Welcome to the Efficient-DVS repository! This repository focuses on developing an efficient distributed optical fiber vibration sensing algorithm that achieves real-time recognition in real-world long-distance fiber applications.
-
Shallow CNN: We utilize a shallow Convolutional Neural Network (CNN) architecture to improve throughput and computational efficiency in the vibration sensing algorithm. This allows for faster real-time recognition of vibration events along the fiber.
-
CNN FPGA Implementation: To further enhance efficiency, we propose a CNN FPGA implementation scheme. By leveraging the power of field-programmable gate arrays (FPGAs), we enable high parallelism and real-time processing capabilities.
-
Knowledge Distillation: Addressing the challenge of poor generalizability in shallow CNNs, we employ a knowledge distillation scheme. This technique enhances the generalizability of the lightweight models, resulting in improved accuracy on data from unknown environment.
This repository provides the necessary codes for the evaluation and simulation of the proposed scheme. It includes:
-
Testing Codes: You will find codes for testing the efficiency and accuracy of the distributed optical fiber vibration sensing algorithm.
-
Example Samples: We have included example samples for you to experiment with and evaluate the performance of the algorithm.
-
Trained Models: Pre-trained models are provided for easy evaluation and testing purposes.
The complete dataset for training and evaluation is available on Google Drive. Please access the dataset here.
Please note that the training codes, and hardware design codes will be made available once the associated article is published.
We encourage you to explore this repository and contribute to this aspect. Your contributions can help enhance the efficiency, accuracy, and scalability of distributed optical fiber vibration sensing systems.
Thank you for your interest in the Efficient-DVS repository.
If you have any questions or need further assistance, don't hesitate to reach out. Contact: zluo@hust.edu.cn, wuhaoboom@hust.edu.cn
Best regards.