Author Affiliations
1College of Flight Technology, Civil Aviation University of China, Tianjin 300300, China2College of Electronic Information and Automation, Civil Aviation University of China, Tianjin 300300, Chinashow less
Fig. 1. Schematic diagram of Doppler lidar
Fig. 2. Schematic diagram of part of the data
Fig. 3. Network input layer eddy current dissipation rate image
Fig. 4. Output result graph of the first convolutional layer
Fig. 5. Output of the second convolutional layer
Fig. 6. Diagram convolutional neural network structuream
Fig. 7. Diagram of CNN training model
Fig. 8. Diagram of convolutional neural network training process
Fig. 9. Schematic diagram of the relationship between decreasing learning rate and network accuracy
Fig. 10. Schematic diagram of loss function during training
Fig. 11. Original data image of two false positives
Fig. 12. Diagram of the judgment of turbulence by two methods
Fig. 13. Schematic diagram of two early warning methods hitting turbulence
Parameter | Value | Wavelength/nm | 1550 | Sampling interval/ns | 2.5 | Laser pulse width/ns | 200 | Pulse repetition frequency/kHz | 10 | Accumulated pulse number | 5000 | Range resolution/m | 30 | Maximum detection distance/km | 6 |
|
Table 1. Relevant parameters of lidar
Date | CNN(L/M/H) | Vsf(L/M/H) | Times of false positives | 2016.11.25 | H | L | 1 | 2016.11.26 | H | H | 0 | 2016.11.27 | H | H | 0 | 2016.11.28 | L | L | 0 | 2016.11.29 | H | H | 0 | 2016.11.30 | M | M | 1 | 2016.12.01 | M | M | 0 | 2016.12.02 | L | H | 1 | 2016.12.03 | H | H | 0 | 2016.12.04 | H | H | 0 | 2016.12.05 | M | M | 0 | 2016.12.06 | L | L | 0 | 2016.12.07 | M | M | 0 | 2016.12.08 | H | H | 0 | 2016.12.09 | L | L | 0 | 2016.12.10 | L | L | 0 | 2016.12.11 | H | L | 1 | 2016.12.12 | H | H | 0 | 2016.12.13 | H | H | 0 | 2016.12.14 | H | H | 0 |
|
Table 2. Turbulence warning statistics of the two methods
Date | CNN Alarm statistics(T/F) | Classified statistics(T/F) | Hog-SVM Alarm statistics(T/F) | Classified statistics(T/F) | 2016.07.20 | T | T | T | T | 2016.09.03 | T | T | T | T | 2016.09.07 | F | F | F | F | 2016.09.10 | T | T | T | T | 2016.09.11 | T | T | T | T | 2016.09.20 | T | T | T | T | 2016.09.24 | T | T | F | F | 2016.10.15 | T | F | F | F | 2016.11.06 | T | T | F | T | 2017.04.13 | F | F | T | T | 2017.04.17 | T | T | T | T | 2017.05.06 | T | T | T | T | 2017.05.09 | T | T | T | T | 2017.05.13 | T | T | F | F | 2017.05.14 | T | T | F | F |
|
Table 3. Judgment results made by two methods on 15 sets of unit reports