Convolutional neural network-based method for estimating number of narrowband ultrasound echoes

Abstract

The invention discloses a convolutional neural network-based method for estimating the number of narrowband ultrasound echoes. The method includes the following steps that: narrowband ultrasonic echosignal model parameters are set, and a plurality of echo signal set models under different noise conditions are generated; some echo signal set models are randomly extracted from the plurality of echosignal set models so as to be subjected to short-time Fourier transformation, and a time-frequency spectrogram is generated and is adopted as a training sample set, and short-time Fourier transformation is performed on the remaining echo signal set models, and a time-frequency spectrogram is generated and is adopted as a test sample set; the obtained training sample set is inputted into a convolutional neural network so as to train the convolutional neural network, and the trained convolutional neural network is outputted; and the obtained test sample set is inputted into the trained convolutional neural network, so that the number of echoes corresponding to echo signals can be estimated. With the method of the invention adopted, the number of the narrowband ultrasound echoes can be effectively estimated. The method has the advantages of good robustness, high precision, and learning capacity.

Description

technical field [0001] The present invention relates to signal parameter estimation technology in the field of ultrasonic detection, and more specifically, to a method for estimating the number of narrow-band ultrasonic echoes based on a convolutional neural network. Background technique [0002] Ultrasonic detection imaging technology is a very important means and method in the field of nondestructive testing technology in today's society. It detects various engineering materials, components, Internal and surface defects such as structural parts, and make judgments and evaluations on the type, nature, quantity, shape, position, size, distribution and changes of defects. The purpose of nondestructive testing and diagnosis is to quantitatively grasp the relationship between defects and strength, evaluate the allowable load, life or remaining life of components, and detect structural incompleteness and defects in the process of manufacturing or using equipment (components), s...

AI Technical Summary

Problems solved by technology

Moreover, as the depth of ultrasonic penetration into the target increases, the echo attenuation will be further aggravated

In addition, for the multi-layer material detection technology of ultrasonic echo, the time delay between layers is mainly used to estimate the thickness of the layer, but as the thickness of the layer decreases, the reflected echoes between different layers will overlap. It will cause difficulties and errors in time measurement. This situation will also be manifested in two very close or small-sized defects inside the material. Due to the serious overlap of defect echoes, effective imaging cannot be performed, which will affect the number, size and accuracy of judging the number of defects. Location

At present, the estimation of echo parameters is basically realized by signal separation technology, parameterization method or time-frequency analysis technology. These methods have relatively large estimation errors in complex environments or low signal-to-noise ratios.

Images