CN110245608B - Underwater target identification method based on half tensor product neural network - Google Patents
Underwater target identification method based on half tensor product neural network Download PDFInfo
- Publication number
- CN110245608B CN110245608B CN201910513322.9A CN201910513322A CN110245608B CN 110245608 B CN110245608 B CN 110245608B CN 201910513322 A CN201910513322 A CN 201910513322A CN 110245608 B CN110245608 B CN 110245608B
- Authority
- CN
- China
- Prior art keywords
- neural network
- underwater
- tensor product
- half tensor
- product neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/20—Scenes; Scene-specific elements in augmented reality scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The invention provides an underwater target identification method based on a half tensor product neural network, which is characterized in that an underwater sound signal is received through an underwater sonar sensor, and time domain and frequency domain information of the sound signal is presented in an LOFAR map through short-time Fourier transform; constructing a half tensor product neural network of the data sample by taking the LOFAR atlas sample as an input feature matrix; dividing the received underwater acoustic signals into a training set and a verification set, and inputting a half tensor product neural network for training and verification; selecting different hyper-parameters, performing model training on the half tensor product neural network by using a training set, comparing the test effect of a verification set, and determining the hyper-parameters with high test accuracy; and finally, inputting the currently acquired acoustic signal of the underwater target into the half tensor product neural network after model training, and giving a judgment result. The method can improve the underwater target recognition rate, expand application scenes and be suitable for recognizing the underwater target in the complex marine environmental noise.
Description
Technical Field
The invention belongs to the field of signal processing, and relates to methods of a neural network, acoustic signal processing, half tensor product multiplication, underwater target identification and the like.
Background
The target classification and identification has very important significance for various research fields, and the traditional target classification and identification is to artificially extract various features and then construct a classifier for classification and identification. With the rapid development of computer vision technology, the target classification and recognition technology based on deep learning is widely researched by people, and the classification and recognition effect beyond human is achieved. However, at present, research is mainly carried out in the field of computer vision such as images and videos, research on acoustic signals is also focused on voice signal processing and natural language processing, and classification and identification of various underwater targets are still carried out on the traditional method.
The conventional acoustic signal recognition algorithm generally includes a dynamic time warping technique, a support vector machine, a gaussian mixture model, a hidden markov model, and the like. The traditional method is greatly influenced by artificial feature extraction and environmental noise, and the identification models are only symbolic systems, so that the modeling capacity is reduced, the identification performance of acoustic signals in a complex environment is greatly reduced, and the classification identification rate in complex and variable marine environmental noise is not ideal enough. Artificial neural networks are a research hotspot emerging in the field of artificial intelligence in recent years. With the continuous and deep research work of the artificial neural network, great progress has been made at present.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an underwater target identification method based on a half tensor product neural network, which introduces a novel matrix operation of half tensor product multiplication into the neural network, establishes a neural network model based on a half tensor product, is applied to the processing of acoustic signals, can improve the identification rate of underwater targets, expands application scenes, and is suitable for identifying the underwater targets in complex marine environment noise.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps: receiving underwater acoustic signals including acoustic signals of underwater targets and ocean environment background noise without the underwater targets through an underwater sonar sensor; presenting time domain and frequency domain information of the acoustic signal in a LOFAR map through short-time Fourier transform; constructing a half tensor product neural network of the data sample by taking the LOFAR atlas sample as an input feature matrix; dividing the received underwater acoustic signals into a training set and a verification set, and inputting a half tensor product neural network for training and verification; selecting different hyper-parameters, performing model training on the half tensor product neural network by using a training set, comparing the test effect of a verification set, and determining the hyper-parameters with high test accuracy; and finally, inputting the currently acquired acoustic signal of the underwater target into the half tensor product neural network after model training, and giving a judgment result.
The underwater acoustic signalIn the formula, h (t) represents the impact response of an ocean channel, s (t) represents an underwater target signal, n (t) represents the noise of an ocean environment, represents convolution operation, and t is a time variable.
After the underwater sonar sensor receives underwater acoustic signals, a high-pass filter is designed to filter low-frequency-band noise, and then frame preprocessing is carried out.
The signal duration of each frame in the frame preprocessing is 1 s.
Said short-time Fourier transformWhere x (t) represents the signal, w (t) is a window function, which acts as a time-limit, e-jωtAnd plays a role of frequency limitation.
In the half-tensor product neural network described above,in the formula yiRepresenting the eigenmatrix, x, after a half-tensor product operationiA feature matrix representing the input, WiFormula weight matrix, biRepresenting an offset term, fc() Represents the activation function of the semi-tensor layer,representing a half tensor product operation; after the input features are subjected to feature extraction through half tensor product, the output feature graph is transmitted to a pooling layer for feature selection and information filtering, and the pooling processIs zi=βipool(yi)+biWherein pool () represents yiCarrying out a sampling operation ofiIndicating the multiplication bias.
The underwater acoustic signal is divided into K sub-samples, one single sub-sample is reserved as data of a verification model, and the other K-1 sub-samples are substituted to construct a half tensor product neural network for training; and repeating the cross validation for K times, validating each subsample once, and averaging the results of the K times to be used as the performance index of the evaluation classifier.
The invention has the beneficial effects that: because the neural network model is trained by data, a determined noise model cannot be assumed, errors brought to complex and variable marine environment noise modeling are reduced, the training and operation speed is higher, the robustness is better than that of a traditional machine learning algorithm, and meanwhile, the recognition rate of an underwater sound source target can be improved. The method can be suitable for various complex marine environmental noises, and has profound significance for underwater target intelligent sensing, marine exploration, underwater abnormal target detection and intelligent marine and marine defense construction. The invention successfully applies the artificial intelligence technology to ocean information perception and underwater target detection, and promotes the application and development of the technology in the fields of intelligent ocean, ocean defense and the like.
Drawings
FIG. 1 is a general method flow diagram of the present invention;
FIG. 2 is a schematic diagram of a half tensor product neural network structure of the present invention.
Detailed Description
The present invention will be further described with reference to the following drawings and examples, which include, but are not limited to, the following examples.
The method receives the acoustic signals of the underwater target through the underwater sonar sensor, acquires a large number of acoustic signals of different underwater targets such as ships, fishing boats, speed boats and the like and various ocean environment background noises, generates an LOFAR spectrogram of the acoustic signals through short-time Fourier transform, and divides data samples into a training set and a verification set to be input into a semi-tensor product neural network for training. And combining neural network parameter adjustment to optimize the effect based on the training set and the verification set. And finally, inputting the acoustic signal of the underwater target into a half tensor product neural network to give a judgment result.
As shown in fig. 1, the present invention comprises the steps of:
the first step is as follows: underwater sonar sensor receiving signal
The underwater sonar sensor is placed in the ocean environment, and received signals x (t) are as follows:
in the formula, h (t) represents the ocean channel impact response, s (t) represents the underwater target signal, n (t) represents the ocean environment noise, represents the convolution operation, and t is a time variable. I.e. no object is received as pure marine ambient noise.
The second step is that: signal pre-processing
Because the noise level of the ocean low-frequency environment is high, a high-pass filter is designed, low-frequency band noise is filtered, then frame preprocessing is carried out, and the signal duration of each frame is 1 s. The signal framing processing can increase the number of training samples and the length of the normalized sample signal, and is ready for obtaining visual spectrogram samples with uniform sizes in the next step.
The third step: obtaining acoustic signal time frequency visual LOFAR spectrogram
The time domain and frequency domain information of the acoustic signal is presented in a LOFAR map in a short-time Fourier transform mode:
where x (t) represents the signal (here 1s long received signal samples are represented), w (t) is a window function, which acts as a time limit, e-jωtAnd plays a role of frequency limitation.
The fourth step: constructing a semi-tensor product neural network
Replacing convolution operation in conventional convolution neural network with half tensor matrix multiplication product by a kernel matrixHalf tensor product multiplication is performed directly on each layer of input. Matrix half tensor product matrix multiplication can realize multiplication of two matrixes with any dimension. The general studies are left half tensor products. For a given two matrices A ∈ Mm×nAnd B ∈ Mp×qThen its half tensor product is:
Half tensor stacking: the process of the entire half tensor product can be expressed as follows:
in the formula yiRepresenting the eigenmatrix, x, after a half-tensor product operationiA feature matrix representing the input (with the LOFAR spectrogram sample obtained in the third step as the input feature matrix), WiFormula weight matrix, biRepresenting an offset term, fc() A semi-tensor layer-built activation function is represented,representing a half tensor product operation.
A pooling layer: after the input features are subjected to feature extraction by half tensor product, the output feature map is transmitted to a pooling layer for feature selection and information filtering, namely down-sampling. The process of pooling can be represented by the following formula:
zi=βipool(yi)+bi
wherein, pool () represents sampling yi, i.e. the pooling operation; beta is aiRepresenting the multiplication offset, biA bias term is represented.
Full connection layer: the fully-connected layer is usually built on the last part of the hidden layer of the convolutional neural network, and only signals are transmitted to other fully-connected layers. The output layer is connected behind the full connection layer and outputs the classification labels by using a logic function or a normalized exponential function.
The fifth step: constructing a training set sample set
A data set acquired by an underwater sonar sensor is divided into K sub-samples, an independent sub-sample is reserved as data of a verification model, and other K-1 samples are used for substituting to construct a half tensor product neural network for training. And repeating the cross validation for K times, validating each subsample once, and averaging the results of the K times to be used as the performance index of the evaluation classifier.
And a sixth step: adjusting hyper-parameter training optimization neural network model
By selecting different hyper-parameters, model training is carried out by using a training set, and the hyper-parameters with high testing accuracy are determined by comparing the testing effects of the testing set. The hyper-parameters comprise the learning rate of the neural network, the batch size, the iteration times, the half tensor product kernel size and the activation function. The method is used for optimizing the neural network model and improving the performance and effect of network learning.
The seventh step: realize underwater target recognition classification
After the data to be detected is preprocessed, a classification result can be given through training an optimized half tensor product neural network model, and underwater target recognition and classification are achieved.
Aiming at the problems of low underwater target recognition rate and weak robustness of the traditional method, the embodiment of the invention provides an underwater target recognition method based on a half tensor product neural network, which comprises the following steps:
the first step is as follows: underwater sonar sensor receiving signal
The underwater sonar sensor is placed in the ocean environment, and received signals x (t) are as follows:
in the formula, h (t) represents the ocean channel impact response, n (t) represents the ocean environment noise, x represents the convolution operation, and t is a time variable. I.e. no object is received as pure marine ambient noise.
The second step is that: signal pre-processing
Because the noise level of the ocean low-frequency environment is high, a second-order Butterworth filter is designed, and a digital Butterworth high-pass filter is deduced through bilinear transformation. And (2) the received sound data is subjected to a second-order Butterworth high-pass filter to obtain a filtered sound signal, low-frequency-band noise is filtered, then frame division preprocessing is carried out, and the time length of each frame of signal is 1 s.
The third step: obtaining acoustic signal time frequency visual LOFAR spectrogram
The time domain and frequency domain information of the acoustic signal is presented in a LOFAR map in a short-time Fourier transform mode:
where x (t) represents the signal, w (t) is a window function functioning as a time limit, e-jωtThe window function can select a Hanning window, the number of Fourier transform points is selected to be 1024, and the length of the window is selected to be 1024.
The fourth step: constructing a semi-tensor product neural network
The convolution operation in the conventional convolutional neural network is replaced by the half tensor matrix multiplication product, and half tensor product multiplication is directly carried out on each layer of input through a kernel matrix. Matrix half tensor product matrix multiplication can realize multiplication of two matrixes with any dimension. The general studies are left half tensor products. Given two matrices A ∈ Mm×nAnd B ∈ Mp×qThen its half tensor product is:
Half tensor stacking: the process of the entire half tensor product can be expressed as follows:
in the formula yiRepresenting the eigenmatrix, x, after a half-tensor product operationiA feature matrix representing the input, WiFormula weight matrix, biRepresenting an offset term, fc() A semi-tensor layer-built activation function is represented,representing a half tensor product operation.
A pooling layer: after the input features are subjected to feature extraction by half tensor product, the output feature map is transmitted to a pooling layer for feature selection and information filtering, namely down-sampling. The process of pooling can be represented by the following formula:
zi=βipool(yi)+bi
wherein pool () representsiPerforming down-sampling operation, namely the pooling operation; beta is aiRepresenting the multiplication offset, biRepresenting the bias term.
Full connection layer: the fully-connected layer is usually built on the last part of the hidden layer of the convolutional neural network, and only signals are transmitted to other fully-connected layers. The output layer is connected behind the full connection layer and outputs the classification labels by using a logic function or a normalized exponential function.
And constructing a half tensor product neural network with 5 convolutional layers, 5 pooling layers and 1 full-connection layer, wherein the number of neurons in an output layer is the number of classification categories.
The fifth step: constructing a training set sample set
Tagged data collected from an underwater sonar sensor is then segmented into K sub-sample sets, one individual sub-sample is retained as data for a verification model, and the other K-1 sample sets are used for training. And repeating the cross validation for K times, validating each subsample once, and averaging the results of the K times to be used as the performance index of the evaluation classifier.
And a sixth step: adjusting hyper-parameter training optimization neural network model
The neural network model is optimized by adjusting the learning rate, batch size, iteration times, half tensor product kernel size, activation function and other hyper-parameters of the neural network, and the performance and effect of network learning are improved. The learning rate can be selected from 0.0001-0.01, the batch size can be selected from 10-30, the iteration times are selected from 2-5, the half tensor product kernels are all selected from 3 x 3, the nonlinear excitation function is selected from a Relu function, and the pooling layer is selected from 2 x 2 average pooling.
The seventh step: realize underwater target recognition classification
After the data to be detected are preprocessed, a classification result can be given through a half tensor product neural network model which is well trained and optimized, and underwater target recognition and classification are achieved.
Claims (6)
1. An underwater target identification method based on a half tensor product neural network is characterized by comprising the following steps: receiving underwater acoustic signals including acoustic signals of underwater targets and ocean environment background noise without the underwater targets through an underwater sonar sensor; presenting time domain and frequency domain information of the acoustic signal in a LOFAR map through short-time Fourier transform; the LOFAR atlas is used as an input feature matrix, a half tensor product neural network which replaces convolution operation with half tensor products is constructed, convolution operation in a conventional convolutional neural network is replaced with half tensor matrix multiplication products, and the process of the whole half tensor product in half tensor lamination is expressed by the following formula:
in the formula yiRepresenting the eigenmatrix, x, after a half-tensor product operationiA feature matrix representing the input, WiIs a weight matrix, biRepresenting an offset term, fc() A semi-tensor layer-built activation function is represented,expressing half tensor product operation, after the input features are subjected to feature extraction through half tensor product, the output feature graph is transmitted to a pooling layer for feature selection and information filtering, and the pooling process is zi=βipool(yi)+biWherein pool () represents yiCarrying out a sampling operation ofiRepresents the multiply bias; dividing the received underwater acoustic signals into a training set and a verification set, and inputting a half tensor product neural network for training and verification; selecting different hyper-parameters, performing model training on the half tensor product neural network by using a training set, comparing the test effect of a verification set, and determining the hyper-parameters with high test accuracy; and finally, inputting the currently acquired acoustic signal of the underwater target into the half tensor product neural network after model training, and giving a judgment result.
2. The underwater target identification method based on the half tensor product neural network as claimed in claim 1, wherein: the underwater acoustic signalIn the formula, h (t) represents the ocean channel impact response, s (t) represents the underwater target signal, n (t) represents the ocean environment noise, represents the convolution operation, and t is a time variable.
3. The underwater target identification method based on the half tensor product neural network as claimed in claim 1, wherein: after the underwater sonar sensor receives underwater acoustic signals, a high-pass filter is designed to filter low-frequency-band noise, and then frame preprocessing is carried out.
4. The underwater target identification method based on the half tensor product neural network as claimed in claim 3, wherein: the signal duration of each frame in the frame preprocessing is 1 s.
5. The half tensor product based nerve of claim 1The underwater target identification method of the network is characterized in that: said short-time Fourier transformWhere x (t) represents the underwater acoustic signal, w (t) is a window function, which acts as a time-limit, e-jωtAnd plays a role of frequency limitation.
6. The underwater target identification method based on the half tensor product neural network as claimed in claim 1, wherein:
the underwater acoustic signal is divided into K sub-samples, an individual sub-sample is reserved as data of a verification model, and other K-1 samples are used for substituting the constructed half tensor product neural network for training; and repeating the cross validation for K times, validating each subsample once, and averaging the results of the K times to be used as the performance index of the evaluation classifier.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910513322.9A CN110245608B (en) | 2019-06-14 | 2019-06-14 | Underwater target identification method based on half tensor product neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910513322.9A CN110245608B (en) | 2019-06-14 | 2019-06-14 | Underwater target identification method based on half tensor product neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110245608A CN110245608A (en) | 2019-09-17 |
CN110245608B true CN110245608B (en) | 2022-05-17 |
Family
ID=67887015
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910513322.9A Active CN110245608B (en) | 2019-06-14 | 2019-06-14 | Underwater target identification method based on half tensor product neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110245608B (en) |
Families Citing this family (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110807365B (en) * | 2019-09-29 | 2022-02-11 | 浙江大学 | Underwater target identification method based on fusion of GRU and one-dimensional CNN neural network |
CN110837085B (en) * | 2019-11-14 | 2022-06-14 | 东南大学 | Fluctuation index calculation method for underwater target discrimination |
CN111401548B (en) * | 2020-03-03 | 2022-03-22 | 西北工业大学 | Lofar line spectrum detection method based on deep learning |
CN111523469B (en) * | 2020-04-23 | 2022-02-18 | 苏州浪潮智能科技有限公司 | Pedestrian re-identification method, system, equipment and computer readable storage medium |
CN111505650B (en) * | 2020-04-28 | 2022-11-01 | 西北工业大学 | HPSS-based underwater target passive detection method |
CN111624585A (en) * | 2020-05-21 | 2020-09-04 | 西北工业大学 | Underwater target passive detection method based on convolutional neural network |
CN111931412A (en) * | 2020-06-19 | 2020-11-13 | 中国船舶重工集团公司第七一五研究所 | Underwater target noise LOFAR spectrogram simulation method based on generative countermeasure network |
CN112163461B (en) * | 2020-09-07 | 2022-07-05 | 中国海洋大学 | Underwater target identification method based on multi-mode fusion |
CN112364779B (en) * | 2020-11-12 | 2022-10-21 | 中国电子科技集团公司第五十四研究所 | Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion |
CN112418181B (en) * | 2020-12-13 | 2023-05-02 | 西北工业大学 | Personnel falling water detection method based on convolutional neural network |
CN112885362B (en) * | 2021-01-14 | 2024-04-09 | 珠海市岭南大数据研究院 | Target identification method, system, device and medium based on radiation noise |
CN113109780B (en) * | 2021-03-02 | 2022-08-05 | 西安电子科技大学 | High-resolution range profile target identification method based on complex number dense connection neural network |
CN113189595A (en) * | 2021-05-07 | 2021-07-30 | 山东大学 | Neural network-based two-way echo target positioning method, equipment and storage medium |
CN113420668B (en) * | 2021-06-21 | 2024-01-12 | 西北工业大学 | Underwater target identification method based on two-dimensional multi-scale permutation entropy |
CN113420870B (en) * | 2021-07-04 | 2023-12-22 | 西北工业大学 | U-Net structure generation countermeasure network and method for underwater sound target recognition |
CN113466839B (en) * | 2021-09-03 | 2021-12-07 | 北京星天科技有限公司 | Side-scan sonar sea bottom line detection method and device |
CN115047408B (en) * | 2022-06-13 | 2023-08-15 | 天津大学 | Underwater multi-sound-source positioning method based on single-layer large convolution kernel neural network |
CN116647376B (en) * | 2023-05-25 | 2024-01-26 | 中国人民解放军军事科学院国防科技创新研究院 | Voiceprint information-based underwater acoustic network node identity authentication method |
CN116973901A (en) * | 2023-09-14 | 2023-10-31 | 海底鹰深海科技股份有限公司 | Algorithm application of time-frequency analysis in sonar signal processing |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006039658A (en) * | 2004-07-22 | 2006-02-09 | Hitachi Software Eng Co Ltd | Image classification learning processing system and image identification processing system |
CN102662167A (en) * | 2012-04-11 | 2012-09-12 | 西北工业大学 | Feature extraction method of radiated noise signal of underwater target |
CN106919921A (en) * | 2017-03-06 | 2017-07-04 | 重庆邮电大学 | With reference to sub-space learning and the gait recognition method and system of tensor neutral net |
CN107194404A (en) * | 2017-04-13 | 2017-09-22 | 哈尔滨工程大学 | Submarine target feature extracting method based on convolutional neural networks |
CN108846323A (en) * | 2018-05-28 | 2018-11-20 | 哈尔滨工程大学 | A kind of convolutional neural networks optimization method towards Underwater Targets Recognition |
CN109086824A (en) * | 2018-08-01 | 2018-12-25 | 哈尔滨工程大学 | A kind of sediment sonar image classification method based on convolutional neural networks |
CN109448707A (en) * | 2018-12-18 | 2019-03-08 | 北京嘉楠捷思信息技术有限公司 | Voice recognition method and device, equipment and medium |
CN109767785A (en) * | 2019-03-06 | 2019-05-17 | 河北工业大学 | Ambient noise method for identifying and classifying based on convolutional neural networks |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10809376B2 (en) * | 2017-01-06 | 2020-10-20 | Massachusetts Institute Of Technology | Systems and methods for detecting objects in underwater environments |
-
2019
- 2019-06-14 CN CN201910513322.9A patent/CN110245608B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006039658A (en) * | 2004-07-22 | 2006-02-09 | Hitachi Software Eng Co Ltd | Image classification learning processing system and image identification processing system |
CN102662167A (en) * | 2012-04-11 | 2012-09-12 | 西北工业大学 | Feature extraction method of radiated noise signal of underwater target |
CN106919921A (en) * | 2017-03-06 | 2017-07-04 | 重庆邮电大学 | With reference to sub-space learning and the gait recognition method and system of tensor neutral net |
CN107194404A (en) * | 2017-04-13 | 2017-09-22 | 哈尔滨工程大学 | Submarine target feature extracting method based on convolutional neural networks |
CN108846323A (en) * | 2018-05-28 | 2018-11-20 | 哈尔滨工程大学 | A kind of convolutional neural networks optimization method towards Underwater Targets Recognition |
CN109086824A (en) * | 2018-08-01 | 2018-12-25 | 哈尔滨工程大学 | A kind of sediment sonar image classification method based on convolutional neural networks |
CN109448707A (en) * | 2018-12-18 | 2019-03-08 | 北京嘉楠捷思信息技术有限公司 | Voice recognition method and device, equipment and medium |
CN109767785A (en) * | 2019-03-06 | 2019-05-17 | 河北工业大学 | Ambient noise method for identifying and classifying based on convolutional neural networks |
Non-Patent Citations (5)
Title |
---|
On Semi-tensor Product of Matrices and Its Applications;Daizhan Cheng 等;《Acta Mathematicae Applicatae Sinica》;20031231;第19卷(第2期);第219-228页 * |
Tensor Product Generation Networks for Deep NLP Modeling;Qiuyuan Huang 等;《Proceedings of NAACL-HLT 2018》;20181231;第1263-1273页 * |
Underwater Acoustic Target Recognition Based on Supervised Feature-Separation Algorithm;Xiaoquan Ke 等;《sensors》;20181207;第1-24页 * |
基于声呐图像的水下目标检测、识别与跟踪研究综述;郭戈 等;《控制与决策》;20180531;第33卷(第5期);第906-922页 * |
面向水下目标识别的卷积神经网络优化方法研究;顾正浩;《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》;20190415;第2019年卷(第04期);第I138-755页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110245608A (en) | 2019-09-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110245608B (en) | Underwater target identification method based on half tensor product neural network | |
CN112364779B (en) | Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion | |
CN110807365B (en) | Underwater target identification method based on fusion of GRU and one-dimensional CNN neural network | |
CN113707176B (en) | Transformer fault detection method based on acoustic signal and deep learning technology | |
CN110751044B (en) | Urban noise identification method based on deep network migration characteristics and augmented self-coding | |
CN113191178B (en) | Underwater sound target identification method based on auditory perception feature deep learning | |
Wei et al. | A method of underwater acoustic signal classification based on deep neural network | |
CN111931820A (en) | Water target radiation noise LOFAR spectrogram spectrum extraction method based on convolution residual error network | |
CN112183582A (en) | Multi-feature fusion underwater target identification method | |
CN110598737A (en) | Online learning method, device, equipment and medium of deep learning model | |
CN115830436A (en) | Marine organism intelligent detection method based on deep learning | |
CN113673323B (en) | Aquatic target identification method based on multi-deep learning model joint judgment system | |
Zhang et al. | MSLEFC: A low-frequency focused underwater acoustic signal classification and analysis system | |
CN111785262B (en) | Speaker age and gender classification method based on residual error network and fusion characteristics | |
CN117711442A (en) | Infant crying classification method based on CNN-GRU fusion model | |
CN117310668A (en) | Underwater sound target identification method integrating attention mechanism and depth residual error shrinkage network | |
CN117370832A (en) | Underwater sound target identification method and device based on Bayesian neural network | |
CN116884435A (en) | Voice event detection method and device based on audio prompt learning | |
CN109741733B (en) | Voice phoneme recognition method based on consistency routing network | |
CN112052880A (en) | Underwater sound target identification method based on weight updating support vector machine | |
CN116417011A (en) | Underwater sound target identification method based on feature fusion and residual CNN | |
CN114818789A (en) | Ship radiation noise identification method based on data enhancement | |
CN115267672A (en) | Method for detecting and positioning sound source | |
CN113488069A (en) | Method and device for quickly extracting high-dimensional voice features based on generative countermeasure network | |
CN112926383A (en) | Automatic target identification system based on underwater laser image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |