Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Residual unit, network, target identification method, system and device and medium

A residual and network technology, applied in the field of residual unit, network and target recognition, can solve the problems of fixed spectrum resolution, loss of resolution information, loss of original signal waveform fine structure information, etc., and improve the accuracy of target recognition , the effect of retaining data characteristics

Active Publication Date: 2021-08-24
CHENGDU SHULIANYUNSUAN TECH CORP
View PDF8 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

But this method will lose the fine structure information of the original signal waveform
In addition, the generation of time-frequency diagrams is usually limited by parameters such as the window size and window interval step of the Fourier transform.
On the one hand, determining the appropriate transformation parameters requires prior knowledge, and the time and frequency resolution cannot be optimal at the same time
On the other hand, once the parameters are determined, the resolution of the generated spectrum is fixed accordingly
For end-to-end models with a fixed input size, this results in loss of additional resolution information
This becomes a bottleneck for spectrum-based identification methods

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Residual unit, network, target identification method, system and device and medium
  • Residual unit, network, target identification method, system and device and medium
  • Residual unit, network, target identification method, system and device and medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0070] Embodiment 1 of the present invention provides a multi-scale residual unit for processing raw underwater acoustic data. The structure of the multi-scale residual unit is as follows figure 2 As shown, it includes a residual network architecture composed of a batch normalization layer, an activation layer, and a convolutional layer for processing one-dimensional original audio waveforms. In the residual network architecture, there is at least one convolutional intermediate layer, The convolutional middle layer is formed by multiple convolutional layers in parallel, and at least one soft threshold calculation unit is provided between the convolutional middle layer and the end of the residual network architecture for soft threshold calculation and filtering.

[0071] Wherein, in practical applications, the number of convolutional intermediate layers and the number of soft threshold calculation units can be flexibly adjusted according to actual needs, which are not specifica...

Embodiment 2

[0092] Based on the multi-scale residual unit in Embodiment 1, the present invention establishes a deep neural network for processing underwater acoustic raw data, the deep neural network includes a network front end, a network middle end and a network end, and the output of the network front end The data is input to the middle end of the network, and the output data of the middle end of the network is input to the end of the network, and the middle end of the network includes at least one multi-scale residual unit described in Embodiment 1; when the middle end of the network includes multiple When there are one multi-scale residual unit, multiple multi-scale residual units are connected in sequence.

[0093] Wherein, in the embodiment of the present invention, the network front end is used to sequentially process the input data of the deep neural network through convolution processing, batch normalization processing, activation processing and pooling processing to obtain outpu...

Embodiment 3

[0117] Embodiment 3 of the present invention provides a training method for a deep neural network, where the deep neural network is the described deep neural network, and the training method includes:

[0118] Constructing the deep neural network to obtain the first deep neural network;

[0119] labeling the target radiated sound signal in the first audio data, obtaining first labeling data, and obtaining a training set based on the first labeling data;

[0120] Using the training set to train the first deep neural network to obtain a second neural network.

[0121] Wherein, in the third embodiment of the present invention, the training method also includes:

[0122] obtaining a test set based on the first labeled data;

[0123] The second deep neural network is tested using the test set.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a residual unit, a network, a target recognition method, system and device and a medium, and relates to the field of underwater acoustic recognition, deep learning and artificial intelligence, and the multi-scale residual unit comprises a residual network architecture composed of a batch normalization layer, an activation layer and a convolution layer, and is characterized in that the residual network architecture is used for processing a one-dimensional original audio waveform; the residual error network architecture is provided with at least one convolution middle layer, the convolution middle layer is formed by a plurality of convolution layers in parallel, and at least one soft threshold value operation unit is arranged between the convolution middle layer and the tail end of the residual error network architecture and is used for carrying out soft threshold value calculation and filtering. According to the invention, the underwater sound one-dimensional signal waveform can be effectively sensed.

Description

technical field [0001] The present invention relates to the fields of underwater acoustic recognition, deep learning and artificial intelligence, in particular, to a residual unit and a network, a target recognition method and system, a device and a medium. Background technique [0002] Underwater acoustic target recognition is to use the target radiation sound signal collected by the hydrophone to identify the target. It plays an important role in ocean transportation and waterway management. Due to the complexity of the ocean acoustic environment, the collected radiation signals are often accompanied by a large amount of interference noise. Because these noises are not explicit or semantic, it is difficult to label and automatically classify them. How to improve the automatic detection and classification performance of signals is still a challenging problem. [0003] The traditional method is to extract artificially designed features from the original audio data, then u...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G10L17/26G10L25/30G06N3/04
CPCG10L17/26G10L25/30G06N3/045
Inventor 不公告发明人
Owner CHENGDU SHULIANYUNSUAN TECH CORP
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products