Scholarlink

To address the challenges of Remote Sensing Scene Classification (RSSC), including considerable classes, intricate spatial data, intra-class variability, and high inter-class similarity, Lightweight Heterogeneous Neural Network (LH-Net) is proposed. LH-Net facilitates feature extraction across diverse dimensions, including spatial information, by leveraging HetConv. Subsequently, the Heterogeneous Block replaces the DSC in MobileNetV1. Evaluation on the UCMerced_LandUse dataset demonstrates LH-Net’s superior performance over traditional CNN models in RSSC, achieving significantly improved accuracy with only a modest increase in FLOPs and parameters.

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

LH-Net: A Lightweight Heterogeneous Convolutional Network for RSSC

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhenxi Gao

摘要

To address the challenges of Remote Sensing Scene Classification (RSSC), including considerable classes, intricate spatial data, intra-class variability, and high inter-class similarity, Lightweight Heterogeneous Neural Network (LH-Net) is proposed. LH-Net facilitates feature extraction across diverse dimensions, including spatial information, by leveraging HetConv. Subsequently, the Heterogeneous Block replaces the DSC in MobileNetV1. Evaluation on the UCMerced_LandUse dataset demonstrates LH-Net’s superior performance over traditional CNN models in RSSC, achieving significantly improved accuracy with only a modest increase in FLOPs and parameters.