Laser & Optoelectronics Progress, Volume. 62, Issue 7, 0706002(2025)
Fast Classification Method for Unbalanced Few-Sample Events for Φ-OTDR Intrusion Detection System
Figures & Tables(25)
Fig. 1. Spatial correlation of event phase changes. (a) Intrusion event phase diagram; (b) intrusion event phase sequence diagram
Fig. 7. Deeply separable convolutional mode. (a) Depthwise separable convolution based on SE; (b) depthwise separable convolution based on ECA improvement
Fig. 10. Equipment and fiber optic cable deployment. (a) Distributed optical fiber sensing equipment; (b) optical cable fence bundling deployment; (c) optical cable tube deployment; (d) exposed fixed deployment of optical cable
Fig. 11. Phase diagram of intrusion events. (a) Raw phase diagram from das device; (b) phase change after cropping; (c) data matrix of phase change; (d) visualization of the phase change matrix
Fig. 12. Experimental results enhanced by SCE method. (a) KO; (b) KF; (c) OF; (d) SF; (e) PA; (f) VA
Fig. 13. Experimental results of spatial correlation denoising enhancement method. (a) KO; (b) KF; (c) OF; (d) SF; (e) PA; (f) VA
Fig. 14. Cosine similarity of results obtained by different data enhancement methods. (a) SCE; (b) SCDE
Fig. 15. Experimental results of spatial correlation noise fusion enhancement method. (a) KO; (b) KF; (c) OF; (d) SF; (e) PA; (f) VA
Fig. 18. Confusion matrixs of MobileNetV3-DAS experimental results. (a) SCNFE; (b) SCDE; (c) SCE
Table 1. Depth separable convolution parameter configuration
View tableTable 1. Depth separable convolution parameter configuration
Input Operator Exp size Out ECA NL S 1122×16 BNeck 3×3 16 16 √ RE 2 562×16 BNeck 3×3 72 24 ‒ RE 2 282×24 BNeck 3×3 88 24 ‒ RE 1 282×24 BNeck 5×5 96 40 √ HS 2 142×40 BNeck 5×5 240 40 √ HS 1 142×40 BNeck 5×3 240 40 √ HS 1 142×40 BNeck 5×5 120 48 √ HS 1 142×48 BNeck 5×5 144 48 √ HS 1 142×48 BNeck 5×5 288 96 √ HS 2 72×96 BNeck 5×5 576 96 √ HS 1 72×96 BNeck 5×5 576 96 √ HS 1
Table 2. Description of experimental data
View tableTable 2. Description of experimental data
Serial number Event name Abbreviations of events Time scale /s Number of original samples Descriptive 1 Knocking on fiber optic cables KO 1‒5 23 Fiber optics on the fence were struck continuously with wooden sticks 2 Knocking on the fence KF 1‒5 29 Continuous tapping on the fence with a wooden stick 3 Over the fence. OF 5‒15 21 Over the fence into the interior of the critical area 4 Shaking fiber optic cables SF 2‒5 5 Shake fiber optics fixed to the fence by hand 5 Pedestrian access PA 2‒5 23 Pedestrians traveling through key area entrances and exits 6 Vehicle access VA 2‒7 7 Vehicle movement through key area entrances and exits 7 Environmental noise EN ‒ ‒ Noise data for different time periods and different fence locations
Table 3. Sample size of different data enhancement methods
View tableTable 3. Sample size of different data enhancement methods
Event label Event name Original sample SCE SCDE SCNFE Noise sample 1 KO 23 230 184 736 4 2 KF 29 290 232 696 3 3 OF 21 210 168 672 4 4 SF 6 60 48 720 15 5 PA 10 100 76 684 9 6 VA 7 84 55 715 13
Table 4. Classification results of intrusion signals using different data enhancement methods
View tableTable 4. Classification results of intrusion signals using different data enhancement methods
Data augmentation Event label Precision Recall Specificity F1 SCE 1 0.86 1.00 0.99 0.92 2 0.90 0.86 0.97 0.88 3 1.00 0.83 1.00 0.91 4 0.81 0.86 0.91 0.83 5 1.00 0.80 1.00 0.89 6 0.67 0.70 0.89 0.68 SCDE 1 0.83 1.00 0.99 0.91 2 0.94 1.00 0.98 0.97 3 0.95 0.87 0.98 0.91 4 1.00 1.00 1.00 1.00 5 0.83 0.83 0.95 0.83 6 1.00 1.00 1.00 1.00 SCNFE 1 1.00 1.00 1.00 1.00 2 0.97 1.00 0.99 0.98 3 0.98 0.98 1.00 0.99 4 1.00 0.98 1.00 0.99 5 1.00 0.92 0.98 0.92 6 0.98 0.97 0.99 0.97
Table 5. Experimental results of different classification methods
View tableTable 5. Experimental results of different classification methods
Model Data augmentation Precision Recall Specificity F1 Model size /MB Time /ms AlexNet SCE 0.85 0.85 0.96 0.84 55.60 17.79 SCDE 0.87 0.83 0.96 0.83 20.00 SCNFE 0.88 0.83 0.96 0.83 17.90 VGG16 SCE 0.92 0.92 0.98 0.92 512.00 155.79 SCDE 0.92 0.92 0.98 0.92 141.63 SCNFE 0.96 0.95 0.99 0.95 154.32 ResNet34 SCE 0.84 0.87 0.97 0.85 81.30 81.61 SCDE 0.94 0.89 0.98 0.91 98.80 SCNFE 0.98 0.98 1.00 0.98 88.59 MobileNetV2 SCE 0.85 0.83 0.96 0.84 8.74 45.36 SCDE 0.88 0.90 0.97 0.88 43.55 SCNFE 0.97 0.97 0.99 0.97 41.90 MobileNetV3 SCE 0.88 0.82 0.96 0.83 5.94 18.87 SCDE 0.88 0.85 0.96 0.85 18.96 SCNFE 0.96 0.96 0.99 0.96 18.52 MobileNetV3-DAS SCE 0.87 0.84 0.96 0.85 4.17 18.43 SCDE 0.93 0.95 0.98 0.94 17.43 SCNFE 0.97 0.97 0.99 0.97 16.62
Table 6. Experimental results of improved MobileNetV3 with different attention mechanisms
View tableTable 6. Experimental results of improved MobileNetV3 with different attention mechanisms
Attention Method Accuracy Precision Recall Specificity F1 Model size /MB Time /ms SE SCE 0.81 0.88 0.82 0.96 0.83 5.94 18.87 SCDE 0.84 0.88 0.85 0.96 0.85 18.96 SCNFE 0.96 0.96 0.96 0.99 0.96 18.52 CBAM SCE 0.81 0.86 0.83 0.96 0.84 4.72 26.10 SCDE 0.85 0.82 0.86 0.97 0.83 23.06 SCNFE 0.89 0.89 0.89 0.98 0.89 22.07 CA SCE 0.79 0.85 0.82 0.95 0.83 4.84 21.12 SCDE 0.89 0.93 0.93 0.87 0.93 21.83 SCNFE 0.97 0.98 0.98 1.00 0.99 25.87 ECA SCE 0.82 0.87 0.84 0.96 0.85 4.17 18.43 SCDE 0.92 0.93 0.95 0.98 0.94 17.43 SCNFE 0.97 0.97 0.97 0.99 0.97 16.62
Tools
Get Citation
Copy Citation Text
Zhanping Zhang, Haoyin Lü, Fangping Yang, Chen Zhang, Jin Yan. Fast Classification Method for Unbalanced Few-Sample Events for Φ-OTDR Intrusion Detection System[J]. Laser & Optoelectronics Progress, 2025, 62(7): 0706002
Paper Information
Category: Fiber Optics and Optical Communications
Received: Oct. 22, 2024
Accepted: Dec. 2, 2024
Published Online: Apr. 8, 2025
The Author Email: Zhanping Zhang (622023340015@smail.nju.edu.cn)
CSTR:32186.14.LOP242137
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20