论文笔记8：Feature importance analysis for local climate zone classification using a residual convolution

最新推荐文章于 2025-04-03 23:33:09 发布

原创最新推荐文章于 2025-04-03 23:33:09 发布 · 429 阅读

1 ·

CC 4.0 BY-SA版权

LCZ 专栏收录该内容

19 篇文章

订阅专栏

使用残差卷积神经网络对多源数据集进行本地气候区分类，分析特征重要性。研究发现，简单堆叠数据集并不提高效果，类别不平衡影响准确性，小类别的平衡方法有所改善。

Feature importance analysis for local climate zone classification using a residual convolutional neural network with multi-source datasets

Unfortunately, a simple stacking of both datasets together does not provide improvement
在这里插入图片描述

Class Imbalance Effect
It shows that the balancing does not improve the accuracy of the big classes, while for small classes, several balancing methods improve the accuracy. However, of all the four exemplary methods, no one performs the best for all LCZs.
在这里插入图片描述