特征工程结合HRNet网络的景观分类研究  

Research on Landscape Classification Using Feature Engineering Combined with HRNet Network

作　　者：崔金敏 CUI Jinmin(Anhui Zhongao Institute of Technology,Hefei Anhui 230001,China)

机构地区：[1]安徽中澳科技职业学院,安徽合肥230001

出　　处：《鞍山师范学院学报》2025年第2期61-68,共8页Journal of Anshan Normal University

基　　金：2023年省级质量工程技能大师工作室(2023JNDS059);2024年省级质量工程一流核心金课《装饰构造与施工》;2024年度安徽省教育厅科学研究重点科研项目(2024AH052707).

摘　　要：针对传统景观分类准确率不高的问题,提出一种特征工程结合HRNet网络的景观分类方法.首先对绿地植被特征、绿地植被物候特征、绿地植被纹理特征等进行提取,然后结合标准假彩色影像特征作为HRNet网络的输入,对特征进行融合,并实现遥感图像中景观的分类.结果表明,景观绿地植被特征、植被物候特征、植被纹理特征与标准假彩色影像特征相融合,能够有效提高特征丰富度,改善景观绿地植被的分类性能;HRNet网络的损失函数优化与网络训练策略在保证网络训练有效性的同时,能够有效提高网络训练的速率;特征工程与HRNet网络相结合,使得遥感图像中景观分类性能明显提高.基于上述结果,说明以上方案能实现对景观绿地植被的准确分类.A landscape classification method combining feature engineering and HRNet network is proposed to address the problem of low accuracy in traditional landscape classification.Firstly,the characteristics of green vegetation,phenological features of green vegetation,and texture features of green vegetation are extracted.Then,standard false color image features are combined as inputs to the HRNet network to fuse the features and achieve landscape classification in remote sensing images.The results indicate that the integration of vegetation characteristics,phenological features,texture features,and standard false color image features in landscape green spaces can effectively improve feature richness and enhance the classification performance of landscape green space vegetation.The optimization of loss function and network training strategy for HRNet network can effectively improve the training speed while ensuring the effectiveness of network training.The combination of feature engineering and HRNet network significantly improves the landscape classification performance in remote sensing images.Based on the above results,it indicates that the above scheme can achieve accurate classification of landscape green space vegetation.

关 键 词：HRNet网络 绿地植被纹理特征 LBP算法 特征工程 景观分类 准确率 

分 类 号：S771.8[农业科学—森林工程]