CN109886214A - A kind of chirm characteristic strengthening method based on image procossing - Google Patents
A kind of chirm characteristic strengthening method based on image procossing Download PDFInfo
- Publication number
- CN109886214A CN109886214A CN201910139801.9A CN201910139801A CN109886214A CN 109886214 A CN109886214 A CN 109886214A CN 201910139801 A CN201910139801 A CN 201910139801A CN 109886214 A CN109886214 A CN 109886214A
- Authority
- CN
- China
- Prior art keywords
- spectrogram
- chirm
- image
- noise
- birds
- 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.)
- Withdrawn
Links
Landscapes
- Image Analysis (AREA)
Abstract
The chirm characteristic strengthening method based on image procossing that the invention discloses a kind of, is related to image procossing, deep learning and birds identification technology.This method is: 1. pre-processing to chirm data set, including resampling and normalization, obtains the spectrogram of chirm；2. carrying out SNR estimation and compensation to spectrogram, signal spectrogram and noise spectrum figure are obtained；3. carrying out piecemeal to all spectrograms；4. carrying out data enhancing to the signal spectrogram after piecemeal: 5. being mapped by Jet and carry out Pseudo Col ored Image to the enhanced gray level image of data, obtain triple channel RGB color image；6. obtaining recognition result by transfer learning mode.The beneficial effects of the present invention are: 1. for efficiently carrying out birds automatic identification；2. passing through the biological spectrum information that the chirping of birds acoustical signal of one-dimensional time domain is converted to two-dimentional time-frequency domain；It is that deep learning is further in the exploration of birds automatic identification field 3. image procossing includes: chirm SNR estimation and compensation, specific data enhancing and the enhancing of visual perception power.
Description
Technical field
The present invention relates to image procossing, deep learning and birds identification technology fields, are based on image more particularly to one kind
The chirm characteristic strengthening method of processing.
Background technique
Bio-diversity is condition for the survival of mankind, is the strategic resource of social sustainable development, is ecological safety
With the important leverage of grain security.Due to having a very wide distribution, complete and to environment the sensibility of research data, birds are that biology is more
The important instruction monoid of sample.Birds population status and its dynamic-change information are grasped, is all had to protecting and assessing the ecosystem
It is significant.Birdvocalization is varied, contains important behavior meaning, species specificity and biology abundant letter
Breath, while being also the main means that species identification is carried out to it.The significance for taxonomy of chirm, have become at present bird sound research with
The hot spot that the research of birds systematics intersects.
The important component that biological spectrum is analyzed as Acoustic landscape ecological theory is monitoring, research and analysis ecology
The important channel of system Biodiversity.The time-frequency characteristic that can sufficiently show sound is analyzed it using method appropriate, from
And efficiently identify out the otherness of biology.Important channel of the chirping of birds sonograph as the species attribute of research bird, frequency spectrum point
Analysis monitors Avian diversity most important.
Deep learning is the most flourishing branch of current machine learning discipline development, and is applied in entire artificial intelligence field
Prospect technology the most wide.As deep learning is in the continuous breakthrough of computer vision field, image recognition is handled with it
Task has become a kind of efficient and professional technology.Under the support of the technical background, present invention combination chirping of birds sound characteristics, to bird
Song spectrogram is targetedly handled.
Summary of the invention
The present invention is intended to provide a kind of chirm characteristic strengthening method based on image procossing, is being known with solving the prior art
Song characteristics during not do not protrude, noise jamming is serious and the incomplete problem of biological spectrum information.
To achieve the above object, the technical solution of the present invention is as follows:
Specifically, a kind of chirm characteristic strengthening method based on image procossing the following steps are included:
1. pre-processing to chirm data set, including resampling and normalization, the spectrogram of chirm is obtained；
2. carrying out SNR estimation and compensation to spectrogram, signal spectrogram and noise spectrum figure are obtained；Wherein, signal spectrogram includes and sings and pipe
Part, noise spectrum figure include noise and mute part, and signal spectrogram is used for ambient noise as original training sample, noise spectrum figure
A kind of approach of enhancing；
3. carrying out piecemeal to all spectrograms after step 2. middle SNR estimation and compensation, and the size of every fritter spectrogram is adjusted to be suitble to
Input dimension for trained neural network；
4. data enhancing is carried out to the signal spectrogram after step 3. middle piecemeal, since spectrogram is different from traditional images, between
Difference limits the direct application of extensive image processing techniques；The characteristic of comprehensive consideration chirm and spectrogram, notebook data increase
Strength reason adhoc approach includes frequency-domain transform, noise addition and the mixing of similar sample:
5. for the visual perception power of enhancing gray level image, while conveniently carrying out transfer learning to different neural networks, pass through Jet
Mapping carries out Pseudo Col ored Image to the enhanced gray level image of data, triple channel RGB color image is obtained, by these triple channels
RGB color image is divided into training set and test set, and wherein training set accounts for 80%, and test set accounts for 20%；
6. suitable neural network model is chosen by transfer learning mode, it is finely adjusted and use step 5. in instruction
Practice collection to be trained, collects verifying model accuracy rate eventually by verifying, obtain recognition result.
Compared to the prior art, the beneficial effects of the present invention are:
1. proposing a kind of bird based on image procossing for the species taxonomy problem based on chirm under nature complexity acoustic enviroment
Song characteristics intensifying method；Alleviate that ambient noise of the chirm in identification process is high, song characteristics are not prominent, song data
The problems such as uneven and biological spectrum information is not comprehensive, for efficiently carrying out birds automatic identification；
2. by the biological spectrum information that the chirping of birds acoustical signal of one-dimensional time domain is converted to two-dimentional time-frequency domain, then flexibly using a system
The method of column image procossing highlights the chirping of birds acoustic signature in spectrogram, and enhancing biological spectrum information visuallization is presented, and can be led to
The method for crossing deep learning is verified；
3. image procossing includes: chirm SNR estimation and compensation, specific data enhancing and the enhancing of visual perception power；The present invention is different from
The image classification method of general visualization tasks is made and being directed to always around each presentation for identifying chirping of birds acoustic signature in link
Property strong image procossing strategy, be that deep learning is further in the exploration of birds automatic identification field.
Detailed description of the invention
Fig. 1 is the flow chart of this method.
Specific embodiment
One, method
1, chirm SNR estimation and compensation
2. step carries out SNR estimation and compensation to spectrogram, obtain signal spectrogram and noise spectrum figure；
The separation method of signal spectrogram are as follows: one threshold value N of setting, if some pixel value is higher than corresponding line and its phase in spectrogram
N times of the intermediate value that should be arranged then is set to 1, is otherwise set to 0；
The separation method of noise spectrum figure are as follows: one threshold value n (n < N) of setting, if some pixel value is higher than corresponding line and its accordingly
N times of the intermediate value of column is then set to 0, is otherwise set to 1.
2, specific data enhances
4. step carries out data enhancing, including the addition of frequency-domain transform, noise and the mixing of similar sample to the signal spectrogram after piecemeal；
A, frequency-domain transform
1. 2. a, the pitch of original chirping of birds sound audio is changed at random, amplitude of fluctuation is no more than 5%, and then repeatedly step
③；
1. 2. b, the volume of original chirping of birds sound audio is changed at random, amplitude of fluctuation is no more than 5%, and then repeatedly step
③；
B, noise adds
A, by random Gaussian be added to step 2. in signal spectrogram, the image that new standardized of laying equal stress on obtains；
B, by step 2. in noise section be added at random step 2. in signal spectrogram, as training sample；
C, similar sample mixing
Signal spectrogram after the different audio SNR estimation and compensations of same bird is mixed at random.
3, visual perception power enhances
For the visual perception power for further enhancing gray level image, while in view of transfer learning is to data dimension before mode input
It limits, is mapped by Jet and carry out Pseudo Col ored Image, increase the contrast between varying strength region to improve recognition performance.No
Be mapped to three monochrome images of red, green, blue respectively with region, and it is corresponding it is high, in and low-power spectrum information, red indicate most
The sound property of high-energy is approximately to sing/pipe characteristic.
According to the image processing method of above-mentioned chirm spectrogram, chooses mixed model SE-ResNeXt-50 and migrated
Study can obtain accurate efficient recognition result from a large amount of chirping of birds sonographs.
Two, embodiment
This method is the chirping of birds acoustical signal of one-dimensional time domain to be converted to the biological spectrum information of two-dimentional time-frequency domain, then flexibly use one
The method of image series processing highlights the chirping of birds acoustic signature in spectrogram, compared to general general image-recognizing method, this hair
Bright more specific aim and identification high efficiency.
Experimental data derives from Xeno-Canto database, and the most of audio file sample rates of the database are 44.1kHz,
16bit, monophonic, the also unified standard as Primary Stage Data format.
1. pre-processing to known chirm data set, resampling is 44.1kHz sample rate, using with Hanning window
The Short Time Fourier Transform (STFT) of function calculates the spectrogram of chirm, and carries out maximum value normalization to spectrogram, makes frequency
The dynamic range of spectrum information is mapped in [0,1] range, then handles spectrogram for gray level image.
2. carrying out SNR estimation and compensation to spectrogram: signal spectrogram includes to sing and pipe part, noise spectrum figure include noise and
Mute part；In most of chirping of birds sound audio, the amplitude of prospect chirping of birds acoustical signal is higher than ambient noise；We utilize this rule
Rule reduces ambient noise to isolate signal spectrogram: one threshold value N of setting, if some pixel value is higher than corresponding line in spectrogram
And its N times of the intermediate value of respective column, then it is set to 1, is otherwise set to 0.This way approximation has highlighted all heavy in spectrogram
The chirping of birds acoustical signal wanted, because high amplitude generally corresponds to singing or piping for birds；Noise in different frequency region simultaneously
Level is compensated and is reduced, and the WBD wideband distortion as caused by the ambient noise of uncontrollable factor is attenuated；
For the step reasons for its use noise, filter is corroded and expanded to application binary to eliminate noise and linkage section, or
The means of bound fraction morphological image process；
Separation for noise spectrum figure, we follow similar step: one threshold value n (n < N) of setting, if some pixel value is high
In n times of corresponding line and its intermediate value of respective column, then it is set to 0, is otherwise set to 1；With the institute of the separating step of signal spectrogram
It is because threshold value N has been to highlight the excessive selection that signal section is suitably made to use different threshold values, it is intended that
There is provided a safety thus extenuates leeway, and the signal for being in the buffer area neither has clearly song characteristics, nor affects on
It is subsequently used for carrying out the information content of the noise section of data enhancing；
To sum up, all the elements for being not chosen as signal or noise spectrum figure hardly provide any effective letter to subsequent neural network
Breath.
3. to signal spectrogram and noise spectrum figure progress piecemeal after SNR estimation and compensation, it is contemplated that subsequent migration learns the mind used
Through network model, 299 × 299 pixels are cropped to by each piece；
4. carrying out data enhancing to the gray scale spectrogram after piecemeal, data enhancing technology can alleviate portion generally existing in data set
Divide between the rare and different birds of chirm data and the case where data serious unbalance occurs；And by enriching training dataset,
It can reduce the over-fitting during model training, enhance the generalization ability of model；Increase different from the frequently-used data of normal image
Rival's section, for the time-frequency characteristic of the spectrogram of chirm, the present invention carries out data enhancing using following technology:
(1) frequency-domain transform: pitch and volume including the original chirping of birds sound audio to input are changed at random, and amplitude of fluctuation is not
More than 5%；
(2) noise adds: noise includes noise sample and random Gaussian；Step 2. when, chirm is divided into letter
Number spectrogram and noise spectrum figure, can be randomly selected the sample of noise section, are added in the training sample of signal spectrogram, should
Step can improve classification results and accelerate entire training process；Random Gaussian equally can also help neural network to highlight image
Feature, the step can restore the ambient noise under truth, assist in the characteristic of model learning noise, or even can support
Noise source in anti-reality；
(3) similar sample mixing: under natural environment, it often will appear more birds while singing/piping, to simulate this true feelings
Condition adds the spectrogram of same bird difference audio, and random combine；The step will not influence the distribution of sample label, and
It can be improved the convergence rate of model, increase identification accuracy.
5. increasing the comparison between varying strength region to further enhance the visual perception power of signal section gray level image
Degree carries out Pseudo Col ored Image to it to improve recognition performance, by Jet mapping.Quantify the dynamic range of spectrogram to not same district
Domain, different zones are mapped to three monochrome images of red, green, blue respectively, and it is corresponding it is high, in and low-power spectrum information, red table
Characteristic is sung/piped to the sound property for showing highest energy；The step for another main purpose be that converting gradation spectrogram is
Triple channel RGB image, using the input as subsequent neural network.Obtained triple channel RGB subgraph is finally divided into training set
And test set, the ratio of training set and test set are 4:1；
6. especially normal with the prior art to embody the above are the main contents of the chirm recognition methods based on image procossing
The superiority of image classification method is advised, the present invention did not did multiprocessing to subsequent neural network, and took common in visual task
Transfer learning method；It selects in 2017ILSVR contest and obtains the image recognition structure Squeeze-and- of champion
Excitation Networks (SENeT), it is modeled by the correlation to feature interchannel, strengthens important channel
Feature weakens the feature in insignificant channel, " highlights bird when chirm data it is considered herein that the identical processing band of this thinking is made an uproar
Song inhibits noise " idea；
Therefore, final choice mixed model SE-ResNeXt-50, is finely adjusted it, freezes preceding several layers of weights of neural network,
And combine and need to identify that the kind number of birds redefines full articulamentum, obtain pre-training model.By will be used to test 80%
Triple channel RGB subgraph, which is input in pre-training model, to be trained, and is retained neural network parameter and is obtained identification model, then will remain
The triple channel RGB subgraph of remaining 20% is used for prediction model accuracy rate, obtains recognition result.
The foregoing is only a preferred embodiment of the present invention, limitation in any form not is done to the present invention,
Any person skilled in the art, without departing from the scope of the present invention, using in the technology of the disclosure above
Hold the equivalent embodiment made a little change or be modified to equivalent variations, but anything that does not depart from the technical scheme of the invention content,
Any simple modification, equivalent change and modification to made by above embodiments according to the technical essence of the invention, still fall within
In the range of technical solution of the present invention.
Claims (3)
1. a kind of chirm characteristic strengthening method based on image procossing, it is characterised in that the following steps are included:
1. pre-processing to chirm data set, including resampling and normalization, the spectrogram of chirm is obtained；
2. carrying out SNR estimation and compensation to spectrogram, signal spectrogram and noise spectrum figure are obtained；Wherein, signal spectrogram includes and sings and pipe
Part, noise spectrum figure include noise and mute part, and signal spectrogram is used for ambient noise as original training sample, noise spectrum figure
A kind of approach of enhancing；
3. carrying out piecemeal to all spectrograms after step 2. middle SNR estimation and compensation, and the size of every fritter spectrogram is adjusted to be suitble to
Input dimension for trained neural network；
4. data enhancing is carried out to the signal spectrogram after step 3. middle piecemeal, since spectrogram is different from traditional images, between
Difference limits the direct application of extensive image processing techniques；The characteristic of comprehensive consideration chirm and spectrogram, notebook data increase
Strength reason adhoc approach includes frequency-domain transform, noise addition and the mixing of similar sample:
5. for the visual perception power of enhancing gray level image, while conveniently carrying out transfer learning to different neural networks, pass through Jet
Mapping carries out Pseudo Col ored Image to the enhanced gray level image of data, triple channel RGB color image is obtained, by these triple channels
RGB color image is divided into training set and test set, and wherein training set accounts for 80%, and test set accounts for 20%；
6. suitable neural network model is chosen by transfer learning mode, it is finely adjusted and use step 5. in instruction
Practice collection to be trained, collects verifying model accuracy rate eventually by verifying, obtain recognition result.
2. chirm characteristic strengthening method according to claim 1, it is characterised in that step 2.:
The separation method of signal spectrogram are as follows: one threshold value N of setting, if some pixel value is higher than corresponding line and its phase in spectrogram
N times of the intermediate value that should be arranged then is set to 1, is otherwise set to 0；
The separation method of noise spectrum figure are as follows: one threshold value n (n < N) of setting, if some pixel value is higher than corresponding line and its accordingly
N times of the intermediate value of column is then set to 0, is otherwise set to 1.
3. chirm characteristic strengthening method according to claim 1, it is characterised in that step 4.:
Including frequency-domain transform, noise addition and the mixing of similar sample；
A, frequency-domain transform
1. 2. a, the pitch of original chirping of birds sound audio is changed at random, amplitude of fluctuation is no more than 5%, and then repeatedly step
③；
1. 2. b, the volume of original chirping of birds sound audio is changed at random, amplitude of fluctuation is no more than 5%, and then repeatedly step
③；
B, noise adds
A, by random Gaussian be added to step 2. in signal spectrogram, the image that new standardized of laying equal stress on obtains；
B, by step 2. in noise section be added at random step 2. in signal spectrogram, as training sample；
C, similar sample mixing
Signal spectrogram after the different audio SNR estimation and compensations of same bird is mixed at random.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910139801.9A CN109886214A (en) | 2019-02-26 | 2019-02-26 | A kind of chirm characteristic strengthening method based on image procossing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910139801.9A CN109886214A (en) | 2019-02-26 | 2019-02-26 | A kind of chirm characteristic strengthening method based on image procossing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109886214A true CN109886214A (en) | 2019-06-14 |
Family
ID=66929354
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910139801.9A Withdrawn CN109886214A (en) | 2019-02-26 | 2019-02-26 | A kind of chirm characteristic strengthening method based on image procossing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886214A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127360A (en) * | 2019-12-20 | 2020-05-08 | 东南大学 | Gray level image transfer learning method based on automatic encoder |
| CN111178699A (en) * | 2019-12-15 | 2020-05-19 | 贵州电网有限责任公司 | Method for constructing intelligent checking system of scheduling operation tickets |
| CN111449644A (en) * | 2020-03-19 | 2020-07-28 | 复旦大学 | Bioelectricity signal classification method based on time-frequency transformation and data enhancement technology |
| CN112560673A (en) * | 2020-12-15 | 2021-03-26 | 北京天泽智云科技有限公司 | Thunder detection method and system based on image recognition |
| CN112617789A (en) * | 2020-07-28 | 2021-04-09 | 上海大学 | Laser speckle blood flow imaging method and system |
| WO2021195967A1 (en) * | 2020-03-31 | 2021-10-07 | 深圳市大疆创新科技有限公司 | Image processing method and apparatus, control terminal, and movable platform |
| CN113591561A (en) * | 2021-06-18 | 2021-11-02 | 新绎健康科技有限公司 | Method and system for identity recognition based on spectrogram |
| CN113643194A (en) * | 2021-06-25 | 2021-11-12 | 陕西西宇无损检测有限公司 | Pseudo-color display method for high-gray-scale welding seam negative image |
| CN113889124A (en) * | 2021-11-08 | 2022-01-04 | 国网山东省电力公司乳山市供电公司 | Power transmission line bird prevention platform based on voice recognition |
| CN113900147A (en) * | 2020-07-06 | 2022-01-07 | 中国石油天然气股份有限公司 | Sand dune ringing suppression method and system |
| CN117727332A (en) * | 2024-02-18 | 2024-03-19 | 百鸟数据科技(北京)有限责任公司 | Ecological population assessment method based on language spectrum feature analysis |
| CN117727309A (en) * | 2024-02-18 | 2024-03-19 | 百鸟数据科技(北京)有限责任公司 | Automatic identification method for bird song species based on TDNN structure |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102624652A (en) * | 2011-01-27 | 2012-08-01 | 日电（中国）有限公司 | LDPC decoding method and apparatus, and receiving terminal |
| CN103474072A (en) * | 2013-10-11 | 2013-12-25 | 福州大学 | Rapid anti-noise twitter identification method by utilizing textural features and random forest (RF) |
| CN104658538A (en) * | 2013-11-18 | 2015-05-27 | 中国计量学院 | Mobile bird recognition method based on birdsong |
| US9367887B1 (en) * | 2013-09-05 | 2016-06-14 | Google Inc. | Multi-channel audio video fingerprinting |
| US20170142515A1 (en) * | 2015-09-21 | 2017-05-18 | Northeastern Univerisity | Systems and methods for monitoring and classifying marine animals based on acoustic signals |
| CN108898164A (en) * | 2018-06-11 | 2018-11-27 | 南京理工大学 | A kind of chirping of birds automatic identifying method based on Fusion Features |
-
2019
- 2019-02-26 CN CN201910139801.9A patent/CN109886214A/en not_active Withdrawn
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102624652A (en) * | 2011-01-27 | 2012-08-01 | 日电（中国）有限公司 | LDPC decoding method and apparatus, and receiving terminal |
| US9367887B1 (en) * | 2013-09-05 | 2016-06-14 | Google Inc. | Multi-channel audio video fingerprinting |
| CN103474072A (en) * | 2013-10-11 | 2013-12-25 | 福州大学 | Rapid anti-noise twitter identification method by utilizing textural features and random forest (RF) |
| CN104658538A (en) * | 2013-11-18 | 2015-05-27 | 中国计量学院 | Mobile bird recognition method based on birdsong |
| US20170142515A1 (en) * | 2015-09-21 | 2017-05-18 | Northeastern Univerisity | Systems and methods for monitoring and classifying marine animals based on acoustic signals |
| CN108898164A (en) * | 2018-06-11 | 2018-11-27 | 南京理工大学 | A kind of chirping of birds automatic identifying method based on Fusion Features |
Non-Patent Citations (4)
| Title |
|---|
| GUIFENG93: "数据增强 总结", 《HTTPS://BLOG.CSDN.NET/GUIFENG93/ARTICLE/DETAILS/87625745》 * |
| HONGYI ZHANG 等: "mixup: BEYOND EMPIRICAL RISK MINIMIZATION", 《ICLR 2018》 * |
| MARIO LASSECK: "Audio-based Bird Species Identification with Deep Convolutional Neural Networks", 《HTTPS://WWW.RESEARCHGATE.NET/PUBLICATION/338422871》 * |
| 谢将剑 等: "基于Chirplet语图特征和深度学习的鸟类物种识别方法", 《北京林业大学学报》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178699A (en) * | 2019-12-15 | 2020-05-19 | 贵州电网有限责任公司 | Method for constructing intelligent checking system of scheduling operation tickets |
| CN111127360A (en) * | 2019-12-20 | 2020-05-08 | 东南大学 | Gray level image transfer learning method based on automatic encoder |
| CN111127360B (en) * | 2019-12-20 | 2023-08-29 | 东南大学 | Gray image transfer learning method based on automatic encoder |
| CN111449644A (en) * | 2020-03-19 | 2020-07-28 | 复旦大学 | Bioelectricity signal classification method based on time-frequency transformation and data enhancement technology |
| WO2021195967A1 (en) * | 2020-03-31 | 2021-10-07 | 深圳市大疆创新科技有限公司 | Image processing method and apparatus, control terminal, and movable platform |
| CN113900147B (en) * | 2020-07-06 | 2024-05-28 | 中国石油天然气股份有限公司 | Sand dune ringing suppression method and system |
| CN113900147A (en) * | 2020-07-06 | 2022-01-07 | 中国石油天然气股份有限公司 | Sand dune ringing suppression method and system |
| CN112617789A (en) * | 2020-07-28 | 2021-04-09 | 上海大学 | Laser speckle blood flow imaging method and system |
| CN112560673A (en) * | 2020-12-15 | 2021-03-26 | 北京天泽智云科技有限公司 | Thunder detection method and system based on image recognition |
| CN113591561A (en) * | 2021-06-18 | 2021-11-02 | 新绎健康科技有限公司 | Method and system for identity recognition based on spectrogram |
| CN113643194A (en) * | 2021-06-25 | 2021-11-12 | 陕西西宇无损检测有限公司 | Pseudo-color display method for high-gray-scale welding seam negative image |
| CN113889124A (en) * | 2021-11-08 | 2022-01-04 | 国网山东省电力公司乳山市供电公司 | Power transmission line bird prevention platform based on voice recognition |
| CN117727309A (en) * | 2024-02-18 | 2024-03-19 | 百鸟数据科技(北京)有限责任公司 | Automatic identification method for bird song species based on TDNN structure |
| CN117727309B (en) * | 2024-02-18 | 2024-04-26 | 百鸟数据科技(北京)有限责任公司 | Automatic identification method for bird song species based on TDNN structure |
| CN117727332B (en) * | 2024-02-18 | 2024-04-26 | 百鸟数据科技(北京)有限责任公司 | Ecological population assessment method based on language spectrum feature analysis |
| CN117727332A (en) * | 2024-02-18 | 2024-03-19 | 百鸟数据科技(北京)有限责任公司 | Ecological population assessment method based on language spectrum feature analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886214A (en) | A kind of chirm characteristic strengthening method based on image procossing | |
| US20200159490A1 (en) | Method, system and artificial neural network | |
| CN105488466B (en) | A kind of deep-neural-network and Acoustic Object vocal print feature extracting method | |
| CN105023573B (en) | It is detected using speech syllable/vowel/phone boundary of auditory attention clue | |
| CN110136731A (en) | Empty cause and effect convolution generates the confrontation blind Enhancement Method of network end-to-end bone conduction voice | |
| Zacharakis et al. | An interlanguage unification of musical timbre: Bridging semantic, perceptual, and acoustic dimensions | |
| CN101819638B (en) | Establishment method of pornographic detection model and pornographic detection method | |
| CN107609488A (en) | A kind of ship noise method for identifying and classifying based on depth convolutional network | |
| CN109524020A (en) | A kind of speech enhan-cement processing method | |
| CN109065072A (en) | A kind of speech quality objective assessment method based on deep neural network | |
| CN106469560A (en) | A kind of speech-emotion recognition method being adapted to based on unsupervised domain | |
| CN110148400A (en) | The pronunciation recognition methods of type, the training method of model, device and equipment | |
| NO20015723L (en) | Improving knowledge discovery from multiple datasets using multiple support vector machines | |
| Kadandale et al. | Multi-channel u-net for music source separation | |
| CN110163855B (en) | Color image quality evaluation method based on multi-path deep convolutional neural network | |
| CN113111786A (en) | Underwater target identification method based on small sample training image convolutional network | |
| Reghunath et al. | Transformer-based ensemble method for multiple predominant instruments recognition in polyphonic music | |
| CN114897884A (en) | No-reference screen content image quality evaluation method based on multi-scale edge feature fusion | |
| CN102930865B (en) | Coarse emotion soft cutting and classification method for waveform music | |
| Pons et al. | Gass: Generalizing audio source separation with large-scale data | |
| Liu et al. | Sound Transformation: Applying Image Neural Style Transfer Networks to Audio Spectograms | |
| Perrotin et al. | Evaluating the extrapolation capabilities of neural vocoders to extreme pitch values | |
| CN115472179A (en) | Automatic detection method and system for digital audio deletion and insertion tampering operation | |
| Hamadicharef et al. | Intelligent and perceptual-based approach to musical instruments sound design | |
| CN110909132B (en) | Police service learning content analysis classifying method based on semantic analysis |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190614 |