CN103323532B - Fish identification method and system based on psychoacoustics parameters - Google Patents

Abstract

The invention relates to a fish identification method and system based on psychoacoustics parameters. The method is an identification strategy based on an active acoustics method, and specifically comprises: step 101) transmitting a sound signal to an underwater fish body, and acquiring a backscatter echo signal of the underwater fish body; step 102) carrying out filtering and amplitude normalization pretreatments for the fish body echo signal; step 103) extracting characteristics from the pretreated echo signal, and the specific method is as follows: carrying out frequency shifting for the pretreated echo signal, and shifting the frequency into a human audible frequency range, then dividing the frequency-shifted signal into a plurality of frequency bands according to a human auditory model, calculating specific loudness of various frequency bands, and taking the specific loudness as characteristic quantities; step 104) inputting the characteristic quantities to a classifier, classifying, and finishing species identification of the underwater fish body.

Description

A kind of fish identification method based on psychologic acoustics parameter and system

Technical field

The present invention relates to the fish recognition technology based on acoustic method, particularly a kind of fish identification method based on psychologic acoustics parameter and system.

Background technology

Along with the mankind are to the increase day by day of ocean resources demand, marine fishery resources exploitation more and more comes into one's own.Fishery resources survey and evaluation work are the important foundations of reasonable development marine fishery resources, and this possesses the ability of the kind of fish being carried out to identification fast with regard to needing urgently.

Traditional resource investigation of fish method mainly based on trawl fishing method, compared with the methods such as traditional trawl fishing identification, acoustic method have quick and convenient, do not damage the advantage such as living resources and sustainable observation.Therefore, external related research institutes greatly develops the fish recognition technology based on acoustic method in the last few years.

Wherein, the fish recognition technology of external acoustic method comprises: the people such as Alexander B.Kulinchenko use rope system method to test Pacific Ocean halibut and rockfish, and utilize echo envelope and statistics spectrum signature two kinds of methods successfully to classify to halibut, rockfish, seabed, but this method is carried out down-sampled rear as proper vector due to the direct echo envelope signal to fish body, fail to find out the characteristic quantity reflecting target essence, intrinsic dimensionality is very high, redundancy is comparatively large, brings very large burden to sorter.Harold M.Brundage III and Jae-Byung Jung utilizes the method for statistics frequency spectrum from demersal fishes, identify brachyrhinia sturgeon.The people such as Eric O.Rogers utilize the method for statistics frequency spectrum successfully to identify catfish, rain herrian and salmon.The method of the equal Corpus--based Method frequency spectrum of above-mentioned fish recognition technology, therefore needs to have very wide band transducer as emissive source, higher to equipment requirement.In addition, the people such as Sunardi determine two kinds of different scad target strengths under two kinds of frequencies by echo sounder, successfully classify to the kind of fish.But this method utilizes the target strength of fish body under two kinds of frequencies to be identified amount, and the factor such as size, shape of the attitude of fish body target strength and fish body, air bladder is relevant, under only selecting two frequency bins, target strength is as characteristic quantity, and feature is unstable, uses limitation larger.In a word, people carry out fish identification by multiple method, but the echoed signal of fish body is very complicated, and existing recognition methods fails to find out the feature that can reflect target essence, makes intrinsic dimensionality high, and redundancy is large, and Classification and Identification effect often can not be satisfactory.And at home, the research based on the fish recognition technology of acoustic method is still in the starting stage at present.

Research shows, having the ability to tell the sound needing to listen in the speech environment of people's ear more than two people, can be extract useful voice messaging the noisy speech signal of-12dB from signal to noise ratio (S/N ratio).As can be seen here, people's ear to the feature extraction of sound and the Classification and Identification ability of human brain to sound quite strong.Psychologic acoustics parameter describes the objective physical amount that alternative sounds signal institute causes subjective feeling difference degree, adopts psychologic acoustics parametric analysis acoustical signal, can the difference of quantitative test auditory perception, eliminates the impact of individuality.

First the present invention carries out frequency translation to fish echo signal, makes its frequency spectrum fall into audible sound frequency band range, then to the signal extraction psychologic acoustics parameter after process, and it can be used as characteristic quantity.Extract the feature that characteristic quantity can reflect fish body preferably, and intrinsic dimensionality is few, can realize Fast Classification.

Summary of the invention

The object of the invention is to, for overcoming prior art, feature extraction is carried out to fish echo signal, cause the intrinsic dimensionality of the echoed signal obtained high, redundancy is large, Classification and Identification effect often can not be satisfactory etc. problem, thus provide a kind of fish identification method based on psychologic acoustics parameter and system.

For achieving the above object, the invention provides a kind of fish identification method based on psychologic acoustics parameter, the method is a kind of recognition strategy based on active acoustical method, and described method specifically comprises:

Step 101) to the acoustical signal of fish body transmitting under water, and obtain the backscattered echoed signal of fish body under water;

Step 102) filtering and amplitude normalization pre-service are carried out to fish echo signal;

Step 103) feature extraction is carried out to pretreated echoed signal, concrete grammar is as follows:

Frequency translation is carried out to pretreated echoed signal, by its frequency translation to people's ear audible sound scope, then according to human auditory model, some frequency ranges are divided into the signal after frequency translation and the loudness calculating each frequency range as characteristic quantity;

Step 104) characteristic quantity inputted sorter and carry out classification process, complete the category identification to fish body under water.

Optimize, described characteristic quantity also comprises: the total loudness obtained by the loudness of each frequency range and/or total sharpness.

Optionally, described human auditory model adopts Zwicker model.

When adopting Zwicker model, described loudness adopts following formulae discovery to obtain:

$N^{\′}\left(z\right)=0.08{\left(\frac{E_{\mathrm{TQ}}}{E_0}\right)}^{0.23}[{(0.5+0.5\frac{E}{E_{\mathrm{TQ}}})}^{0.23}-1]$

Wherein, E _tQfor the excitation that the threshold of audibility under quiet situation is corresponding; E ₀for reference sound intensity I ₀corresponding excitation, and its value is 10 ^-12watt/meter ²; E is calculated excitation corresponding to sound, and z is the 20Hz-16000Hz frequency partition that can be heard by people's ear according to the definition of Zwicker theory by 24 critical bands are obtained each critical band.

Optionally, described total sharpness adopts following formulae discovery to obtain:

$S=0.11\frac{{\∫}_0^{24}{N}^{\′}\left(z\right)g\left(z\right)\mathrm{dz}}{{\∫}_0^{24}{N}^{\′}\left(z\right)\mathrm{dz}};$

Wherein, g (z) is an additional factor, and computing formula is as follows:

$g\left(z\right)=\left\{\begin{array}{cc}1& z\&le;16\\ 0.06e^{0.17\mathrm{lz}}& 16<z\&le;24\end{array}\right..$

Based on said method, present invention also offers a kind of fish recognition system based on psychologic acoustics parameter, this system comprises: for the transmitting terminal subsystem of fish body emission sound source signal and the receiving terminal system for identifying fish body analogy under water under water, and described receiving terminal system comprises further: the echoed signal acquisition module of fish body and sorter, it is characterized in that, described receiving terminal system also comprises:

Pretreatment module, for carrying out filtering and amplitude normalization pre-service to the fish echo data collected;

Characteristic extracting module, for carrying out feature extraction to pretreated echoed signal, concrete grammar is:

Frequency translation is carried out to pretreated echoed signal, by its frequency translation to people's ear audible sound scope, then according to human auditory model, some frequency ranges are divided into the signal after frequency translation and the loudness calculating each frequency range as characteristic quantity; With

Sort module, the proper vector input sorter for characteristic extracting module being exported carries out the category identification of fish body.

Optimize, total loudness that the loudness by each frequency range also obtains by described characteristic extracting module and/or total sharpness are as characteristic quantity.

Optionally, when adopting Zwicker model, described characteristic extracting module adopts the characteristic loudness value of each frequency range of following formulae discovery:

$N^{\&prime;}\left(z\right)=0.08{\left(\frac{E_{\mathrm{TQ}}}{E_0}\right)}^{0.23}[{(0.5+0.5\frac{E}{E_{\mathrm{TQ}}})}^{0.23}-1]$

Wherein, E _tQfor the excitation that the threshold of audibility under quiet situation is corresponding; E ₀for reference sound intensity I ₀corresponding excitation, and its value is 10 ^-12watt/meter ²; E is calculated excitation corresponding to sound, and z is the 20Hz-16000Hz frequency partition that can be heard by people's ear according to the definition of Zwicker theory by 24 critical bands are obtained each critical band.

In technique scheme, described characteristic extracting module comprises further:

Frequency spectrum shift module, for carrying out frequency translation to pretreated fish echo signal, within the scope of the sound can listen its frequency translation to people's ear; With

Processing module, for calculating the psychoacoustic parameter of the signal that frequency spectrum shift module exports, described psychoacoustic parameter comprises: loudness, total loudness and/or sharpness.

Compared with prior art, technical advantage of the present invention is:

(1) data source needed for is the echo data utilizing narrow-band transducer to obtain as emissive source, low to the cost requirement of equipment;

(2) intrinsic dimensionality is few, and redundancy is few, can realize quick and precisely classifying.

Technical scheme of the present invention utilizes simple device in a word, reduce equipment cost, extract the characteristic quantity that can reflect fish bulk properties in fish echo signal simultaneously, thus can intrinsic dimensionality be reduced, reduce feature redundancy, final realization identifies fast and accurately to the kind of fish.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the fish identification method based on psychologic acoustics parameter of the present invention.

Embodiment

Below in conjunction with drawings and the specific embodiments, the invention will be further described.

Fish echo signal transacting: the sonar system being usually used in fishery resources survey, its frequency of sound wave is general higher, beyond the audible sound scope of people's ear, so directly can not carry out the extraction of psychologic acoustics parameter to fish echo signal.Here, need first frequency translation is carried out to echoed signal, namely with the place frequency of fish echo signal for foundation, be multiplied by the cosine signal of a certain frequency, and carry out low-pass filtering, obtain the echoed signal within the scope of audible sound.Such as: suppose that the centre frequency of fish echo signal is 200KHz, frequency coverage is 190KHz-210KHz, then echoed signal can be multiplied by the cosine signal that frequency is 200KHz, and carry out low-pass filtering, filter cutoff frequency is 10kHz, then obtain the baseband signal of fish echo signal, its frequency range is just within the scope of audible sound.

The fish identification method based on psychologic acoustics parameter recorded in order to the content part that carries out an invention the invention provides a kind of fish recognition system based on psychologic acoustics parameter, this system comprises: for the transmitting terminal subsystem to fish body generation sound-source signal under water and the receiving terminal system for identifying fish body kind under water, and described receiving terminal system comprises further: the echoed signal acquisition module of fish body and sorter, compared with prior art receiving terminal system provided by the invention also comprises:

Pretreatment module, for carrying out filtering and amplitude normalization pre-service to the fish echo data collected;

Characteristic extracting module, for carrying out feature extraction to pretreated echoed signal, concrete grammar is: carry out frequency translation to pretreated echoed signal, within the scope of the sound that its frequency translation to people's ear can be listened, again to signal estimated performance loudness, the always degree of thinking after frequency translation and total sharpness, and it can be used as characteristic quantity; With

Sort module, the proper vector input sorter for characteristic extracting module being exported carries out the category identification of fish body.

Above-mentioned characteristic extracting module comprises further:

Frequency spectrum shift module, for carrying out frequency translation to pretreated fish echo signal, by its frequency translation within the scope of people's ear audible sound; And processing module, for calculating the psychoacoustic parameter of the signal that frequency spectrum shift module exports, this psychoacoustic parameter comprises: loudness, total loudness and total sharpness.

Above-mentioned transmitting terminal subsystem specifically adopts narrow-band transducer to the acoustical signal of fish body transmitting under water.

The present invention propose the fish identification method based on psychologic acoustics parameter be implemented as follows description:

Step 101) to the acoustical signal of fish body transmitting under water, and obtain the backscattered echoed signal of fish body under water;

Above-mentioned both can be broadband emission source to the emissive source that fish body emission sound source signal adopts under water also can be narrow emission source, but considers from cost-saving angle, and suggestion uses narrow emission source.

Step 102) filtering and amplitude normalization pre-service are carried out to fish echo signal;

Step 103) feature extraction is carried out to pretreated echoed signal, concrete grammar is as follows:

Frequency translation is carried out to pretreated echoed signal, by its frequency translation to people's ear audible sound scope, then according to human auditory model, some frequency ranges are divided into the signal after frequency translation and the loudness calculating each subsegment as characteristic quantity;

Wherein, described human auditory model is including but not limited to employing Zwicker model, and those skilled in the art also can adopt Moore model.

Following examples we adopt Zwicker model to describe the leaching process of psychoacoustic parameter in detail.

The extracting method of 1 psychologic acoustics parameter

Psychologic acoustics parameter be describe alternative sounds signal cause the objective physical amount of subjective feeling difference degree.Adopt psychologic acoustics parametric analysis acoustical signal, can the difference of quantitative test auditory perception, eliminate individual impact.For people, to the sensation of sound with judge that identifying not is the logistics capacity such as acoustic pressure, the sound intensity, acoustical power adopting method for objectively evaluating, but carry out according to the psychoacoustic parameter of subjective assessment standard.Main psychoacoustic parameter has: loudness, sharpness, roughness, cymomotive force etc., the present invention mainly uses loudness and sharpness feature.

1.1 loudness

Loudness is that reflection people ear is to the psychoacoustic parameter of sound intensity subjective feeling.It depends on the size of sonic wave amplitude, simultaneously relevant with frequency.The unit of loudness is Song (sone).The calculating of loudness is relevant with the masking effect of people's ear.According to the definition of Zwicker theory, the 20Hz-16000Hz frequency partition can heard by people's ear is 24 critical bands (Bark), and the conversion formula between frequency (Hz) and critical band (Bark) is as follows:

Z＝13*arctan(0.00076f)+3.5*arctan(f/7500) ²(1)

Wherein, Z is critical band, and f is frequency.The relation dividing critical band and the frequency obtained is as shown in table 1.

The relation of table 1 critical band and frequency

As shown in Table 1, according to the theoretical division to frequency band of Zwicker, in low-frequency range, frequency band division is relatively more even, and precision is higher; And at high band, frequency span presents nonlinear increase, frequency resolution reduces gradually.During for fish echo signal analysis, meticulous analysis can be done to the signal of low-frequency range.

Loudness is the loudness in each critical band, and its computing formula is as follows:

$N^{\&prime;}\left(z\right)=0.08{\left(\frac{E_{\mathrm{TQ}}}{E_0}\right)}^{0.23}[{(0.5+0.5\frac{E}{E_{\mathrm{TQ}}})}^{0.23}-1]---\left(2\right)$

Wherein, E _tQfor the excitation that the threshold of audibility under quiet situation is corresponding; E ₀for reference sound intensity I ₀=10 ^-12w/m ²corresponding excitation; E is calculated excitation corresponding to sound.

Total loudness value N be each critical band internal characteristic loudness and:

$N={\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)\mathrm{dz}---\left(3\right)$

1.2 sharpness

Sharpness is a weighting square of loudness, and general computing method adopt the frequency spectrum of critical band to loudness weighted integral, and it is the psychoacoustic parameter describing radio-frequency component proportion in sound spectrum, and the ear-piercing degree of reflection sound, unit is acum.Computing formula is as follows:

$S=0.11\frac{{\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)g\left(z\right)\mathrm{dz}}{{\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)\mathrm{dz}}---\left(4\right)$

In formula, g (z) is an additional factor, and its value changes with the change of critical band, and computing formula is as follows:

$g\left(z\right)=\left\{\begin{array}{cc}1& z\&le;16\\ 0.06e^{0.17\mathrm{lz}}& 16<z\&le;24\end{array}\right.---\left(5\right)$

Wherein, above-mentioned loudness can be classified as characteristic quantity input sorter separately, but in order to total loudness and/or total sharpness can also be inputted sorter as characteristic quantity by the accuracy rate of further Optimum Classification identification, carry out the Classification and Identification of fish body under water.

Step 104) characteristic quantity inputted sorter and carry out classification process, complete the category identification to fish body under water.

This step adopts sorter to carry out Classification and Identification according to the characteristic quantity of the reaction fish body characteristics to be measured extracted to fish body under water.Described sorter can be chosen BP neural network classifier and classify, and this sorting technique is the sorting technique having supervision, that is: know training sample generic in advance, then trains sorter according to appointment classification.Specific to this programme, need first to the fish echo signal of a large amount of Known Species according to step 101)-103) complete the extraction of characteristic quality of sample, these characteristic quantities are sent into BP neural network classifier as training sample characteristic quantity, and then sorter is trained, sorter finally can be completed the fish echo signal of Known Species is sorted out.After having trained, preserve sorter.Underwater target echo completing steps 101 to the unknown)-103), obtain the characteristic quantity of unknown object, and characteristic quantity sent in the sorter trained and classify, complete the category identification to fish body under water.

In sum, the treatment scheme of fish body signal provided by the invention can be represented by Fig. 1, and in order to realize optimum technique effect, the concrete steps of fish body identification are as follows under water:

(1) carry out pre-service to fish echo signal, comprise filtering and normalization, wherein, filtering is in order to cake resistancet external noise; Normalization is in order to follow-up process is convenient;

(2) frequency translation is carried out to pretreated echoed signal, by its frequency translation to audible sound scope;

(3) to the signal after frequency translation according to formula (2), formula (3) and formula (4) estimated performance loudness, total loudness and total sharpness, and it can be used as characteristic quantity;

(4) according to extracted characteristic quantity, Classification and Identification is completed.

Wherein, the sorter described in step (4) can be chosen BP neural network classifier and classify, and this sorting technique is the sorting technique having supervision, that is: know training sample generic in advance, then trains sorter according to appointment classification.Specific to this programme, need the extraction first the fish echo signal of a large amount of Known Species being completed to characteristic quality of sample according to step (1)-(3), these characteristic quantities are sent into BP neural network classifier as training sample characteristic quantity, and then sorter is trained, sorter finally can be completed the fish echo signal of Known Species is sorted out.After having trained, preserve sorter.To underwater target echo completing steps (1)-(3) of the unknown, obtain the characteristic quantity of unknown object, and characteristic quantity is sent in the sorter trained and classify, complete the category identification to fish body under water.

It should be noted that, embodiment of the present invention of above introduction and and unrestricted.It will be understood by those of skill in the art that any amendment to technical solution of the present invention or the equivalent alternative spirit and scope not departing from technical solution of the present invention, it all should be encompassed in right of the present invention.

Claims (10)

1., based on a fish identification method for psychologic acoustics parameter, the method is a kind of recognition strategy based on active acoustical method, and described method specifically comprises:

Step 101) to the acoustical signal of fish body transmitting under water, and obtain the backscattered echoed signal of fish body under water;

Step 102) filtering and amplitude normalization pre-service are carried out to fish echo signal;

Step 103) feature extraction is carried out to pretreated echoed signal, concrete grammar is as follows:

Frequency translation is carried out to pretreated echoed signal, by its frequency translation to people's ear audible sound scope, more some frequency ranges is divided into according to human auditory model to the signal after frequency translation, and the loudness calculating each frequency range is as characteristic quantity;

Step 104) characteristic quantity inputted sorter and carry out classification process, complete the category identification to fish body under water.

2. the fish identification method based on psychologic acoustics parameter according to claim 1, is characterized in that, described characteristic quantity also comprises: the total loudness obtained by the loudness of each frequency range and/or total sharpness.

3. the fish identification method based on psychologic acoustics parameter according to claim 1 and 2, is characterized in that, described human auditory model adopts Zwicker model.

4. the fish identification method based on psychologic acoustics parameter according to claim 3, is characterized in that, described loudness adopts following formulae discovery to obtain:

$N^{\&prime;}\left(z\right)=0.08{\left(\frac{E_{\mathrm{TQ}}}{E_0}\right)}^{0.23}[{(0.5+0.5\frac{E}{E_{\mathrm{TQ}}})}^{0.23}-1]$

Wherein, E _tQfor the excitation that the threshold of audibility under quiet situation is corresponding; E ₀for reference sound intensity I ₀corresponding excitation, and its value is 10 ^-12watt/meter ²; E is calculated excitation corresponding to sound, and z is the 20Hz-16000Hz frequency partition that can be heard by people's ear according to the definition of Zwicker model theory by 24 critical bands are obtained each critical band.

5. the fish identification method based on psychologic acoustics parameter according to claim 4, is characterized in that, described total sharpness adopts following formulae discovery to obtain:

$S=0.11\frac{{\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)g\left(z\right)\mathrm{dz}}{{\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)\mathrm{dz}};$

Wherein, g (z) is an additional factor, and computing formula is as follows:

$g\left(z\right)=\left\{\begin{array}{cc}1& z\&le;16\\ 0.06e^{0.17\mathrm{lz}}& 16<z\&le;24\end{array}\right..$

6. the fish recognition system based on psychologic acoustics parameter, this system comprises: for the transmitting terminal subsystem to fish body emission sound source signal under water and the receiving terminal system for identifying fish body kind under water, and described receiving terminal system comprises further: the echoed signal acquisition module of fish body and sorter, it is characterized in that, described receiving terminal system also comprises:

Pretreatment module, for carrying out filtering and amplitude normalization pre-service to the fish echo data collected;

Characteristic extracting module, for carrying out feature extraction to pretreated echoed signal, concrete grammar is:

Frequency translation is carried out to pretreated echoed signal, by its frequency translation to people's ear audible sound scope, then according to human auditory model, some frequency ranges are divided into the signal after frequency translation and the loudness calculating each frequency range as characteristic quantity; With

Sort module, the proper vector input sorter for characteristic extracting module being exported carries out the category identification of fish body.

7. the fish recognition system based on psychologic acoustics parameter according to claim 6, is characterized in that, total loudness that the loudness by each frequency range also obtains by described characteristic extracting module and/or total sharpness are as characteristic quantity.

8. the fish recognition system based on psychologic acoustics parameter according to claim 6, is characterized in that, when adopting Zwicker model, described characteristic extracting module adopts the characteristic loudness value of each frequency range of following formulae discovery:

$N^{\&prime;}\left(z\right)=0.08{\left(\frac{E_{\mathrm{TQ}}}{E_0}\right)}^{0.23}[{(0.5+0.5\frac{E}{E_{\mathrm{TQ}}})}^{0.23}-1]$

Wherein, E _tQfor the excitation that the threshold of audibility under quiet situation is corresponding; E ₀for reference sound intensity I ₀corresponding excitation, and its value is 10 ^-12watt/meter ²; E is calculated excitation corresponding to sound, and z is the 20Hz-16000Hz frequency partition that can be heard by people's ear according to the definition of Zwicker theory by 24 critical bands are obtained each critical band.

9. the fish recognition system based on psychologic acoustics parameter according to claim 8, is characterized in that, described characteristic extracting module also adopts the total sharpness of following formulae discovery:

$S=0.11\frac{{\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)g\left(z\right)\mathrm{dz}}{{\&Integral;}_0^{24}{N}^{\&prime;}\left(z\right)\mathrm{dz}}$

Wherein, g (z) is an additional factor, and computing formula is as follows:

$g\left(z\right)=\left\{\begin{array}{cc}1& z\&le;16\\ 0.06e^{0.17\mathrm{lz}}& 16<z\&le;24\end{array}\right..$

10. the fish recognition system based on psychologic acoustics parameter according to claim 6, it is characterized in that, described characteristic extracting module comprises further:

Frequency spectrum shift module, for carrying out frequency translation to pretreated fish echo signal, by its frequency translation within the scope of people's ear audible sound; With

Processing module, for calculating the psychoacoustic parameter of the signal that frequency spectrum shift module exports, described psychoacoustic parameter comprises: loudness, total loudness and/or sharpness.