Convolutional neural network time delay estimation method based on quadratic rational kernel function

Abstract

The invention discloses a convolutional neural network time delay estimation method based on a quadratic rational kernel function. The method comprises the following steps: S1, collecting a time delay value between a radar observation signal and a multipath echo signal; S2, transforming the time delay value by adopting a quadratic rational kernel function; S3, carrying out the preprocessing of the radar observation signal; S4, constructing a convolutional neural network, and inputting the preprocessed radar observation signals into the neural network for nonlinear transformation to obtain time domain transformation representation of the input signals; S5, inputting the time domain transformation representation into a neural network to extract time domain delay features; carrying out interlayer relation compression transformation to obtain a time domain time delay feature value of the input signal; S6, fitting the quadratic rational kernel function representation of the time delay characteristic value and the real time delay value by using a minimum variance iterative algorithm, and carrying out multiple times of training to obtain a time delay estimation network; and S7, inputting a test sample into the time delay estimation network, and obtaining a time delay estimation value by adopting a peak searching function.

Description

technical field [0001] The invention belongs to a reflection detection signal processing method, in particular to a convolutional neural network time delay estimation method based on a quadratic rational kernel function. Background technique [0002] At present, frequency sweep signals are widely used in the field of reflection detection, such as vibrator, radar, sonar and ultrasonic non-destructive testing. It is used as a source signal to transmit to the detected medium, then propagates in the medium, and forms a reflected echo after encountering different targets, and then is received by the detector, and the received signal contains multiple frequency sweep signals with different time delays echo. In the signal processing of reflection detection, in order to estimate the distance of each target, it is necessary to detect each echo and estimate its time delay. [0003] In the application scenarios of radar and sonar detection, radar usually obtains the distance from the...

AI Technical Summary

Problems solved by technology

Although this patent solves the interference of environmental noise on the radar echo signal, it does not solve the multipath effect caused by the aliasing of multi-angle multipath echo signals

Images