CN116660876A - Automatic detection and positioning method for echo positioning signals of Chinese white dolphin - Google Patents
Automatic detection and positioning method for echo positioning signals of Chinese white dolphin Download PDFInfo
- Publication number
- CN116660876A CN116660876A CN202310608930.4A CN202310608930A CN116660876A CN 116660876 A CN116660876 A CN 116660876A CN 202310608930 A CN202310608930 A CN 202310608930A CN 116660876 A CN116660876 A CN 116660876A
- Authority
- CN
- China
- Prior art keywords
- echo
- chinese white
- signal
- dolphin
- time
- 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.)
- Pending
Links
- 241001479123 Sousa chinensis Species 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000001514 detection method Methods 0.000 title claims abstract description 17
- 238000010586 diagram Methods 0.000 claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000013528 artificial neural network Methods 0.000 claims abstract description 21
- 238000005070 sampling Methods 0.000 claims abstract description 18
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 11
- 239000012634 fragment Substances 0.000 claims abstract description 8
- 238000009432 framing Methods 0.000 claims abstract description 4
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 241001481833 Coryphaena hippurus Species 0.000 claims description 17
- 238000011176 pooling Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 4
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 238000012216 screening Methods 0.000 abstract description 2
- 210000002569 neuron Anatomy 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000002592 echocardiography Methods 0.000 description 2
- 230000007340 echolocation Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 241000283153 Cetacea Species 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000002431 foraging effect Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/52—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00
- G01S7/52001—Auxiliary means for detecting or identifying sonar signals or the like, e.g. sonar jamming signals
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The invention relates to an automatic detection and positioning method for echo positioning signals of Chinese white dolphin, which comprises the steps of preprocessing acquired data; framing is carried out, the frame is divided into a plurality of data fragments, and each possible click signal position index result is obtained in each data fragment; calculating short-time Fourier transform of sampling points near each echo positioning signal index to form a two-dimensional time-frequency diagram and automatically storing the two-dimensional time-frequency diagram; constructing a convolutional neural network model; training to obtain the best network parameters; the time-frequency diagram is reset to be the designated size and then is input into a trained neural network for further judgment, and relevant information of the obtained echo positioning signals is stored, so that the echo positioning signals of the Chinese white dolphin can be automatically detected and positioned from the acquired data containing ship sonar signals, the echo positioning signals of the Chinese white dolphin and the echo positioning signals of the Chinese white dolphin, the detection efficiency is improved, the problem that time and labor are wasted when a manual means is used for screening massive signals is solved, and the detection efficiency and the detection accuracy are improved.
Description
Technical Field
The invention belongs to the fields of aquatic biology, aquatic science and underwater sound signal processing, and particularly relates to an automatic detection and positioning method for echo positioning signals of Chinese white dolphin.
Background
The echo positioning signal of the Chinese white dolphin is generally used for foraging and navigation in various complex environments, the echo positioning signal emits a sound to the surrounding environment through the head, and receives echoes reflected by various nearby objects, the echoes are used for positioning and identifying the objects, the biological sonar of the Chinese white dolphin is still far superior to the artificial sonar at present, so that the biological sonar and the communication mode of the whale are very worthy of learning, the development of underwater detection positioning and underwater communication technology can play a heuristic role, and the echo positioning signal has important significance for protecting rare wild animals in China, namely the Chinese white dolphin. However, the echo positioning signals of the white dolphin are collected from the ocean of great vast and quick extent and are still very difficult at present, and the main characteristic is that the echo positioning signals of the white dolphin usually appear in the form of sequences, each sequence contains a large number of single echo positioning signals, the duration of each echo positioning signal is very short, usually only tens of microseconds, the traditional method for identifying by manpower consumes time and labor, the requirement of people for processing a large number of white dolphin voice signals cannot be met, and the identification accuracy of some detection methods based on time domains is still deficient at present. Therefore, if the collected signals can be analyzed in an automatic mode, and the echo locating signals of the white dolphin can be automatically detected and extracted from the signals, the research efficiency is greatly improved.
Disclosure of Invention
The invention aims to provide the automatic detection and positioning method for the echo positioning signals of the Chinese white dolphins, which can automatically detect and position the echo positioning signals of the Chinese white dolphins from the acquired data containing ship sonar signals, echo positioning signals of the Chinese white dolphins and the voice signals of the Chinese white dolphins, thereby improving the detection efficiency, overcoming the time and labor waste problem when manual means are used for screening mass signals and improving the detection efficiency and accuracy.
The present invention has been achieved in such a way that,
an automatic detection and positioning method for echo positioning signals of Chinese white dolphin comprises the following steps:
step 1: preprocessing the acquired data containing the echo positioning signals of the Chinese white dolphin;
step 2: framing the data containing the echo positioning signals of the dolphin sinensis obtained in the step 1, dividing the data into a plurality of data fragments, and obtaining a position index result of each possible click signal in each data fragment;
step 3: according to the possible echo positioning signal position index result obtained by calculation in the step 2, calculating short-time Fourier transform of sampling points near each echo positioning signal index to form a two-dimensional time-frequency diagram and automatically storing the two-dimensional time-frequency diagram;
step 4: constructing a convolutional neural network model;
step 5: manufacturing a training set and a verification set for classifying and identifying echo positioning signals, performing data enhancement processing on training set data, then training the neural network model built in the step 4 according to specified parameters, and storing the best network parameters;
step 6: and (3) resetting the time-frequency diagram stored in the step (3) to a specified size, inputting the time-frequency diagram into a trained neural network for further judgment, and storing the related information of the obtained echo positioning signal.
Further, obtaining each possible click signal location index result within each data segment includes:
step A: setting a designated sample length SampleNum, sampleNum to represent the number of sampling points contained in a sample, calculating and storing signal kurtosis value information in the sample;
and (B) step (B): sequentially moving the initial position of the sample according to a designated step length until the end of the data segment, and calculating kurtosis values of all sample data;
step C: judging whether the data segment contains the Chinese white dolphin echo positioning signal according to the set kurtosis threshold value, and obtaining an echo positioning signal initial position index.
Further, the step a of performing feature calculation on the sampling points in the sample to obtain kurtosis value information includes:
kurtosis calculation is carried out on all sampling points in a sample:
wherein K represents kurtosis value in the sample, which is used for measuring flatness of data distribution, namely data value distribution form steepness degree,represents the fourth order central moment of the sample, +.>Representing the variance of the sample.
Further, the step C of judging whether the data segment contains the echo locating signal of white dolphin according to the set kurtosis threshold value, and obtaining the initial position index thereof includes:
when the maximum kurtosis value calculated in a certain sample in the data segment is larger than a set threshold value, initially judging that the echo locating signal of the Chinese white dolphin exists in the data segment, recording a position index at the maximum kurtosis value as a starting index of the echo locating signal, and storing the position index into a list to wait for the next processing.
Further, the constructing of the neural network model in the step 4 includes two parts of feature extraction and network classification, wherein the feature extraction part includes 9 CBL modules, 5 maximum pooling units, one average pooling unit, and the classification part includes 2 LBL modules and 1 softmax unit;
the CBL module is composed of Conv layer, BN layer and ReLu activation function, and the LBL module is composed of Linear layer, BN layer and ReLu activation function.
Further, the training set data enhancement operation in step 5 includes: the step 5 of training the neural network model constructed in the step 4 according to the specified parameters comprises the following steps: with cross entropy as a loss function, SGD as a network optimizer, batch size 12, learning momentum 0.9, initial learning rate set to 0.001, and then learning rate changed to one tenth of the original one every 10 rounds of training.
Further, in step 6, the time-frequency diagram saved in step 3 is reset to a specified size and then input into the trained neural network for further judgment, and the saved result includes:
loading the network parameters with the best recognition effect obtained in the step 5 into a network model, then adopting a bicubic interpolation method to adjust the two-dimensional time-frequency diagram size of the echo positioning signal to be determined to enable the two-dimensional time-frequency diagram size to meet the network input requirement, inputting the two-dimensional time-frequency diagram size into a neural network, and calculating the two-dimensional time-frequency diagram size by the neural network to generate two outputs which respectively represent the probability that the echo positioning signal is the echo positioning signal of the Chinese white dolphin and the probability that the echo positioning signal is the sonar noise signal of the ship;
if the probability of the Chinese white dolphin echo positioning signal is larger than the other probability in the output of the neural network, finally determining that the Chinese white dolphin echo positioning signal is the Chinese white dolphin echo positioning signal, and then calculating the occurrence time of the Chinese white dolphin echo positioning signal; repeating the steps until all the time-frequency diagrams are judged, and writing the position information, the time information and the probability information of the echo positioning signals of the Chinese white dolphin which are judged to be finished into a csv file for recording and storing.
Compared with the prior art, the invention has the beneficial effects that:
according to the invention, by utilizing the characteristic that the echo positioning signal of the Chinese white dolphin has obvious high-frequency narrow pulse characteristics compared with other signals, the interference of ocean background noise is removed in a proper mode, so that the characteristic of the echo positioning signal is highlighted, the comprehensive judgment of a convolutional neural network is combined after the characteristic signal is extracted in the time domain, the detection accuracy is improved, and the full-automatic detection and positioning of the echo positioning signal of the Chinese white dolphin are realized. According to the invention, the echo positioning signal of the white dolphin can be automatically and accurately extracted and positioned from the data mixed with ship sonar noise acquired in dolphin research, so that the dependence on manpower is reduced, and the recognition efficiency is improved.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is an example of a time-frequency diagram of a resized signal provided in accordance with an embodiment of the present invention;
FIG. 3 is a neural network architecture for use with embodiments of the present invention;
FIG. 4 is a graph of neural network loss provided by an embodiment of the present invention;
FIG. 5 is a graph of accuracy of a neural network on a validation set provided by an embodiment of the present invention;
FIG. 6 is a time-frequency diagram of a segment of actual data provided by an embodiment of the present invention;
fig. 7 is a schematic diagram of a signal result extracted by using the method according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Because the actually collected observation signals do not contain the Chinese white dolphin echo positioning signals at each moment, a proper echo positioning signal detection method is needed, the echo positioning signal positions are detected and positioned from the collected signals, a time-frequency diagram is generated through the detected signal positions, and a convolutional neural network is called to judge the echo positioning signals and retain the final echo positioning signal results. Referring to fig. 1, the method for automatically detecting and positioning the echo positioning signal of the dolphin sinensis comprises the following steps:
step 1: and preprocessing the echo positioning signal of the Chinese white dolphin.
Aiming at the frequency domain characteristics of the echo positioning signals of the Chinese white dolphins, a proper frequency band is selected to carry out high-pass filtering treatment on the acquired signals, clutter which is obviously lower than the frequency band of the echo positioning signals, whistle sounds and ocean background noise which are sent by the Chinese white dolphins are filtered, and the signal to noise ratio is improved. The high-pass filter adopts a direct FIR high-pass filter designed by a Kaiser window, the stop band frequency of the high-pass filter is 12000 Hz, the pass band frequency of the high-pass filter is 13000 Hz, and the order is 2892.
Step 2: and (3) framing the data containing the Chinese white dolphin echo positioning signals obtained in the step (1), dividing the data into a plurality of data fragments, and obtaining each possible click signal position index result in each data fragment to automatically extract the Chinese white dolphin echo positioning signal time domain.
For the Chinese white dolphin echo locating signals, the duration range is 20-80 mu s, the interval range between the echo locating signals is 30-60ms, the denoised signals are divided into a plurality of time segments according to the minimum interval of the echo locating signals, each segment lasts for 28ms (so as to ensure that each time segment only contains one echo locating signal), namely each time segment contains 16000 sampling points (the sampling rate of the acquired signals is 576000). Then one sample segment at a time is taken during the time segment: s (n), n=1, 2,3,..2500), and calculating kurtosis information for sample points within the sample segment:
wherein K represents the kurtosis value in the sample, which is used for measuring the flatness (flat) of the data distribution, namely the sharpness of the data value distribution form.Represents the fourth order central moment of the sample, +.>Representing the variance of the sample. Then taking the step length as 100 points, sequentially calculating all samples S in the time segment i (n), i=1, 2,3,..kurtosis information, and thus kurtosis values for all samples within the time segment are obtained and saved. And setting the kurtosis threshold value to 90, and if the maximum kurtosis of the sample exceeds the set threshold value in a certain time segment, preliminarily judging that an echo positioning signal exists in the segment, wherein the position index of the echo positioning signal is the index at the maximum kurtosis. Taking into account the presence of echolocation signalsThe situation occurs in the middle of two time slices, when the following is done:
if the difference between the last index of the segment and the calculated position index of the echo positioning signal is less than 500 sampling points, the echo positioning signal is considered to be between the time segment and the next time segment, and at this time, the first 800 sampling points of the time segment are skipped before the kurtosis information of the samples in the next time segment is calculated, so that the echo positioning signal is prevented from being repeatedly calculated. I.e. dividing the echolocation signal in between two segments into the previous time segment.
And repeating the operation for each time segment to obtain and store all the position indexes which are preliminarily judged to be the echo positioning signals in the data.
Step 3: and (5) automatically storing the echo positioning signal time-frequency diagram.
According to all the undetermined echo positioning signal position indexes stored in the step 2, for each signal index, taking the first 500 sampling points and the last 4500 sampling points of the index position as a range interval of a generated time-frequency diagram, applying a hanning window with the length of 256 to the sampling points in the region, a 50% window overlapping rate and 256 Fourier transform points, calculating short-time Fourier transform of the sampling points in the range, generating a signal time-frequency diagram with the size of 3 x 31 x 224, and storing the signal time-frequency diagram. Repeating the operation to obtain the time-frequency diagrams at all the signal index positions.
Step 4: and constructing a convolutional neural network model.
The convolutional neural network is constructed based on a pytorch framework, the overall network structure and the output size of each stage of feature map are shown as figure 3, the basic constituent units of the convolutional neural network of the method are a CBL module and an LBL module, wherein the CBL module is formed by a Conv layer, a BatchNorm layer and a ReLu activation function, and the LBL module is formed by a Linear layer, a BatchNorm layer and a ReLu activation function. The BatchNorm layer is added, so that the setting of the super parameters of the network is more free, the convergence rate of the network is faster, and the performance is better. The principle formula is expressed as follows, and the purpose is to enable a Batch (Batch) feature map to meet the distribution rule with the mean value of 0 and the variance of 1.
Where E (x) is the mean μ and Var (x) is the variance σ 2 Obtained by statistics in forward propagation calculations. Gamma and beta are trained in the back propagation process.
The neural network constructed by the method mainly comprises two parts, wherein the first part is a feature extraction part, and features of training data are extracted through combination of a CBL unit and a Max pool unit (the sizes of pooling cores of pooling units used by the method are all 2 x 2, and the step sizes are all 2 pixels).
Specifically, the network inputs are RGB images of 3×224×224, and the 1 st CBL module performs the following settings: input 3 channels, output 32 channels, convolution kernel size 3*3, step size 1 pixel, fill 1 pixel. At this time, the size of the feature map becomes 32×224×224, and the 2 nd CBL module is set as follows: the input 32 channels, the output 64 channels, and the rest are set to be the same as the first CBL module. At this time, the size of the feature map becomes 64×224×224, and the 3 rd CBL module is set as follows: the input 64 channels and the output 128 channels are arranged in the same way as the first CBL module. At this time, the size of the feature map becomes 128×224×224. And then carrying out dimension reduction compression on the feature map through a maximum pooling unit to extract texture features. At this time, the feature map size becomes 128×112×112.
The fourth CBL module is set as follows: the input 128 channels and the output 128 channels are the same as the first CBL module, and at this time, the feature map size becomes 128×112×112, and the fifth CBL module performs the following settings: the input 128 channels and the output 256 channels are arranged in the same way as the first CBL module. At this time, the feature map size becomes 256×112×112. The feature map is further reduced in dimension by a maximum pooling unit, and the feature map becomes 256×56×56.
The sixth CBL module is set as follows: 256 channels are input, 512 channels are output, the step size is 1 pixel, and no filling exists. At this time, the feature map size becomes 512×54×54, and the seventh CBL module performs the following settings: the input 512 channels and the output 512 channels are arranged in the same way as the sixth CBL module. At this time, the feature map size becomes 512×52×52. And then continuing to reduce the dimension through the maximum pooling unit, wherein the size of the feature map is 512 x 26.
The eighth CBL module is set as follows: the input 512 channels and the output 512 channels are arranged in the same way as the sixth CBL module. At this time, the feature map size becomes 512×24×24. And then the dimension reduction is carried out through the maximum pooling unit, and the size of the characteristic diagram is changed to 512 x 12. The last CBL module is set as follows: the input 512 channels, the output 512 channels, and the rest are set to be the same as the first CBL module. At this time, the feature map size becomes 512×12×12. And then reducing the dimension by using a maximum pooling unit, wherein the size of the characteristic diagram is 512 x 6. After the last maximum pooling, the feature map is processed using an averaging pooling unit for further compression of the feature map and more likely retention of the feature information. At this time, the feature map size becomes 512×3×3. Thus, the feature extraction of the data is completed.
The second part is a classification part which mainly consists of two LBL units and a Soft Max layer and mainly aims to flatten the two-dimensional feature map obtained by the feature extraction part into a one-dimensional vector so as to facilitate the final classification.
Specifically, the feature map with a size of 512×3×3 obtained by the feature extraction section is flattened to obtain a one-dimensional vector with a length of 4609. After passing through the first CBL module, the vector length becomes 256. And then Drop-out is added, which has the effect of randomly setting the output of a certain proportion of neurons to 0 in the training process, namely discarding the output of some neurons, thereby reducing the complexity of the model, alleviating the problem of over-fitting and improving the generalization capability of the model. At the time of testing, all neurons will be activated and therefore will not have any effect on the network. The vector length then continues to be 2 by the second LBL module. The last Soft Max layer is used to normalize the resulting one-dimensional numerical vector of length 2 into a probability distribution vector, the first value in the probability vector representing the probability that it is an echo locating signal and the second value representing the probability that it is a noise signal (ship sonar noise). Thus, the construction of the convolutional neural network is completed.
Step 5: training the convolutional neural network model built in the step 4, and storing the model parameters with the best effect.
And (3) extracting a plurality of Chinese white dolphin echo positioning signals and noise signals (ship sonar noise) from a section of actually acquired data by using the methods introduced in the steps 1,2 and 3. From which echo locating signals and noise signals are randomly chosen as the original data set.
After the selection is completed, the original data set contains 1204 pictures, and 604 positive (echo positioning signal) negative (ship sonar noise signal) samples are respectively contained. With 80% of the data as training set and 20% as validation set. Before converting the original data set picture into tensor and inputting the tensor to the convolutional neural network, performing data enhancement operation on the data set, including random center clipping and random horizontal overturn to improve the sample quality and the generalization capability and robustness of the model.
During training, cross entropy is used as a loss function, SGD is used as a network optimizer, batch size is 12, initial learning rate is set to be 0.001, then, after 10 training rounds, learning rate becomes one tenth of the original training, 50 training rounds are performed on the network, the loss value of the neural network continuously decreases until convergence along with the training, and a loss curve of the network after training is shown in fig. 4. The recognition accuracy on the training set is also improved, and the recognition accuracy gradually reaches 100%. Finally, the recognition accuracy on the verification set also reaches 100%, and the accuracy curve of the neural network on the verification set is shown in fig. 5. And finally, saving the network parameters with the best recognition results, and loading the network parameters into a network model.
Step 6: and (3) resetting the time-frequency diagram stored in the step (3) to a specified size, inputting the time-frequency diagram into a trained neural network for further judgment, and storing the related information of the obtained echo positioning signal.
The collected data contains ship sonar noise, the frequency band range and the signal characteristics of the data are close to those of the white dolphin sonar signal, most sonar noise is eliminated according to the set kurtosis threshold value in the step 2, but a small part of sonar noise still exists, so that the stored echo positioning signal is further judged by using a neural network.
And (5) calling the network model obtained in the step (5) to identify all the signal time-frequency diagrams obtained in the step (2). In order to meet the input requirement of the neural network, all time-frequency diagrams are adjusted to be 3 x 224 by adopting a bicubic interpolation method, the time-frequency diagrams of the signals after the adjustment are shown in fig. 2, the adjusted pictures are sent to a network model for judgment, two numerical values are output for each picture, the neural network respectively represents the probability that the picture is an echo positioning signal and a ship sonar noise signal, if the probability of the echo positioning signal is greater than the probability of the noise signal, the echo positioning signal is finally considered to be the echo positioning signal, at the moment, the position index i of the signal is recorded, the occurrence time t, t=i/fs, fs is the signal sampling rate, and the probability p of the echo positioning signal is obtained. Repeating the above operation, judging all the signal time-frequency diagrams obtained in the step 2, and finally writing the relevant information of all the echo positioning signals obtained by the complaint operation into a csv file for recording and storing.
The method is actually tested to detect the condition of the echo positioning signal of the white dolphin. The method comprises the steps of using a segment of acquired actual data, applying a Hanning window with the length of 256, calculating short-time Fourier transform with the overlapping rate of 50% window, making a time-frequency chart of the short-time Fourier transform as shown in figure 6, firstly, manually obtaining that the segment of signal contains 16 Chinese white dolphin echo positioning signals, then successfully extracting 16 echo positioning signals by using the method introduced by the invention, matching the calculated position with the actual position, and obtaining the probability that the calculated signal position corresponds to the calculated position as shown in figure 7. The accuracy of extracting the echo locating signal of the dolphin sinensis on the data segment reaches 100%.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (7)
1. The automatic detection and positioning method for the echo positioning signal of the Chinese white dolphin is characterized by comprising the following steps of:
step 1: preprocessing the acquired data containing the echo positioning signals of the Chinese white dolphin;
step 2: framing the data containing the echo positioning signals of the dolphin sinensis obtained in the step 1, dividing the data into a plurality of data fragments, and obtaining a position index result of each possible click signal in each data fragment;
step 3: according to the possible echo positioning signal position index result obtained by calculation in the step 2, calculating short-time Fourier transform of sampling points near each echo positioning signal index to form a two-dimensional time-frequency diagram and automatically storing the two-dimensional time-frequency diagram;
step 4: constructing a convolutional neural network model;
step 5: manufacturing a training set and a verification set for classifying and identifying echo positioning signals, performing data enhancement processing on training set data, then training the neural network model built in the step 4 according to specified parameters, and storing the best network parameters;
step 6: and (3) resetting the time-frequency diagram stored in the step (3) to a specified size, inputting the time-frequency diagram into a trained neural network for further judgment, and storing the related information of the obtained echo positioning signal.
2. The method for automatically detecting and locating echo locating signals of white dolphin according to claim 1, wherein obtaining each possible click signal location index result in each data segment comprises:
step A: setting a designated sample length SampleNum, sampleNum to represent the number of sampling points contained in a sample, calculating and storing signal kurtosis value information in the sample;
and (B) step (B): sequentially moving the initial position of the sample according to a designated step length until the end of the data segment, and calculating kurtosis values of all sample data;
step C: judging whether the data segment contains the Chinese white dolphin echo positioning signal according to the set kurtosis threshold value, and obtaining an echo positioning signal initial position index.
3. The method for automatically detecting and locating the echo locating signal of the Chinese white dolphin according to claim 2, wherein,
and (C) performing characteristic calculation on the sampling points in the sample to obtain kurtosis value information, wherein the step (A) comprises the following steps of:
kurtosis calculation is carried out on all sampling points in a sample:
wherein K represents kurtosis value in the sample, which is used for measuring flatness of data distribution, namely data value distribution form steepness degree,represents the fourth order central moment of the sample, +.>Representing the variance of the sample.
4. The method for automatically detecting and locating the echo locating signal of the Chinese white dolphin according to claim 2, wherein,
and C, judging whether the data segment contains the Chinese white dolphin echo positioning signal according to the set kurtosis threshold value, and obtaining the initial position index comprises the following steps:
when the maximum kurtosis value calculated in a certain sample in the data segment is larger than a set threshold value, initially judging that the echo locating signal of the Chinese white dolphin exists in the data segment, recording a position index at the maximum kurtosis value as a starting index of the echo locating signal, and storing the position index into a list to wait for the next processing.
5. The method for automatically detecting and locating echo locating signals of dolphin sinensis according to claim 2, wherein the constructing of the neural network model in the step 4 comprises two parts of feature extraction and network classification, wherein the feature extraction part comprises 9 CBL modules, 5 maximum pooling units, one average pooling unit, and the classification part comprises 2 LBL modules and 1 softmax unit;
the CBL module is composed of Conv layer, BN layer and ReLu activation function, and the LBL module is composed of Linear layer, BN layer and ReLu activation function.
6. The method for automatically detecting and locating echo locating signals of white dolphin according to claim 2, wherein the training set data enhancement operation in step 5 comprises: the step 5 of training the neural network model constructed in the step 4 according to the specified parameters comprises the following steps: with cross entropy as a loss function, SGD as a network optimizer, batch size 12, learning momentum 0.9, initial learning rate set to 0.001, and then learning rate changed to one tenth of the original one every 10 rounds of training.
7. The method for automatically detecting and locating the echo locating signal of the dolphin Chinese according to claim 2, wherein in step 6, the time-frequency diagram stored in step 3 is reset to a specified size and then is input into a trained neural network for further judgment, and the storing of the obtained result comprises:
loading the network parameters with the best recognition effect obtained in the step 5 into a network model, then adopting a bicubic interpolation method to adjust the two-dimensional time-frequency diagram size of the echo positioning signal to be determined to enable the two-dimensional time-frequency diagram size to meet the network input requirement, inputting the two-dimensional time-frequency diagram size into a neural network, and calculating the two-dimensional time-frequency diagram size by the neural network to generate two outputs which respectively represent the probability that the echo positioning signal is the echo positioning signal of the Chinese white dolphin and the probability that the echo positioning signal is the sonar noise signal of the ship;
if the probability of the Chinese white dolphin echo positioning signal is larger than the other probability in the output of the neural network, finally determining that the Chinese white dolphin echo positioning signal is the Chinese white dolphin echo positioning signal, and then calculating the occurrence time of the Chinese white dolphin echo positioning signal; repeating the steps until all the time-frequency diagrams are judged, and writing the position information, the time information and the probability information of the echo positioning signals of the Chinese white dolphin which are judged to be finished into a csv file for recording and storing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310608930.4A CN116660876A (en) | 2023-05-26 | 2023-05-26 | Automatic detection and positioning method for echo positioning signals of Chinese white dolphin |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310608930.4A CN116660876A (en) | 2023-05-26 | 2023-05-26 | Automatic detection and positioning method for echo positioning signals of Chinese white dolphin |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116660876A true CN116660876A (en) | 2023-08-29 |
Family
ID=87711162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310608930.4A Pending CN116660876A (en) | 2023-05-26 | 2023-05-26 | Automatic detection and positioning method for echo positioning signals of Chinese white dolphin |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116660876A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110273964A1 (en) * | 2010-05-10 | 2011-11-10 | Wildlife Acoustics, Inc. | Method for listening to ultrasonic animal sounds |
| CN108680245A (en) * | 2018-04-27 | 2018-10-19 | 天津大学 | Whale globefish class Click classes are called and traditional Sonar Signal sorting technique and device |
| CN110927706A (en) * | 2019-12-10 | 2020-03-27 | 电子科技大学 | Convolutional neural network-based radar interference detection and identification method |
| CN111444832A (en) * | 2020-03-25 | 2020-07-24 | 哈尔滨工程大学 | Whale cry classification method based on convolutional neural network |
| CN112712047A (en) * | 2021-01-08 | 2021-04-27 | 自然资源部第一海洋研究所 | Marine mammal echo positioning signal detection method based on image processing |
| CN113870870A (en) * | 2021-12-02 | 2021-12-31 | 自然资源部第一海洋研究所 | Convolutional neural network-based real-time recognition method for marine mammal vocalization |
| WO2022052367A1 (en) * | 2020-09-10 | 2022-03-17 | 中国科学院深圳先进技术研究院 | Neural network optimization method for remote sensing image classification, and terminal and storage medium |
| CN114488100A (en) * | 2022-01-20 | 2022-05-13 | 哈尔滨工程大学 | Whale echo positioning monopulse signal extraction method |
| CN114881093A (en) * | 2022-07-05 | 2022-08-09 | 北京理工大学 | Signal classification and identification method |
| CN115050386A (en) * | 2022-05-17 | 2022-09-13 | 哈尔滨工程大学 | Automatic detection and extraction method for Chinese white dolphin whistle sound signal |
-
2023
- 2023-05-26 CN CN202310608930.4A patent/CN116660876A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110273964A1 (en) * | 2010-05-10 | 2011-11-10 | Wildlife Acoustics, Inc. | Method for listening to ultrasonic animal sounds |
| CN108680245A (en) * | 2018-04-27 | 2018-10-19 | 天津大学 | Whale globefish class Click classes are called and traditional Sonar Signal sorting technique and device |
| CN110927706A (en) * | 2019-12-10 | 2020-03-27 | 电子科技大学 | Convolutional neural network-based radar interference detection and identification method |
| CN111444832A (en) * | 2020-03-25 | 2020-07-24 | 哈尔滨工程大学 | Whale cry classification method based on convolutional neural network |
| WO2022052367A1 (en) * | 2020-09-10 | 2022-03-17 | 中国科学院深圳先进技术研究院 | Neural network optimization method for remote sensing image classification, and terminal and storage medium |
| CN112712047A (en) * | 2021-01-08 | 2021-04-27 | 自然资源部第一海洋研究所 | Marine mammal echo positioning signal detection method based on image processing |
| CN113870870A (en) * | 2021-12-02 | 2021-12-31 | 自然资源部第一海洋研究所 | Convolutional neural network-based real-time recognition method for marine mammal vocalization |
| CN114488100A (en) * | 2022-01-20 | 2022-05-13 | 哈尔滨工程大学 | Whale echo positioning monopulse signal extraction method |
| CN115050386A (en) * | 2022-05-17 | 2022-09-13 | 哈尔滨工程大学 | Automatic detection and extraction method for Chinese white dolphin whistle sound signal |
| CN114881093A (en) * | 2022-07-05 | 2022-08-09 | 北京理工大学 | Signal classification and identification method |
Non-Patent Citations (4)
| Title |
|---|
| 孙蕴瀚;史金龙;孙正兴;: "利用自监督卷积网络估计单图像深度信息", 计算机辅助设计与图形学学报, no. 04, 31 January 2020 (2020-01-31) * |
| 杨少凡;郭中源;贾宁;郭圣明;肖东;黄建纯;陈庚;: "海豚Whistles为信息载体的正交频分复用循环移位键控扩频伪装水声通信", 声学学报, no. 05, 13 September 2018 (2018-09-13) * |
| 杨蔚: "鲸豚动物叫声检测识别技术研究", 中国优秀硕士论文电子期刊网, 31 December 2022 (2022-12-31), pages 2 - 28 * |
| 高敏;尹雪飞;陈克安;: "时频图像特征用于声场景分类", 声学技术, no. 05, 15 October 2017 (2017-10-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245608B (en) | Underwater target identification method based on half tensor product neural network | |
| CN110927706B (en) | Convolutional neural network-based radar interference detection and identification method | |
| CN110020684B (en) | Image denoising method based on residual convolution self-coding network | |
| CN110675411B (en) | Cervical squamous intraepithelial lesion recognition algorithm based on deep learning | |
| CN113191178B (en) | Underwater sound target identification method based on auditory perception feature deep learning | |
| CN112819732A (en) | B-scan image denoising method for ground penetrating radar | |
| CN116052064B (en) | Method and device for identifying feeding strength of fish shoal, electronic equipment and bait casting machine | |
| CN115830436A (en) | Marine organism intelligent detection method based on deep learning | |
| CN115546622A (en) | Fish shoal detection method and system, electronic device and storage medium | |
| CN116778309A (en) | Residual bait monitoring method, device, system and storage medium | |
| CN113052866B (en) | Ultrasonic image tongue contour extraction method based on local binary fitting model | |
| CN110349106A (en) | A kind of wavelet soft-threshold image de-noising method based on Renyi entropy | |
| CN112381845B (en) | Rock core image generation method, model training method and device | |
| CN112270404A (en) | Detection structure and method for bulge defect of fastener product based on ResNet64 network | |
| CN116660876A (en) | Automatic detection and positioning method for echo positioning signals of Chinese white dolphin | |
| CN115601301B (en) | Fish phenotype characteristic measurement method, system, electronic equipment and storage medium | |
| CN116884435A (en) | Voice event detection method and device based on audio prompt learning | |
| CN117197454A (en) | Liver and liver tumor data segmentation method and system | |
| CN115690546A (en) | Shrimp length measuring method and device, electronic equipment and storage medium | |
| CN115223033A (en) | Synthetic aperture sonar image target classification method and system | |
| CN116756527B (en) | Data processing method for breeding screening | |
| CN112906458B (en) | Group intelligent optimized underwater laser multi-target end-to-end automatic identification system | |
| CN115169407B (en) | Weak transient signal detection method and system based on time domain depth feature learning | |
| CN109064477A (en) | With the method for improved U-Net detection cell nuclear periphery | |
| CN117974736B (en) | Underwater sensor output signal noise reduction method and system based on machine learning |
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 |