Modified Data Augmentation Integration Method for Robust Intrusion Events Recognition with Fiber Optic DAS System

Due to the poor generalization performance of the artificial intelligence recognition models, high false alarm rate (FAR) is still a challenge for intrusion events recognition with fiber optic distributed acoustic sensor (DAS) system in practical applications. Intrinsically, the mismatch between the training and test set is more seriously under complex condition such as low signal-to-noise ratio (low-SNR), input-shift and events-mixing exist simultaneously, resulting in the poor generalization performance of the recognition models. Previous researches mostly focus on the solution under single typical condition or require extra computing resources, while the conventional data augmentation integration (conventional-DAI) methods for image signals in computer vision offer a new way for intrusion recognition of DAS signals, which can tackle several typical conditions simultaneously. In this article, according to the motivation of conventional-DAI and the characteristics difference between DAS signals and image signals, the modified data augmentation integration (modified-DAI) method is proposed. Through the effective simulation of DAS signals variability about low-SNR, input-shift and events-mixing, this method can reduce the mismatch between the training and test set and improve model generalization performance under complex condition without sacrificing recognition speed. The results from the fiber optic DAS system based field test demonstrates that the recognition accuracy of modified-DAI is 85% under complex condition, which is 20% higher than the conventional-DAI. Apparently, the modified-DAI is promising to reduce FAR in practical applications, which is further conducive to the applications of DAS technology for intrusion events recognition.