King Abdulaziz City for Science and Technology, National Center for Data Analytics and Artificial Intelligence, P.O. Box 6086, Riyadh 11442, Saudi Arabia.
Comput Intell Neurosci. 2022 Feb 18;2022:5008854. doi: 10.1155/2022/5008854. eCollection 2022.
Recently, settlement planning and replanning process are becoming the main problem in rapidly growing cities. Unplanned urban settlements are quite common, especially in low-income countries. Building extraction on satellite images poses another problem. The main reason for the problem is that manual building extraction is very difficult and takes a lot of time. Artificial intelligence technology, which has increased significantly today, has the potential to provide building extraction on high-resolution satellite images. This study proposes the differentiation of buildings by image segmentation on high-resolution satellite images with U-net architecture. The open-source Massachusetts building dataset was used as the dataset. The Massachusetts building dataset includes residential buildings of the city of Boston. It was aimed to remove buildings in the high-density city of Boston. In the U-net architecture, image segmentation is performed with different encoders and the results are compared. In line with the work done, 82.2% IoU accuracy was achieved in building segmentation. A high result was obtained with an F1 score of 0.9. A successful image segmentation was achieved with 90% accuracy. This study demonstrated the potential of automatic building extraction with the help of artificial intelligence in high-density residential areas. It has been determined that building mapping can be achieved with high-resolution antenna images with high accuracy achieved.
最近，定居点规划和重新规划过程成为快速发展城市的主要问题。无规划的城市住区很常见，尤其是在低收入国家。从卫星图像中提取建筑物也带来了另一个问题。问题的主要原因是手动提取建筑物非常困难且耗时。人工智能技术的飞速发展，有可能为高分辨率卫星图像提供建筑物提取。本研究提出了利用 U-net 架构对高分辨率卫星图像进行图像分割来区分建筑物。该研究使用了麻省理工学院的开源建筑物数据集作为数据集。该数据集包括波士顿市的居民楼。目的是去除波士顿高密度城市中的建筑物。在 U-net 架构中，使用不同的编码器进行图像分割，并比较结果。根据所做的工作，建筑物分割的 IoU 精度达到了 82.2%。F1 分数为 0.9，获得了较高的分数。图像分割的准确率达到了 90%。这项研究展示了人工智能在高密度住宅区自动提取建筑物的潜力。已经确定可以使用高分辨率天线图像并实现高精度来进行建筑物测绘。
