北京信息科技大学学报（自然科学版）2023，Vol.38Issue(6)：95-100,6.DOI:10.16508/j.cnki.11-5866/n.2023.06.013
基于深度迁移学习的多标签遥感影像地物分类
Multi-label terrain classification of remote sensing images based on deep transfer learning
张博涵 1徐晓敏1
作者信息
- 1. 北京信息科技大学信息管理学院,北京 100192
- 折叠
摘要
Abstract
In order to effectively monitor the damage of human activities to the natural environment,a multi-label remote sensing image feature classification system based on deep transfer learning was constructed.Based on the deep transfer learning method,5 residual network(ResNet)models with different depths were trained.By analyzing the loss curve,testing the precision and F1 score,the optimal model was determined,and a user-friendly interactive interface system was designed according to the model.The results show that the classification performance of ResNet34 model optimized by deep transfer learning is better than that of other networks;the precision of classification index can reach 96.02％,and F1 score can reach 90.38％.The results can provide decision support for the relevant government management departments to understand the ecological changes and take management measures,which is of positive significance for the sustainable development of ecological environment.关键词
地物分类/深度残差网络/深度迁移学习/卷积神经网络/多标签Key words
terrain classification/deep residual network/deep transfer learning/convolutional neural network/multi-label分类
信息技术与安全科学引用本文复制引用
张博涵,徐晓敏..基于深度迁移学习的多标签遥感影像地物分类[J].北京信息科技大学学报（自然科学版）,2023,38(6):95-100,6.
