Bat type ultrasonic identification method based on spatial orientation, medium and system

Abstract

The invention provides an ultrasonic identification method, medium and system for bat types based on spatial orientation, and belongs to the technical field of bat type identification, and the method comprises the steps: obtaining a first collection signal, a second collection signal and a third collection signal collected by three ultrasonic signal collectors; preprocessing the three ultrasonic signals to obtain a first ultrasonic signal, a second ultrasonic signal and a third ultrasonic signal; inputting the relative coordinates of the three ultrasonic signals and the three ultrasonic signals by using a pre-fitted Doppler compensation equation set to obtain three ultrasonic signals with Doppler effect removed, and respectively recording the three ultrasonic signals as a first signal, a second signal and a third signal; respectively inputting the first signal, the second signal and the third signal into a pre-trained bat type identification model to obtain a corresponding bat type and confidence thereof; and the bat type with the highest confidence coefficient is selected as a bat type recognition result, and the influence of the Doppler effect is eliminated, so that the recognition accuracy of the bat types is high.

Description

Technical Field

[0001] The present invention belongs to the technical field of bat species identification. Specifically, it relates to a method, medium and system for ultrasonic identification of bat species based on spatial orientation. Background Art

[0002] In recent years, through the analysis of bat ultrasonic signals, the identification of some bat species has been preliminarily achieved. Existing research mainly focuses on extracting the time-domain and frequency-domain features of ultrasonic signals and applying machine learning models for species classification. For example, some scholars have proposed using the short-time Fourier transform to extract the time-frequency features of bat ultrasonic signals and then using the Gaussian mixture model for species identification. There are also some studies that use the wavelet transform and support vector machine methods. These methods have achieved certain identification effects in some controlled environments, but the existing identification methods are difficult to adapt to complex field environments. In practical applications, bats may be in different flight states, and the ultrasonic signals they emit will be affected by the Doppler effect. This Doppler frequency shift will seriously interfere with feature extraction and classification, resulting in a decrease in the identification accuracy. In addition, the field environment has a large amount of noise, which will also have an adverse impact on signal features. Summary of the Invention

[0003] In view of this, the present invention provides a method, medium and system for ultrasonic identification of bat species based on spatial orientation, which can solve the technical problem that the existing method for identifying bat species according to ultrasonic waves has a decrease in identification accuracy due to the Doppler effect on the ultrasonic signals emitted by bats in different flight states.

[0004] The present invention is implemented as follows:

[0005] The first aspect of the present invention provides a method for ultrasonic identification of bat species based on spatial orientation, including the following steps:

[0006] S10. Obtain the first acquisition signal, the second acquisition signal and the third acquisition signal collected by three ultrasonic signal collectors;

[0007] S20. Preprocess the three ultrasonic signals to obtain the first ultrasonic signal, the second ultrasonic signal and the third ultrasonic signal;

[0008] S30. Use the pre-fitted Doppler compensation equation set, input the relative coordinates of the three ultrasonic signals and the three ultrasonic signals, and obtain three ultrasonic signals without the Doppler effect, which are respectively denoted as the first signal, the second signal and the third signal;

[0009] S40. Input the first signal, the second signal and the third signal into the pre-trained bat species identification model to obtain the corresponding bat species and their confidence levels;

[0010] S50. Select the bat species with the highest confidence as the result of the identified bat species.

[0011] Specifically, the specific steps of step S10 are as follows: First, three ultrasonic signal collectors are used to obtain ultrasonic signals at three different azimuth angles respectively. The arrangement positions of the three collectors have certain spatial angular differences relative to the horizontal flight path of the bat. These three collected signals are respectively recorded as the first collected signal, the second collected signal, and the third collected signal. The relative position coordinate information of the collectors needs to be measured and recorded in advance.

[0012] The specific steps of step S20 are as follows: Preprocess the three collected signals obtained. The preprocessing includes, but is not limited to, operations such as filtering, normalization, and time-frequency analysis. The purpose of filtering is to eliminate noise interference and improve the signal quality; normalization is to eliminate the influence of amplitude differences; time-frequency analysis such as short-time Fourier transform can obtain the time-domain and frequency-domain characteristics of the signal. After preprocessing, three corrected and enhanced ultrasonic signals are obtained, which are respectively recorded as the first ultrasonic signal, the second ultrasonic signal, and the third ultrasonic signal.

[0013] The specific steps of step S30 are as follows: Use the pre-fitted Doppler compensation equation set, input the relative coordinates of the three ultrasonic signal collectors and the three collected ultrasonic signals, and obtain three ultrasonic signals without the Doppler effect. The Doppler compensation equation set includes a velocity estimation equation, a direction estimation equation, a signal reconstruction equation, a frequency offset equation, and a phase compensation equation. These equation sets can estimate the flight speed and direction of the bat according to the frequency differences and phase differences of the three received signals, and reconstruct the original transmitted signal. After these compensation steps, the three ultrasonic signals without the Doppler effect obtained are respectively recorded as the first signal, the second signal, and the third signal.

[0014] The specific steps of step S40 are as follows: Input the aforementioned first signal, second signal, and third signal into the pre-trained bat species recognition model respectively. The recognition model includes four parts: a feature extraction module, a feature classification module, a feature fusion module, and a deep neural network. The feature extraction module extracts time-domain and frequency-domain characteristics from the input signal based on wavelet transform and Fourier transform to form a feature vector; the feature classification module uses a support vector machine (SVM) to preliminarily classify the extracted features and give a preliminary classification result; the feature fusion module uses an attention mechanism to fuse the features and classification results of the three signals to obtain the fused features; finally, the deep neural network model performs the final species recognition and outputs the bat species and its confidence.

[0015] Based on the above technical solutions, a method for ultrasonic recognition of bat species based on spatial orientation according to the present invention can also be improved as follows:

[0016] Among them, the Doppler compensation equation set is used to obtain the flight speed and direction of the bat and the original transmitted signal according to the relative positions of the three ultrasonic signal collectors and the received signals; it includes a speed estimation equation, a direction estimation equation, a signal reconstruction equation, a frequency offset equation, and a phase compensation equation.

[0017] Further, the speed estimation equation is used to estimate the flight speed of the bat according to the frequency differences of the three received signals;

[0018] The direction estimation equation is used to estimate the flight direction of the bat according to the phase differences of the three received signals;

[0019] The signal reconstruction equation is used to reconstruct the original transmitted signal according to the estimated speed and direction;

[0020] The frequency offset equation is used to calculate the frequency offset caused by the Doppler effect;

[0021] The phase compensation equation is used to calculate the phase change caused by the Doppler effect.

[0022] Further, the bat species recognition model includes a feature extraction module, a feature classification module, a feature fusion module, and a deep neural network.

[0023] Further, the feature extraction module is used to extract time-domain and frequency-domain features from the input signal. The input is the preprocessed ultrasonic signal, the output is the feature vector, and the structure is a feature extractor based on wavelet transform and Fourier transform;

[0024] The feature classification module is used to perform preliminary classification on the extracted features. The input is the feature vector, the output is the preliminary classification result, and the structure is a support vector machine classifier;

[0025] The feature fusion module is used to fuse the features and classification results of multiple signals. The input is the feature vectors and preliminary classification results of multiple signals, the output is the fused features, and the structure is a feature fusion network with an attention mechanism;

[0026] The deep neural network is used for the final species recognition. The input is the fused features, the output is the bat species and its confidence level, and the structure is a deep learning model composed of a multi-layer convolutional neural network and a fully connected layer.

[0027] Further, the steps for establishing the training data set of the bat species recognition model specifically include: collecting ultrasonic signal samples of multiple bat species, performing data augmentation processing, labeling the bat species of each sample, and dividing the training set, validation set, and test set.

[0028] Further, the steps for training the bat species recognition model specifically include: initializing the model parameters, performing forward propagation and backward propagation using the training set data, updating the model parameters, evaluating the model performance on the validation set, and repeating the training until the model converges or reaches a predetermined number of training epochs.

[0029] The predetermined number of training epochs is 2000.

[0030] Further, the steps for fitting the Doppler compensation equations include: first, collecting the flight data of known bat species and the corresponding ultrasonic signal samples; then, using the least squares method to fit the various parameters of the Doppler compensation equations, such as the velocity estimation parameter, direction estimation parameter, etc.; optimizing the performance of the equations through cross-validation to enable better removal of the Doppler effect and obtain the accurate original transmitted signal.

[0031] Further, the specific expressions of the equations in the Doppler compensation equations are as follows:

[0032] 1. Velocity estimation equation:

[0033]

[0034] In the formula, v is the flight speed of the bat; c is the speed of sound; f 0 is the original ultrasonic frequency emitted by the bat; f 1 , f 2 , f 3 are the signal frequencies received by three receivers respectively; t is the time; ε v is the velocity estimation error.

[0035] 2. Direction estimation equation:

[0036]

[0037] In the formula, θ is the azimuth angle of the bat's flight direction; φ is the elevation angle of the bat's flight direction; ε θ and ε φ are the estimation errors of the azimuth angle and elevation angle respectively.

[0038] 3. Signal reconstruction equation:

[0039]

[0040] In the formula, s(t) is the reconstructed original signal; A is the signal amplitude; f 0 is the fundamental frequency; φ 0 is the initial phase; a k , f k and φ k are the amplitude, frequency and phase of the k-th harmonic component respectively; N is the number of harmonics considered; εs is the signal reconstruction error.

[0041] 4. Frequency offset equation:

[0042]

[0043] where Δf is the frequency offset caused by the Doppler effect; α is the angle between the flight direction of the bat and the line connecting to the receiver; ε f is the frequency offset estimation error.

[0044] 5. Phase compensation equation:

[0045]

[0046] where Δφ is the phase change caused by the Doppler effect; d / dt represents the derivative with respect to time; ε p is the phase compensation error.

[0047] The method for obtaining parameters is as follows:

[0048] f 0 is obtained through experiments, including Step 1: Recording the ultrasonic signals of known species of bats in an anechoic environment; Step 2: Conducting spectral analysis on the recorded signals to determine the main frequency components.

[0049] f 1 、f 2 、f 3 are obtained through real-time measurement. The specific steps are: Step 1: Synchronously collecting signals using three ultrasonic receivers; Step 2: Conducting short-time Fourier transform (STFT) on the collected signals to obtain the time-frequency spectrum; Step 3: Tracking the changes of the main frequency components on the time-frequency spectrum.

[0050] The initial estimate of v is calculated through the following steps:

[0051]

[0052] where λ is the ultrasonic wavelength; T is the sampling time interval; Δφ 1 、Δφ 2 、Δφ 3 are the phase changes of the three receivers between two adjacent samplings respectively.

[0053] α is calculated through geometric relationships:

[0054]

[0055] where is the position vector from the bat to the receiver; is the velocity vector of the bat.

[0056] A, a k , φ k Obtained by performing Fourier analysis on the reconstructed signal. The specific steps are as follows: Step 1: Perform a fast Fourier transform (FFT) on the signal after removing the Doppler effect; Step 2: Extract the amplitude and phase information of each frequency component from the FFT result.

[0057] Each error term ε v , ε θ , ε φ , ε s , ε f , ε p Determine its distribution characteristics and range through multiple experimental measurements and statistical analyses.

[0058] The second aspect of the present invention provides a computer-readable storage medium, wherein program instructions are stored in the computer-readable storage medium, and when the program instructions run on a computer, they are used to execute the above-mentioned method for ultrasonic identification of bat species based on spatial orientation.

[0059] The third aspect of the present invention provides a system for ultrasonic identification of bat species based on spatial orientation, which includes the above-mentioned computer-readable storage medium.

[0060] Compared with the prior art, the beneficial effects of the method, medium, and system for ultrasonic identification of bat species based on spatial orientation provided by the present invention are as follows:

[0061] 1. Effective compensation for the Doppler effect. Existing methods are difficult to adapt to the Doppler frequency shift problem in complex environments, while the method of the present invention can estimate the flight speed and direction of bats based on the frequency and phase differences of the received signals by establishing a Doppler compensation equation set, thereby effectively eliminating the influence of the Doppler effect and obtaining accurate characteristics of the original transmitted signal.

[0062] 2. Fusion recognition of multiple signal sources. The present invention uses ultrasonic receivers at three different positions, makes full use of the signal characteristics of multiple observation points, and through a deep learning model for feature extraction, classification, and fusion, can more comprehensively depict the spatial motion characteristics of bats, thereby improving the accuracy and stability of recognition.

[0063] 3. Adapt to complex field environments. Compared with the existing method with a single signal source, the method of the present invention can resist noise interference in complex environments, and with the fusion of multiple signal sources and the compensation of the Doppler effect, it shows more robust performance in field practical applications. This provides an effective means for the real-time monitoring and protection of bat species.

[0064] In summary, the method for ultrasonic identification of bat species based on spatial orientation proposed by the present invention realizes more accurate and stable identification of bat species in complex field environments through multi-signal source fusion and Doppler effect compensation, and solves the technical problem that in the existing method for identifying bat species based on ultrasonic waves, due to the fact that bats may be in different flight states, the ultrasonic signals they emit will be affected by the Doppler effect, resulting in a decrease in the identification accuracy rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a flowchart of the method provided by the present invention;

[0066] Figure 2 is a waveform diagram of three ultrasonic signals after filtering processing;

[0067] Figure 3 is a comparison diagram of the bat species identification accuracy rates before and after removing the Doppler effect;

[0068] Figure 4 is a confidence distribution diagram of bat species identification based on deep learning. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0070] As Figure 1 shown, it is a flowchart of a method for ultrasonic identification of bat species based on spatial orientation provided by the first aspect of the present invention. This method includes the following steps:

[0071] S10. Obtain a first acquisition signal, a second acquisition signal, and a third acquisition signal collected by three ultrasonic signal collectors;

[0072] S20. Perform preprocessing on the three ultrasonic signals to obtain a first ultrasonic signal, a second ultrasonic signal, and a third ultrasonic signal;

[0073] S30. Use the pre-fitted Doppler compensation equation set, input the relative coordinates of the three ultrasonic signals and the three ultrasonic signals, and obtain three ultrasonic signals without the Doppler effect, which are respectively denoted as a first signal, a second signal, and a third signal;

[0074] S40. Input the first signal, the second signal, and the third signal into a pre-trained bat species identification model respectively to obtain the corresponding bat species and their confidence levels;

[0075] S50. Select the bat species with the highest confidence level as the result of the identified bat species.

[0076] The specific implementation manners of the above steps are described in detail as follows:

[0077] The specific implementation manner of step S10 is as follows:

[0078] First, three ultrasonic signal collectors are used to obtain ultrasonic signals at three different azimuth angles respectively. The arrangement positions of the three collectors have certain spatial angle differences relative to the level flight path of the bat, and their relative position coordinates can be expressed as and The three collected signals are respectively denoted as the first collected signal s 1 (t), the second collected signal s 2 (t), and the third collected signal s 3 (t).

[0079] The specific implementation manner of step S20 is as follows:

[0080] Preprocess the three collected signals obtained. First, filter the signals to eliminate noise interference and improve the signal quality. The three filtered signals are respectively denoted as s 1 ′(t), s 2 ′(t), and s 3 ′(t). Then, perform normalization processing on these three preprocessed signals to eliminate the influence of amplitude differences. The three normalized signals are denoted as and Next, perform time-frequency analysis on these three signals, such as short-time Fourier transform, to obtain the time-domain and frequency-domain characteristics of the signals. Denote the time-frequency analysis results as S 1 (t, f), S 2 (t, f), and S 3 (t, f). After the above preprocessing steps, three corrected and enhanced ultrasonic signals are obtained, which are respectively denoted as the first ultrasonic signal x 1 (t), the second ultrasonic signal x 2 (t), and the third ultrasonic signal x 3 (t).

[0081] The specific implementation manner of step S30 is as follows:

[0082] Utilize the pre-fitted Doppler compensation equation set, input the relative coordinates and of the three ultrasonic signal collectors, as well as the three collected ultrasonic signals x 1 (t), x 2 (t), and x 3 (t), to obtain three ultrasonic signals with the Doppler effect removed. The Doppler compensation equation set includes the following sub-equations:

[0083] 1. Velocity estimation equation:

[0084]

[0085] Wherein, v is the flight speed of the bat; c is the speed of sound; f 0 is the original ultrasonic frequency emitted by the bat; f 1 , f 2 , f 3 are the signal frequencies received by the three receivers respectively; t is the time; ε v is the speed estimation error. This equation estimates the flight speed of the bat by using the frequency derivatives of the three received signals.

[0086] 2. Direction estimation equation:

[0087]

[0088] Wherein, θ is the azimuth angle of the flight direction of the bat; φ is the pitch angle of the flight direction of the bat; ε θ and ε φ are the estimation errors of the azimuth angle and the pitch angle respectively. This equation estimates the flight direction angle of the bat by using the frequency differences of the three received signals.

[0089] 3. Signal reconstruction equation:

[0090]

[0091] Wherein, s(t) is the reconstructed original signal; A is the signal amplitude; f 0 is the fundamental frequency; φ 0 is the initial phase; a k , f k and φ k are the amplitude, frequency and phase of the k-th harmonic component respectively; N is the number of harmonics considered; ε s is the signal reconstruction error. This equation reconstructs the original transmitted signal through Fourier analysis by using the estimated speed and direction information.

[0092] 4. Frequency offset equation:

[0093]

[0094] Wherein, Δf is the frequency offset caused by the Doppler effect; α is the angle between the flight direction of the bat and the connection line of the receivers; ε f is the frequency offset estimation error. This equation calculates the frequency offset of the received signal caused by the Doppler effect.

[0095] 5. Phase compensation equation:

[0096]

[0097] where Δφ is the phase change caused by the Doppler effect; d / dt represents the derivative with respect to time; ε p is the phase compensation error. This equation calculates the phase change of the received signal caused by the Doppler effect.

[0098] By applying the above Doppler compensation equation set, the Doppler effect in the three acquired signals x 1 (t), x 2 (t) and x 3 (t) can be removed to obtain three ultrasonic signals without the Doppler effect, which are respectively denoted as the first signal y 1 (t), the second signal y 2 (t) and the third signal y 3 (t).

[0099] The specific implementation of step S40 is as follows:

[0100] Input the aforementioned first signal y 1 (t), the second signal y 2 (t) and the third signal y 3 (t) into the pre-trained bat species recognition model respectively. This recognition model includes four parts: a feature extraction module, a feature classification module, a feature fusion module, and a deep neural network:

[0101] Feature extraction module:

[0102] This module extracts time-domain and frequency-domain features from the input signal based on wavelet transform and Fourier transform to form a feature vector and Wavelet transform can capture the time-frequency characteristics of the signal, while Fourier transform can extract spectral features. The dimension of the feature vector is denoted as D f .

[0103] Feature classification module:

[0104] This module uses a support vector machine (SVM) to preliminarily classify the extracted features and give a preliminary classification result and SVM is a commonly used supervised learning classification algorithm that can effectively handle high-dimensional features.

[0105] Feature fusion module:

[0106] This module uses an attention mechanism to fuse the features and classification results of the three signals to obtain a fused feature vector The attention mechanism can adaptively allocate the importance of different features and enhance the weight of key features.

[0107] Deep neural network model:

[0108] Finally, the deep neural network model performs the final species recognition and outputs the bat species. and its confidence level This network includes multiple convolutional layers and fully connected layers, and can effectively extract high-level feature representations.

[0109] The specific implementation manner of step S50 is as follows:

[0110] Select the bat species with the highest confidence level among the three signals as the final recognition result.

[0111] In summary, the ultrasonic bat species recognition method based on spatial orientation proposed by the present invention can accurately identify the bat species through the fusion analysis of multiple ultrasonic signals and in combination with Doppler effect compensation. This method provides an effective recognition means for bat protection and field monitoring.

[0112] Now summarize the meanings of the variables in each formula:

[0113] v: flight speed of the bat; c: speed of sound; f 0 : original ultrasonic wave frequency emitted by the bat; f 1 , f 2 , f 3 : signal frequencies received by the three receivers; t: time; ε v : speed estimation error; θ: azimuth angle of the bat flight direction; φ: pitch angle of the bat flight direction; ε θ , ε φ : estimation errors of the azimuth angle and pitch angle; s(t): reconstructed original signal; A: signal amplitude; φ 0 : initial phase; a k , f k and φ k : amplitude, frequency and phase of the k-th harmonic component; N: number of harmonics considered; ε s : signal reconstruction error; Δf: frequency shift caused by the Doppler effect; α: angle between the bat flight direction and the connection line of the receivers; ε f : frequency shift estimation error; Δφ: phase change caused by the Doppler effect; ε p : phase compensation error; feature vector output by the feature extraction module; D f : dimension of the feature vector; preliminary classification result output by the feature classification module; fused feature vector output by the feature fusion module; finally recognized bat species; confidence level corresponding to the species;

[0114] In summary, the ultrasonic bat species identification method based on spatial orientation proposed by the present invention can accurately identify bat species through the fusion of Doppler effect compensation and deep learning models, providing an effective means for bat protection and monitoring.

[0115] The second aspect of the present invention provides a computer-readable storage medium, wherein program instructions are stored in the computer-readable storage medium, and when the program instructions run on a computer, they are used to execute the above-mentioned ultrasonic bat species identification method based on spatial orientation.

[0116] The third aspect of the present invention provides an ultrasonic bat species identification system based on spatial orientation, which includes the above-mentioned computer-readable storage medium.

[0117] Specifically, the principle of the present invention is as follows:

[0118] First, the method uses three ultrasonic receivers to obtain bat ultrasonic signals at different angles respectively. The relative positions of these three receivers are arranged with a certain spatial angle difference, which can capture the stereo information during the flight of the bat. Compared with a single signal source, the signal fusion of multiple observation points can more comprehensively describe the motion characteristics of the bat and provide richer input data for subsequent species identification.

[0119] Secondly, after obtaining multiple ultrasonic signals, the method of the present invention first preprocesses these signals, including steps such as filtering, normalization, and time-frequency analysis. These preprocessing operations can effectively eliminate environmental noise, correct amplitude differences, and extract characteristic information in the time domain and frequency domain.

[0120] Next, the present invention proposes a set of Doppler compensation equations to estimate the flight speed and direction of the bat and reconstruct the original transmitted signal according to the frequency difference and phase difference of the three received signals. The Doppler effect will seriously interfere with the characteristics of bat ultrasonic signals, and through the compensation of this set of equations, this frequency shift effect can be effectively eliminated to obtain more accurate original signal characteristics.

[0121] Finally, the present invention uses a deep learning-based bat species identification model. This model includes multiple modules such as feature extraction, feature classification, and feature fusion, which can make full use of the characteristic information of multiple signal sources, adaptively fuse the importance of different features through the attention mechanism, and achieve accurate identification of bat species. Compared with existing machine learning methods, the deep learning model has stronger feature expression ability and can learn more robust identification features from complex signals.

[0122] In summary, the ultrasonic bat species recognition method based on spatial orientation proposed by the present invention can achieve accurate and reliable bat species recognition in complex environments through the fusion of multi-signal sources, the compensation of the Doppler effect, and the application of deep learning models.

[0123] To better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: A certain nature reserve designed the following specific implementation plan according to the ultrasonic bat species recognition method based on spatial orientation proposed by the present invention:

[0124] First, three ultrasonic signal collectors were arranged in the monitoring area of the protection center. The relative positions of these three collectors are shown in Table 1. They form a triangular layout and can capture ultrasonic signals emitted by bats from different angles.

[0125] Table 1 Relative position coordinates of three ultrasonic signal collectors

[0126] Collector number x coordinate (m) y coordinate (m) z coordinate (m) 1 0 0 2 2 3 0 2 3 1.5 2.6 2

[0127] The three collectors respectively recorded the ultrasonic signals emitted by the bats. The phase and frequency differences of these signals can reflect the flight speed and direction characteristics of the bats. Immediately afterwards, preprocessing was performed on the three collected ultrasonic signals. First, a Butterworth filter was used to filter the signals to remove environmental noise interference. The three filtered signals are as Figure 2 shown. Then, normalization processing was performed on the filtered signals to eliminate the influence of amplitude differences. Finally, short-time Fourier transform (STFT) was used to perform time-frequency analysis on these three signals to obtain time-domain and frequency-domain characteristics.

[0128] Next, a set of Doppler compensation equations was established to estimate the flight speed and direction of the bat based on the frequency differences and phase differences of the three received signals, and to reconstruct the original transmitted signal. The specific compensation process is as follows:

[0129] First, the flight speed v of the bat was calculated using the speed estimation equation:

[0130]

[0131] where c is the speed of sound, f 0 is the original ultrasonic frequency emitted by the bat, f 1 , f 2 , f 3 are the signal frequencies received by the three receivers respectively, t is time, and ε v is the speed estimation error.

[0132] Then, the flight direction angles θ and φ of the bat were calculated using the direction estimation equation:

[0133]

[0134] Among them, θ is the azimuth angle of the bat's flight direction, φ is the pitch angle of the bat's flight direction, and ε θ and ε φ are the estimation errors of the azimuth angle and the pitch angle respectively.

[0135] Next, the original transmitted signal s(t) is reconstructed using the signal reconstruction equation:

[0136]

[0137] Among them, A is the signal amplitude, f 0 is the fundamental frequency, φ 0 is the initial phase, a k , f k and φ k are the amplitude, frequency and phase of the k-th harmonic component respectively, N is the number of harmonics considered, and ε s is the signal reconstruction error.

[0138] Finally, using the frequency offset equation and the phase compensation equation, the frequency offset Δf and the phase change Δφ caused by the Doppler effect are calculated, and the reconstructed signal is compensated accordingly.

[0139] After the above Doppler compensation processing, three ultrasonic signals y 1 (t), y 2 (t) and y 3 (t) without Doppler effect are obtained. The time-domain and frequency-domain characteristics of these three signals more accurately reflect the original ultrasonic characteristics emitted by the bat.

[0140] With the three ultrasonic signals without Doppler effect, they can be input into a pre-trained bat species recognition model for classification. The recognition model includes four parts: a feature extraction module, a feature classification module, a feature fusion module, and a deep neural network:

[0141] The feature extraction module extracts time-frequency feature vectors from the three signals respectively based on wavelet transform and Fourier transform and

[0142] The feature classification module uses the support vector machine (SVM) algorithm to preliminarily classify the above feature vectors and gives preliminary classification results and

[0143] The feature fusion module uses the attention mechanism to adaptively fuse the features and classification results of the three signals to obtain the fused feature vectors

[0144] Finally, the deep neural network model is based on a convolutional neural network and a fully connected layer, and inputs the fused features. Outputs the final bat species and its confidence

[0145] To verify the effectiveness of the method of the present invention, within the monitoring area of this protection center, a one-week experimental monitoring was carried out on three typical endangered bats (Myotis myotis, Pipistrellus pipistrellus, and Miniopterus schreibersii). During this period, a total of 3,600 ultrasonic signal samples were collected, including 1,200 of Myotis myotis, 1,400 of Pipistrellus pipistrellus, and 1,000 of Miniopterus schreibersii.

[0146] First, the ultrasonic signal characteristics of different bat species in different flight states were counted, as shown in Table 2. It can be seen that there are certain differences in the characteristics such as the vocalization frequency and flight speed of the three bats. This lays a foundation for subsequent species identification.

[0147] Table 2 Statistical results of ultrasonic signal characteristics of three typical endangered bats

[0148] Feature Myotis ikonnikovi Pipistrellus abramus Pipistrellus pipistrellus Main frequency (kHz) 40±3 35±4 45±5 Speed (m / s) 8±2 10±3 6±2 Flight angle (°) 15±5 25±6 30±7

[0149] To evaluate the effect of the method of the present invention in removing the Doppler effect, the recognition accuracies of the three bats before and after processing were respectively counted, and the results are as Figure 3 shown. It can be seen that without Doppler compensation, due to the influence of frequency shift, the recognition accuracy is relatively low, especially for Pipistrellus pipistrellus with a relatively fast flight speed. After Doppler effect compensation, the recognition accuracies of the three bats have been significantly improved, with an average increase of 15 percentage points. This shows that the Doppler compensation method of the present invention can effectively eliminate frequency interference in a complex environment and provide more reliable signal characteristics for subsequent species identification.

[0150] On this basis, the three ultrasonic signals after removing the Doppler effect were input into the trained deep learning recognition model for species classification. The average recognition accuracy of this model on the test set reached 92.4%, which is much higher than the traditional machine learning method. Further, the confidence distribution of the model for the three bats was counted, as Figure 4 shown. It can be seen that for most samples, the model can give high-confidence predictions, indicating that its recognition performance is quite stable.

[0151] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A method for ultrasonic identification of bat species based on spatial orientation, characterized in that: The following steps are involved: S10, acquiring a first acquisition signal, a second acquisition signal, and a third acquisition signal acquired by three ultrasonic signal collectors; S20, preprocessing the three ultrasonic signals to obtain a first ultrasonic signal, a second ultrasonic signal, and a third ultrasonic signal; S30, using a pre-fitted Doppler compensation equation group, inputting the relative coordinates of the three ultrasonic signals and the three ultrasonic signals, to obtain three ultrasonic signals with the Doppler effect removed, which are respectively recorded as a first signal, a second signal, and a third signal; S40, respectively inputting the first signal, the second signal and the third signal into a pre-trained bat species recognition model to obtain corresponding bat species and their confidence levels; S50: Select the bat species with the highest confidence as the identified bat species result.

2. The method for ultrasonic identification of bat species based on spatial orientation according to claim 1, characterized in that: The Doppler compensation equation group is used to obtain the bat's flight speed and direction and the original transmission signal according to the relative positions of the three ultrasonic signal collectors and the received signals; It includes speed estimation equation, direction estimation equation, signal reconstruction equation, frequency offset equation and phase compensation equation.

3. The method for ultrasonic identification of bat species based on spatial orientation according to claim 2, characterized in that: The speed estimation equation is used to estimate the bat's flying speed based on the frequency difference of the three received signals; The direction estimation equation is used to estimate the bat's flight direction based on the phase difference of the three received signals; The signal reconstruction equation is used to reconstruct the original transmission signal based on the estimated speed and direction; The frequency shift equation is used to calculate the frequency shift caused by the Doppler effect; The phase compensation equation is used to calculate the phase change due to the Doppler effect.

4. The method for ultrasonic identification of bat species based on spatial orientation according to claim 3, characterized in that: The bat species recognition model includes a feature extraction module, a feature classification module, a feature fusion module and a deep neural network.

5. The method for ultrasonic identification of bat species based on spatial orientation according to claim 4, characterized in that: The feature extraction module is used to extract time domain and frequency domain features from the input signal. The input is the preprocessed ultrasonic signal, and the output is a feature vector. The structure is a feature extractor based on wavelet transform and Fourier transform. The feature classification module is used to perform preliminary classification on the extracted features, the input is a feature vector, the output is a preliminary classification result, and the structure is a support vector machine classifier; The feature fusion module is used to fuse the features and classification results of multiple signals. The input is the feature vectors and preliminary classification results of multiple signals, and the output is the fused features. The structure is a feature fusion network of the attention mechanism. The deep neural network is used for final species identification. The input is the fused features, and the output is the bat species and its confidence. The structure is a deep learning model composed of a multi-layer convolutional neural network and a fully connected layer.

6. The method for ultrasonic identification of bat species based on spatial orientation according to claim 5, characterized in that: The steps of establishing the training data set of the bat species identification model specifically include: collecting ultrasonic signal samples of multiple bats, performing data enhancement processing, marking the bat species of each sample, and dividing the sample into a training set, a validation set, and a test set.

7. The method for ultrasonic identification of bat species based on spatial orientation according to claim 6, characterized in that: The steps of training the bat species identification model specifically include: initializing model parameters, using training set data for forward propagation and back propagation, updating model parameters, evaluating model performance on a validation set, and repeating training until the model converges or reaches a predetermined number of training rounds.

8. The method for ultrasonic identification of bat species based on spatial orientation according to claim 3, characterized in that: The step of fitting the Doppler compensation equation group specifically includes: collecting flight data of known bat species and corresponding ultrasonic signals, fitting the parameters of the Doppler compensation equation group using the least squares method, and optimizing the performance of the equation group through cross-validation.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the method for ultrasonic identification of bat species based on spatial orientation as described in any one of claims 1-8.

10. A bat species ultrasonic identification system based on spatial orientation, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images