Vibration optical fiber signal classification method based on full convolutional neural network

Abstract

The invention discloses a vibration optical fiber signal classification method based on a full convolutional neural network, and the method solves a problem that a conventional technology needs to manually extract features, enables the classification feature information to be completely and automatically extracted from a training sample through a full convolutional neural network algorithm, and greatly reduces the manual intervention process. Effective features in a training sample are extracted by utilizing a full convolutional neural network, depth information in an image gray value is fullyutilized, and the recognition rate is effectively improved under the condition that the calculated amount is not remarkably increased.

Description

technical field [0001] The invention relates to the field of security technology, in particular to an intelligent fence alarm system capable of realizing vibration detection. Background technique [0002] The existing vibration-type optical fiber sensing products have the following technologies for classifying signals: [0003] 1) Extract the vibration signal amplitude and strength, set different thresholds according to experience, and classify the duty cycle within a specific range by statistics; [0004] 2) Multi-point feature supervised pattern recognition algorithm; [0005] 3) After converting the one-dimensional signal to a two-dimensional image, use image processing classification to classify; [0006] 4) Analyze the frequency domain and phase spectrum analysis of the vibration signal. [0007] The main drawbacks of existing vibratory fiber optic sensing products for signal classification technology: [0008] 1) Insufficient representation of amplitude and intensi...

AI Technical Summary

Problems solved by technology

[0010] 2) The extraction of feature points is heavily dependent on manual work and experience

[0011] The accuracy of pattern recognition is determined by the quality of feature point extraction. Existing technologies rely heavily on labor and experience. On the one hand, designing feature extractors is time-consuming and labor-intensive. On the other hand, these features are still relatively shallow features, which seriously affect pattern recognition. accuracy rate

[0012] 3) After the one-dimensional signal is converted to a two-dimensional image, conventional image processing and classification algorithms have many limitations and poor performance

[0013] In the technology of image pattern recognition after one-dimensional signal is converted to two-dimensional image, feature extraction still relies on manual extraction, and the calculation efficiency is low. The size of the pixel block limits the size of the perception area, and only some local features can be extracted, which leads to the performance of classification being affected. limit

[0014] 4) Frequency domain and phase spectrum analysis are difficult

[0015] Algorithms based on frequency domain and phase spectrum analysis, due to the limited data carried by one-dimensional vibration signals, it is often difficult to analyze, not only time-consuming, but also the classification effect is not good

Images