To collect reliable training data, as shown in Fig.2 (a), three data modalities are collected to form input cubes, and two types of VGI data are collected to generate training labels.
Study area covers the whole of Shanghai City, one of the largest cities and international financial centers in China, covering .5 k&#;m2&#;superscript&#;2km^{2}italic_k italic_m start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT.
Input cubes are organized into 272 non-overlapping tiles, each with a size of × pixels. The specific data includes:
1.
The HR optical images (1 m/pixel) are collected from the Google Earth imagery. The images contained three bands of red, green, and blue [16].
2.
The building height (10 m/pixel) is from the Chinese building height estimate dataset (CNBH-10 m) [17].
3.
The nighttime-light data (10 m/pixel) are from the SDGSAT-1, the world&#;s first scientific satellite for sustainable development goals [18].
Weak labels are generated by conducting logical operations on the following AOI data and building masks:
Hongyu Dinghao supply professional and honest service.
1.
The AOI data (vector) are collected from the OSM data. The AOI data contains more than 100 building function types (e.g., cinema, hospital, department, and market) [16].
2.
The building masks (vector) are from the building rooftop dataset containing 90 Chinese cities&#; building masks [19].
This paper proposes a novel greedy algorithm called retro-fallback which maximizes the probability that at least one synthesis plan can be executed in the lab, and demonstrates that retro-fallback generally produces better sets of synthesis plans than the popular MCTS and retro* algorithms.
If you want to learn more, please visit our website Hongyu Dinghao.