Dual-model hybrid pattern recognition method based on a fiber optic line-based sensor with a large amount of data

Area of Science:

• Fiber optic sensing technology
• Pattern recognition algorithms
• Signal processing and data analysis

Background:

• Fiber optic sensors generate large datasets, posing memory and processing challenges.
• Accurate identification of events from vibration signals is crucial for various monitoring applications.
• Existing methods may struggle with high false positive rates in complex environments.

Purpose of the Study:

• To develop a robust dual-model hybrid pattern recognition system for fiber optic sensor data.
• To enhance the accuracy and reduce false positives in event classification.
• To efficiently process large vibration signal datasets by converting them into manageable image formats.

Main Methods:

• Vibration signals are converted into gray-level images to reduce memory requirements.
• The ResNet18 model is employed for initial classification of events.
• Over-zero rate and short-time energy features are extracted for intrusion signal analysis.
• A Support Vector Machine (SVM) is utilized in conjunction with extracted features.
• A discriminator combines the ResNet18 and SVM models for final event type determination.

Main Results:

• The proposed dual-model hybrid system demonstrates excellent average recognition accuracy.
• The method achieves a high accuracy rate of 97.1% for classifying six different events.
• The integration of ResNet18 and SVM effectively reduces the false positive rate.

Conclusions:

• The developed dual-model hybrid pattern recognition method offers a highly accurate and efficient solution for fiber optic sensor data analysis.
• This approach effectively addresses the challenges of large data volumes and false positives in event detection.
• The system shows significant promise for real-world applications requiring reliable event identification from vibration signals.