Aerosol Identification Based on Attention-Unet Neural Network

Fig. 5. Identification result and corresponding observation data on March 22, 2023. (a) Model identification result; (b) depolarization ratio; (c) reflectivity factor; (d) extinction coefficient; (e) mass concentration ratio of PM_2.5 to PM₁₀; (f) mass concentrations of PM_2.5 and PM₁₀

Download full size

Fig. 6. Identification result and corresponding observation data on April 14, 2023. (a) Model identification result; (b) depolarization ratio; (c) reflectivity factor; (d) extinction coefficient; (e) mass concentration ratio of PM_2.5 to PM₁₀; (f) mass concentrations of PM_2.5 and PM₁₀

Download full size

Fig. 7. Identification result and observation data on November 17, 2022. (a) Identification result; (b) depolarization ratio; (c) reflectivity factor; (d) extinction coefficient; (e) mass concentration ratio of PM_2.5 to PM₁₀; (f) mass concentrations of PM_2.5 and PM₁₀

Download full size

Fig. 8. Identification result and observation data on October 19, 2022. (a) Identification result; (b) depolarization ratio; (c) reflectivity factor; (d) extinction coefficient; (e) mass concentration ratio of PM_2.5 to PM₁₀; (f) mass concentrations of PM_2.5 and PM₁₀

Download full size

Table 1. Parameters of instruments
View table
Table 1. Parameters of instruments
Instrument Observationparameter Corresponding spatio-temporal resolution
Lidar Depolarization ratio/extinction coefficient 7.5 m, 1 min
Millimeter-wave cloud radar Reflectivity factor 30 m, 1 min

Table 2. Data sample selection criteria
View table
Table 2. Data sample selection criteria
Weather AQI Extinction coefficient Depolarization ratio
Hazy 100‒500 0.2‒3.0 0.02‒0.10
Dusty 100‒500 0.2‒3.0 0.15‒0.38
Sunny or cloudy 0‒100 0‒0.2 0‒0.1

Table 3. Numbers of labels for each category in dataset
View table
Table 3. Numbers of labels for each category in dataset
Label Pollutant Clear Cloud Dust Polluted dust
Number 2815608 30443514 7581637 1593431 1202130

Table 4. Performance comparison of different neural models
View table
Table 4. Performance comparison of different neural models
Model Accuracy /% Average precision /% Average recall /%
FCN 87.5 81.7 73.9
SegNet 88.4 86.4 81.5
U-Net 95.2 89.3 88.6
Attention-Unet 96.5 91.5 89.9

Tools

Get Citation

Copy Citation Text

Changqing Fu, Zhipeng Yang, Chengli Ji, Tao Fu, Fa Tao, Jianhui Zheng. Aerosol Identification Based on Attention-Unet Neural Network[J]. Chinese Journal of Lasers, 2025, 52(10): 1010001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites