significant improvement in the classification of remotely sensed scenes. However, because of the intricacy of small objects in high-resolution images, CNNs are unable to effectively achieve the scene classification objectives. To address this challenge, we propose a novel framework, termed Adaptive Inter-Intra Level Convolutional Block Feature-Fusion with Level-Based Attention (AIICBFF-LBA), designed to extract discriminative feature representations from remote-sensing images. The model leverages feature maps extracted from the middle and final convolutional blocks of ResNet50. A channel-spatial attention method is employed at the middle layers, while channel attention is applied at the deeper layers to enhance feature representation. The framework integrates a SoftMax classifier with Stochastic Gradient Descent optimization and Label Smoothing Regularization to achieve robust classification. The AIICBFF-LBA framework was evaluated on three openly accessible benchmark datasets, demonstrating superior or comparable performance in classification accuracy and reported metrics compared to state-of-the-art methods.