自主监督学习与三边冗余减少,用于城市功能区识别,使用街景图像
Kun Zhao1, Juan Li1, Shuai Xie1
1School of Information and Control Engineering, Qingdao University of Technology, Qingdao 266520, China.
Sensors (Basel, Switzerland)
|March 17, 2025
概括
一个新的自我监督学习框架,三边冗余减少 (Tri-ReD),有效地解决了城市场景分类中的标签稀缺问题. 它通过从未标记的街景图像中学习基本表示来实现最先进的结果.
科学领域:
- 计算机视觉 计算机视觉
- 机器学习 机器学习
- 城市分析城市分析.
背景情况:
- 城市场景分类的监督学习需要广泛的标记数据,这往往很少.
- 现有的方法难以应对城市数据集中有限的高质量标签的挑战.
研究的目的:
- 提出创新的自我监督学习框架,三边冗余减少 (Tri-ReD),以克服城市场景分类中的标签短缺.
- 引入一种新的数据增强策略,即三分支相互排斥增强 (Tri-MExA),以减少增强引起的不确定性.
主要方法:
- 开发了三边冗余减少 (Tri-ReD) 框架,利用一种新的"三边损失"进行自我监督的预培训.
- 实施三分支相互排斥增强 (Tri-MExA) 增强表示学习.
- 在116,491个未标记的街景图像上预先训练模型,并在标记的数据集上进行微调 (BIC_GSV,BEAUTY).
主要成果:
- 三-ReD框架在城市场景分类的自我监督预培训中取得了最先进的 (SOTA) 性能.
- 在城市功能区识别中,直接监督学习方法的表现平均比直接监督学习方法高19%.
- 超越了在ImageNet上预先训练的模型大约11%.
结论:
- 拟议的Tri-ReD框架有效地学习了没有语义标签的基本城市场景表示,解决了标签稀缺的挑战.
- Tri-ReD是建筑不可知的,它在卷积神经网络 (CNN) 和视觉转换器 (ViT) 方面都表现出有效性.
- 这种自我监督的方法为各种下游城市分析任务提供了强大而适应性的解决方案.
相关概念视频
Reducing Line Loss
140
In a three-phase circuit, line loss is an indicator of energy dissipated as heat due to the resistance of transmission lines. To address this, incorporating transformers into the system—a step-up transformer at the source and a step-down transformer at the load—is a strategic solution. Two three-phase transformers are introduced to improve this.
With a step-up transformer at the source, the voltage is increased, thereby reducing the current in the transmission lines since power loss...
With a step-up transformer at the source, the voltage is increased, thereby reducing the current in the transmission lines since power loss...
140
Thematic Layering in GIS
26
In the past, planning projects such as schools or public facilities required extensive manual effort to gather and compile data. Information such as property boundaries, soil characteristics, road networks, zoning regulations, and flood zones had to be sourced individually from courthouses, utility providers, and registry offices. Assembling these datasets into a coherent format often took several months, delaying project timelines.The introduction of Geographic Information Systems (GIS)...
26


