Extraction of construction site dust nets based on hyperspectral target recognition

Background Introduction

Construction sites, as crucial locations for urban development, have always been a major source of dust pollution. With the increasing demands for urban environmental quality from the government in recent years, effectively controlling dust pollution from construction sites has become a focus of public attention. Against this backdrop, dust control nets, as an efficient and practical dust control tool, are increasingly widely used on construction sites, becoming an indispensable part of dust control efforts. Therefore, by extracting and statistically analyzing dust control net data, we can indirectly assess the distribution and progress of construction sites, identifying hotspots for civil engineering activities.

Methods and Principles

This paper uses hyperspectral imagery of the Xi'an area taken by the Xiguang-1 05 satellite on April 6, 2025. Because dust nets have gaps that allow some light to pass through, their reflectance characteristics vary slightly depending on the surface they cover. Traditional image classification methods are ineffective and easily confused with some building roofs. Therefore, based on target recognition methods, we extracted the spectra of the dust net pixels as endmembers based on visual interpretation and established a spectral library. Dust net extraction was then performed based on this spectral library.

Table 1. Data Introduction

process

1. Radiation calibration:The original DN value data is radiometrically corrected using radiometric calibration coefficients to obtain the apparent radiance data.

2. Atmospheric correction:Atmospheric correction was performed using the 6S model to obtain surface reflectance data.

3. Image registration:The processed data is then georeferenced based on the base map.

4. Endmember extraction:Select multiple dustproof mesh pixels, calculate the average spectrum, and save it as an endmember.

5. Hybrid Pixel Decomposition:Experiments showed that the Adaptive Coherence Estimator algorithm performed best for this scenario, using manual thresholding to filter pixels close to the dust net target.

6. Vectorization and heatmap generation:The extracted raster pixels are converted into points and polygons, and a heat map is generated based on the points using a kernel density estimation algorithm.

Results Display

Figure 1. Distribution of spectral sampling points for dustproof netting

Figure 2. Spectral curve of dustproof netting sampling

Figure 3. Spectral library established based on the average spectrum of the dustproof net.

Figure 1 shows the sampling points of typical spectra of dust control nets. All sampling points in the figure were obtained through visual interpretation of the latest temporal imagery. Figure 2 shows the spectra of each sampling point; slight differences exist in the red edge and near-infrared regions due to variations in surface vegetation. The average spectrum in Figure 2 was calculated and saved in a spectral library format for subsequent mixed pixel decomposition (Figure 3).

Figure 4. Heat map showing the distribution of dust control netting at construction sites in Xi'an and surrounding areas.(April 6, 2025)