High-speed Rail Line Building Dataset Display
This Project is created to introduce and display our proposed High-speed Rail Line (HSR) Building Dataset.The dataset is the first large-span and high-precision pixel-level semantic segmentation dataset of buildings along high-speed rail line in China. Compared with the existing public building datasets, it has obvious advantages in many aspects and can provide important data support for data-driven machine learning methods in the future.
The dataset is introduced in detail in file "[1] High-speed Rail (HSR) Line Building Dataset Introduction.pdf". The dataset includes two sub-datasets, i.e., Zhengzhou-Xi’an HSR Line Building Dataset and Beijing-Shanghai HSR Line Building Dataset.
The Zhengzhou-Xi’an HSR Line Building Dataset is displayed in file "[2] Zhengzhou-Xi’an HSR Line Building Dataset Display.pdf".
The Beijing-Shanghai HSR Line Building Dataset is displayed in file "[3] Beijing-Shanghai HSR Line Building Dataset Display.pdf".
Three original images and corresponding labeled samples are provided in "Data Samples" folder. We will open all the data in the dataset later to support related scientific research.
All data of our proposed HSR dataset can be previewed in the "HSR_dataset" folder.
The proposed method can be found in the "FANet" folder and the results can be found in the "results_HSR" folder.
The image block collection and distribution diagram of the dataset is shown below.
We have developed a set of web-based online potential hazardous building information identification system along the high-speed rail line, which has achieved excellent test performance in the Zhengzhou-Xi'an high-speed rail line. We expecte the system can provide service for the accurate investigation of potential hazardous building in all high-speed rail lines in the future.
Based on our high-speed rail line building dataset, our trained FCN model is applied to building extraction in the entire Zhengzhou-Xi'an high-speed rail line (484 km, 1 km on both sides of line). The results can be shown below. It prove the usability of our proposed dataset and the effectiveness of our proposed method.
