研究目的
Investigating the effectiveness of a novel classification framework, Multi-branch regression network (MBR-Net), for building classification using high-resolution remote sensing images.
研究成果
The proposed Multi-branch regression network (MBR-Net) framework for building classification using high-resolution remote sensing images demonstrates significant improvement over the state-of-the-art U-Net in both accuracy and speed of convergence. A new training strategy was also developed to handle uneven samples effectively.
研究不足
The study does not address the potential limitations of the MBR-Net framework in terms of computational resources required for training or its applicability to other types of remote sensing images beyond building classification.
1：Experimental Design and Method Selection:
The study proposes a Multi-branch regression network (MBR-Net) framework for building classification, which includes a contraction path and an expansion path with multiple output branches for different scale features.
2：Sample Selection and Data Sources:
The Inria aerial image labeling dataset is used, containing images from five cities with two semantic classes (building and not building).
3：List of Experimental Equipment and Materials:
Experiments are carried out using software on a GTX 1060 GPU with 6GB of RAM.
4：Experimental Procedures and Operational Workflow:
The original images are resized into 32×32 pixels for balancing the amount of building and not building categories, and divided into five type groups for training.
5：Data Analysis Methods:
The accuracy of the proposed MBR-Net is compared with the state-of-the-art U-Net framework using classification accuracy as the metric.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
