Global Vision Media Focus
The company was featured on CCTV a total of 15 times, and its reach through Party media, central media, and local official media reached a total of 50 million people.
Red mud field identification comparison based on Xiguang-1 05 satellite (Tianxianpei) and Sentinel-2
Background Introduction
Red mud storage facilities are specialized structures for storing red mud, a highly alkaline waste residue generated during alumina production. Their core function is to prevent red mud leakage and environmental pollution. Because red mud contains fluorides and its accompanying liquid has a pH value far exceeding safety standards, it is classified as Class II general industrial solid waste and requires anti-seepage design according to relevant environmental standards. Early red mud storage facilities, lacking anti-seepage measures, caused groundwater pollution problems. As of 2016, China's accumulated red mud stockpiles exceeded 50 million tons, with a utilization rate of only 15%, and stockpiling remains the primary treatment method.
Principles and methods
Satellite remote sensing can be used to identify and monitor the distribution of open-pit red mud fields and the progress of ecological restoration and transformation, which helps to effectively manage this type of industrial solid waste.
Compared to multispectral satellites, hyperspectral satellites have stronger material resolution capabilities, making them well-suited for scenarios requiring specific ground feature identification. Therefore, a known red mud field area was selected for comparative analysis.
Table 1. Comparison of Band Parameters
A red mud field in an aluminum industry in Shandong Province was selected as the target, and hyperspectral satellite data from Sentinel-2 and Xiguang-1 05 (Tianxianpei) satellites were used for target identification. The Sentinel-2 data used median composite data from March 10 to April 20, 2025, and the Xiguang-1 05 (Tianxianpei) satellite data used data from March 29, 2025.
For Sentinel-2, due to the limited number of bands, spectral feature matching algorithms were not suitable. Therefore, supervised clustering using random forest was employed to extract red mud fields in bauxite mines. The samples were polygonal vectors, and five categories of samples were selected: red mud fields, farmland, construction site dust control nets, water bodies, and red building roofs.
For the Xiguang-1 05 satellite (Tianxianpei), the single-pixel spectral curve of the red mud field was collected as a reference endmember, and the target was extracted using the Adaptive Coherence Estimator algorithm and threshold segmentation.
Area 1
Figure 1. RGB image of Sentinel-2 and extraction results
Figure 2. Red mud extraction results from Xiguang-1 05 satellite (Tianxianpei).
Area 2
Figure 3. RGB image of Sentinel-2 and extraction results
Figure 4. RGB image and extraction results of Xiguang-1 05 satellite (Tianxianpei), and verification of high-resolution base map.
Results Analysis
When using data from Xiguang-1 05 satellite (Tianxianpei), two red mud deposits were found outside the known red mud field area, but Sentinel-2 data did not identify these red mud deposit areas.
The comparison results show that multispectral imagery lacks the ability to identify specific substances in scenarios with few samples for supervised classification. Hyperspectral imagery, on the other hand, can utilize various hybrid pixel decomposition algorithms for substance identification, requiring only a minimum of one pixel sample (pure pixels are optimal, but hybrid pixels still meet the requirements). The supervised clustering algorithms commonly used with multispectral imagery have poor universality; models trained on one image perform poorly on other images. However, endmember extraction based on hyperspectral imagery allows for direct application to other images with better results.
Smaller FWHM and denser bands enable the identification of subtle differences in reflectance characteristics between different materials. Therefore, in material identification scenarios, hyperspectral imagery still shows a significant advantage even when the ground resolution is much lower than that of multispectral imagery.