A bird sound recognition method based on optimized selection of sound features

Abstract

The present invention discloses a bird sound recognition method based on optimized selection of sound features in the field of sound recognition technology, comprising the following steps: First, perform Fourier transform on the bird sound signal and weight it; calculate the power spectrum of the bird sound, design two groups of filters to filter the power spectrum, and respectively extract the coefficients of the filtered signals after filtering, calculate the covariance matrix of the coefficients of the signals after filtering for each group to obtain the optimized bird sound features, and finally form the feature vector of the bird sound by combining the optimized bird sound features of the two groups; construct a non-linear classification model f(x), and use an improved optimization method to find the optimal kernel function in the non-linear classification model; input the extracted bird sound feature vector into the non-linear classification model for learning. Compared with other bird sound recognition methods, the bird sound recognition method of the present invention has better classification accuracy for bird sounds by designing new bird sound features and optimizing selection and optimizing the kernel parameters of the non-linear classification method, and has short calculation time, is simple and easy to implement, and has good real-time performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of sound recognition, and particularly relates to a bird sound recognition method based on optimized selection of sound features. Background Technique

[0002] Birds can sensitively perceive minute changes in the environment and are the most ideal ecological species for monitoring environmental changes. With the development of sound recognition technology, considering that bird sounds and the morphological characteristics of birds are both important biological characteristics, each bird sound is unique and easier to capture than morphological characteristics. Bird monitoring can be completed by collecting and recognizing wild bird sounds. In the field of sound recognition algorithms, the combination of concise and effective sound features and high-precision recognition models is a popular research direction. Commonly used sound features include formant frequency, line spectrum pair, Mel frequency cepstral coefficient, short-time energy, short-time average zero-crossing rate, and amplitude, etc. Currently, most sound recognition technologies are applied to the fields of music and speech recognition, and the effect is not ideal when applied to bird sound recognition. It is necessary to study sound feature parameters suitable for bird sounds in order to better recognize bird sounds. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a bird sound recognition method based on optimized selection of sound features to solve the problems raised in the above background technique.

[0004] The purpose of the present invention can be achieved through the following technical solutions:

[0005] A bird sound recognition method based on optimized selection of sound features includes the following steps:

[0006] Step 1: First, perform Fourier transform and weighting on the bird sound signal d(n), and the weighted result is DQ(k);

[0007] Step 2: Calculate the power spectrum P(k) of DQ(k), design two groups of filters to filter the power spectrum P(k) of the bird sound, and respectively extract the coefficients C1(n) and C2(n) of the filtered signals S1(m) and S2(m). Calculate the covariance matrices C and F for the coefficients of each group of filtered signals, find the eigenvalues and eigenvectors of the covariance matrices, arrange them from large to small, and select the largest first L1 (the typical value of L1 is equal to the integer of L / 2) to obtain the optimized bird sound features. Finally, combine the two groups of optimized bird sound features x1 and x2 to form the feature vector x of the bird sound, x = [x1, x2];

[0008] Step 3: Construct a non-linear classification model f(x), and use an improved optimization method to find the optimal kernel function γ in the non-linear classification model i+1,j, where the fitness of the position during the search process is the number of identical recognition results and actual results of bird sound recognition using a kernel function with the current position as the kernel parameter;

[0009] Step 4: Input the extracted bird sound feature vectors into a non-linear classification model for learning;

[0010] Step 5: Extract the feature vectors of the bird sound to be recognized, and input the feature vectors into the well-learned non-linear classification model to identify the type of bird.

[0011] Preferably, the Fourier transform is as follows:

[0012]

[0013] The zero energy ratio Q of the bird sound is calculated by the following formula:

[0014]

[0015] Preferably, the calculation formula of the filtered signal S1(m) in step 2 is as follows:

[0016]

[0017]

[0018]

[0019] The coefficient C1(n) of the filtered signal S1(m) is calculated by the following formula:

[0020]

[0021] Preferably, calculate the coefficient of the bird sound in section A according to C1(n), denoted as B1(s,t) (s = 1, 2... A, t = 1, 2... L), and de-mean B1 to obtain B:

[0022]

[0023] Calculate the covariance matrix C as shown below:

[0024]

[0025]

[0026] Find the eigenvalue λ1 and eigenvector u1 of C:

[0027] Cu1 = λ1u1

[0028] Calculate the optimized first group of features x1 according to the following formula:

[0029] x1(p) = u1(p) T[C1(1),……C1(L)] T p = 1, 2…L1.

[0030] Preferably, the calculation formula of the filtered signal S2(m) in step 2 is as follows:

[0031]

[0032]

[0033] The coefficient C2(n) of the filtered signal S2(m) is calculated by the following formula:

[0034]

[0035] Preferably, the coefficient of the A-section bird sound is calculated according to C2(n), denoted as E1(s, t), (s = 1, 2…A, t = 1, 2…L), and the mean value of E1 is removed to obtain E:

[0036]

[0037] The covariance matrix F is obtained as follows:

[0038]

[0039]

[0040] The eigenvalues λ2 and eigenvectors u2 of F are obtained according to the following formula:

[0041] Fu2 = λ2u2

[0042] The optimized second set of features x2 is calculated according to the following formula:

[0043] x2(p) = u2(p) T [C2(1),……C2(L)] T p = 1, 2…L1.

[0044] Preferably, the constructed non-linear classification model f(x) in step 3 is as follows:

[0045]

[0046] Preferably, the optimization process of the optimal kernel function in step 3 is as follows:

[0047] First, the kernel parameters are regarded as crows, and the population number P1, the maximum number of iterations maxi, and the upper and lower bounds of the target space are initialized. The initial positions of the crows and the food are set as follows:

[0048] γ0 = rand(1, g)*(ub - lb)+lb

[0049] Then update the position as follows:

[0050]

[0051] D = |C·wz i+1,j - X(i,t)|

[0052] A = 2a·r1

[0053] C = 2·r2

[0054] Next, calculate the fitness of the position as follows:

[0055] shf(γ i,j ) = cur i,j

[0056] shf(γ i+1,j ) = cur i+1,j

[0057] Finally, update the food position wz based on the fitness i+1,j :

[0058]

[0059] Judge whether the iteration number i is less than maxi. If so, return to execute the position update. If not, then use the γ at this time i+1,j as the optimal kernel parameter output.

[0060] Advantages of the present invention:

[0061] 1. Compared with other bird sound recognition methods, the bird sound recognition method of the present invention has better classification accuracy for bird sounds by designing new bird sound features and optimizing the selection and optimization of the kernel parameters of the non - linear classification method, and has short calculation time, is simple and easy to implement, and has good real - time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0063] Figure 1 is the flowchart of the bird sound recognition method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0065] Please refer to Figure 1 As shown, the present invention proposes a bird sound recognition method based on optimized selection of sound features, including the following steps:

[0066] Step 1: First, perform frequency-domain weighting on the bird sound.

[0067] Let the collected time-domain bird sound signal be d(n) (n = 0, 1, 2,..., N - 1), where N is the length of the bird sound;

[0068] Perform Fourier transform and weighting on d(n), where the Fourier transform is as follows:

[0069]

[0070] In the formula, r is the imaginary unit, D(k) represents the kth data in the spectrum of the bird sound signal, where 0 ≤ k ≤ N - 1.

[0071] Calculate the zero-energy ratio Q of the bird sound according to the following formula:

[0072]

[0073] In the formula, d(n) represents the nth data of the bird sound signal, sgn[] is the sign function, num is a very small number, and d(n - 1) represents the (n - 1)th data of the bird sound signal;

[0074] Weight D(k) to obtain DQ(k):

[0075] DQ(k) = D(k)·Q

[0076] Step 2: Filter the power spectrum of the bird sound with two designed filters respectively, extract the coefficients of the filtered signals respectively, calculate the covariance matrix of the coefficients of each group of filtered signals, calculate the eigenvalues of the covariance matrix, arrange them from large to small, and select the first L1 (the typical value of L1 is equal to the integer of L / 2) to obtain the optimized bird sound features. Finally, combine the optimized bird sound features of the two groups to form the feature vector of the bird sound:

[0077] Step 2 specifically includes:

[0078] Step 2.1: Calculate the power spectrum P(k) of DQ(k):

[0079] P(k) = |DQ(k)| 2

[0080] Wherein, P(k) represents the k-th data in the power spectrum of the bird sound signal, and 0 ≤ k ≤ N - 1.

[0081] Step 2.2: Filter the power spectrum P(k) with the first set of filter banks to obtain the filtered signal S1(m):

[0082]

[0083]

[0084] In the above formula, f s ·k < N·F -1 (f1(m - 1)) or f s ·k ≥ N·F -1 (f1(m + 1)), H1 m (k) = 0;

[0085] N·F -1 (f1(m - 1)) ≤ f s ·k ≤ N·F -1 (f1(m)), H1 m (k) is as follows:

[0086]

[0087] N·F -1 (f1(m)) ≤ f s ·k ≤ N·F -1 (f1(m + 1)), H1 m (k) is as follows:

[0088]

[0089]

[0090] In the above three formulas, m represents the filter number, M represents the number of filters used, H1 m (k) represents the m-th filter in the first set of filter banks, f1(m), f1(m - 1), f1(m + 1) represent the center frequencies of the m-th, m - 1-th, and m + 1-th filters in the first set of filter banks, fs represents the sampling frequency, fh represents the highest frequency within the bird sound frequency range, fl represents the lowest frequency within the bird sound frequency range, F(*) = 1127 * ln(1 + * / 700), F -1 (*) = 700(e * / 1125 -1);

[0091] Step 2.3. Extract the coefficient C1(n) of the filtered signal S1(m) in Step 2.2 according to the following formula:

[0092]

[0093] In the formula, L represents the order;

[0094] Step 2.4. Calculate the coefficients of the bird sounds in segment A, denoted as B1(s, t) (s = 1, 2... A, t = 1, 2... L), and de-mean B1 to obtain B:

[0095]

[0096] Obtain the covariance matrix C according to the following formula

[0097]

[0098]

[0099] Step 2.5. Obtain the eigenvalue λ1 and eigenvector u1 of C according to the following formula:

[0100] Cu1 = λ1u1

[0101] There are L eigenvalues. Let the eigenvector corresponding to the nth (n = 1, 2... L) eigenvalue λ1(n) be u1(n). Arrange the eigenvalues from largest to smallest, select the first L1 largest ones, where the typical value of L1 is equal to the integer part of L / 2, and select the corresponding eigenvectors. Calculate the optimized first set of features x1 according to the following formula:

[0102] x1(p) = u1(p) T [C1(1),……C1(L)] T p = 1, 2... L1

[0103] Step 2.6. Filter the power spectrum P(k) with the second set of filter banks to obtain the filtered signal S2(m):

[0104]

[0105]

[0106] In the formula, when f s ·k < N·G -1 (f2(m - 1)) or f s ·k ≥ N·G -1 (f2(m + 1)) then, H2 m (k) = 0;

[0107] When N·G -1 (f2(m - 1)) ≤ fs · k ≤ N·G -1 When (f2(m)), H2 m (k) is as follows:

[0108]

[0109] When N·G -1 (f2(m)) ≤ f s · k ≤ N·G -1 When (f2(m + 1)), H2 m (k) is as follows:

[0110]

[0111]

[0112] In the above three formulas, H2 m (k) represents the m-th filter in the second set of filter banks, f2(m), f2(m - 1), f2(m + 1) represent the center frequencies of the m-th, m - 1-th, and m + 1-th filters in the second set of filter banks, G(*) = 2195 - 2595 * log(1 + (4031 - *) / 700), G -1 (*) = 700(10 * / 2595 -1);

[0113] Step 2.7: Extract the coefficient C2(n) of the filtered signal S2(m) according to the following formula:

[0114]

[0115] Step 2.8: Calculate the coefficients of the A-section bird calls according to Step 2.7, denoted as E1(s, t), (s = 1, 2... A, t = 1, 2... L), and de-mean E1 to obtain E:

[0116]

[0117] Calculate the covariance matrix F according to the following formula:

[0118]

[0119]

[0120] Step 2.9: Calculate the eigenvalue λ2 and eigenvector u2 of F according to the following formula:

[0121] Fu2 = λ2u2

[0122] There are L eigenvalues. The eigenvector corresponding to the nth (n = 1, 2, …, L) eigenvalue λ2(n) is u2(n). Arrange the eigenvalues from largest to smallest, select the first L1 largest ones, and select their corresponding eigenvectors. Calculate the optimized second set of features x2 according to the following formula:

[0123] x2(p) = u2(p) T [C2(1), …… C2(L)] T p = 1, 2, …, L1

[0124] Step 2.10: Combine x1 and x2 to form the feature vector x, x = [x1, x2].

[0125] Step 3: Construct a non - linear classification model and use an improved optimization method to find the optimal kernel function in the non - linear classification model. Among them, the fitness of the position in the search process is the number of the same recognition results and actual results when using the kernel function with the current position as the kernel parameter for bird sound recognition:

[0126] Step 3 specifically includes:

[0127] Step 3.1: Construct a non - linear classification model f(x):

[0128]

[0129] In the formula, x is the feature vector of the bird sound, Q is the total number of segments of the bird sound (i.e., the training set) used to train the classification model, x q is the feature vector of the qth segment of the bird sound used to train the classification model, y q is the classification label of the qth segment of the bird sound, specifically - 1 or 1, - 1 indicates not this type of bird, 1 indicates this type of bird, a q is the weight coefficient, b is the preset bias parameter, a q , b are the parameters to be learned, is the kernel function, d1 is the kernel parameter;

[0130] Step 3.2: Use the variant crow optimization method to find the optimal kernel parameter of the kernel function, so as to obtain the optimal kernel function. The optimization process is as follows:

[0131] Step 3.2.1: Take the kernel parameter as a crow, initialize the population number P1, the maximum number of iterations maxi, the upper and lower bounds of the target space, and set the initial positions of the crow and the food according to the following formula;

[0132] γ0 = rand(1, g) * (ub - lb) + lb

[0133] Among them, γ0 is the initial position of the crow and the food, lb is the minimum value of the target space, ub is the maximum value of the target space, g is the number of variables, and rand(1,g) generates a random number matrix of 0-1 with one row and g columns;

[0134] Step 3.2.2: Update the position according to the following formula:

[0135]

[0136] D = |C·wz i+1,j -X(i,t)|

[0137] A = 2a·r1

[0138] C = 2·r2

[0139] In the formula, X(i+1,j) is the position of the j-th crow and food at the (i+1)-th iteration, X(i,j) is the position of the j-th crow and food at the i-th iteration, wz i,j is the position of the j-th food at the i-th iteration, Z is the adaptive parameter, and X(r1,j) is a random position of the j-th crow before the (i+1)-th iteration. a is a variable that decreases arithmetically from 2 to 0, and r1 and r2 are random vectors with a modulus less than 1. rd, r i and r j are random numbers uniformly distributed in (0,1), is the flight length of crow j at the (t+1)-th iteration, is the perception probability of crow j at the (i+1)-th iteration, where i,j∈P.

[0140] Step 3.2.3: Calculate the fitness of the position according to the following formula:

[0141] shf(γ i,j ) = cur i,j

[0142] shf(γ i+1,j ) = cur i+1,j

[0143] In the formula, shf(γ i,j ), shf(γ i+1,j ) respectively represent the fitness of positions γ i,j and γ i+1,j , cur i,j and cur i+1,j respectively represent the number of cases where the recognition results of bird species identification using kernel functions with γ i,j and γ i+1,j as kernel parameters are the same as the actual results;

[0144] Step 3.2.4: Update the food position wz based on the fitnessi+1,j :

[0145]

[0146] Step 3.2.5. Determine whether the iteration number i is less than maxi. If so, return to execute Step 3.2.2. If not, then use the γ at this time i+1,j as the optimal kernel parameter for output.

[0147] Step 4. Input the extracted bird sound feature vectors into the non-linear classification model for learning:

[0148] Step 4 specifically includes:

[0149] Step 4.1. Randomly divide the bird sound feature vectors in segment A into a training set and a test set according to a ratio of 8:2;

[0150] Step 4.2. Input the training set into the non-linear classification model f(x) to find the optimal a q and b, where:

[0151] When solving for a q a q should satisfy:

[0152]

[0153] and

[0154] When solving for b, randomly select a bird sound feature vector x * and its corresponding label y * , and calculate b according to the following formula:

[0155]

[0156] Step 4.3. Use the test set to evaluate the trained non-linear classification model. If the preset evaluation criteria are met, it is determined that the model training is completed.

[0157] Step 5. Extract the feature vectors of the bird sound to be recognized according to Steps 1 and 2, and input the feature vectors into the learned non-linear classification model to identify the type of bird.

[0158] Sensitivity (SE) and specificity (SP) are important evaluation indicators for the model in bird sound recognition. F1score is the comprehensive evaluation indicator for the model in bird sound recognition. F1score can be calculated through sensitivity and specificity:

[0159]

[0160] The F1 score can take into account both the sensitivity and specificity of the evaluation model, and the bird sound recognition model provided in this embodiment is evaluated through the comprehensive index F1 score.

[0161] To verify the effect of the method of the present invention, it is compared with the existing SVM method. Table 1 shows the recognition results of the two methods for the bird sounds of 5 kinds of birds, namely purple swamphen, cuckoo, house sparrow, kingfisher and house finch. The results show that the bird sound recognition provided in this embodiment has a higher recognition accuracy and better performance.

[0162] Table 1 Comparison results of the two methods

[0163]

[0164] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0165] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A bird sound recognition method based on optimized selection of sound features, characterized in that, It includes the following steps: Step 1: First, perform Fourier transform on the bird sound signal d(n) and weight it to obtain DQ(k) after weighting; Step 2: Calculate the power spectrum P(k) of DQ(k), design two groups of filters to filter the power spectrum P(k) of the bird sound, and respectively extract the coefficients C1(n) and C2(n) of the filtered signals S1(m) and S2(m) after filtering. Calculate the covariance matrices C and F for the coefficients of each group of filtered signals, find the eigenvalues and eigenvectors of the covariance matrices, arrange them from largest to smallest, select the first L1 largest ones to obtain the optimized bird sound features. Finally, form the feature vector x of the bird sound by combining the two groups of optimized bird sound features x1 and x2, x = [x1, x2]; Step 3: Construct a non-linear classification model f(x), and use an improved optimization method to find the optimal kernel function γ in the non-linear classification model i+1，j , where the fitness of the position in the search process is the number of the same recognition results between the recognition result of the bird sound using the kernel function with the current position as the kernel parameter and the actual result; Step 4: Input the extracted bird sound feature vector into a non-linear classification model for learning; Step 5: Extract the feature vector of the bird sound to be recognized and input the feature vector into the well-learned non-linear classification model to identify the type of the bird.

2. The bird sound recognition method based on optimized selection according to claim 1, wherein In the said Step 1, let the bird sound signal be d(n), n = 0, 1, 2, …, N - 1, and N is the length of the bird sound; The Fourier transform is as follows: where \(r\) is the imaginary unit, \(D(k)\) represents the \(k\)-th data in the spectrum of the bird sound signal, where \(0\leq k\leq N - 1\); The zero energy ratio Q of the bird sound is calculated by the following formula: In the formula, d(n) represents the n-th data of the bird sound signal, sgn[] is the sign function, num is a very small number, d(n - 1) represents the (n - 1)-th data of the bird sound signal, and DQ(k) = D(k)·Q is obtained by weighting D(k).

3. The bird sound recognition method based on optimized selection according to claim 1, characterized in that, In step 2, the power spectrum P(k) = |DQ(k)| 2 , where P(k) represents the k-th data in the power spectrum of the bird sound signal, 0 ≤ k ≤ N - 1, and the calculation formula for the filtered signal S1(m) is as follows: In the formula, m represents the filter number, M represents the number of filters used, H1 m (k) represents the m-th filter in the first group of filters, f1(m), f1(m - 1), f1(m + 1) represent the center frequencies of the m-th, (m - 1)-th, and (m + 1)-th filters in the first group of filters, fs represents the sampling frequency, fh represents the highest frequency within the bird sound frequency range, fl represents the lowest frequency within the bird sound frequency range, F(*) = 1127 * ln(1 + * / 700), F -1 (*) = 700(e * / 1125 - 1); The coefficient C1(n) of the filtered signal S1(m) is calculated by the following formula: In the formula, L represents the order.

4. The bird sound recognition method based on optimized selection according to claim 3, characterized in that, Calculate the coefficients of the bird sound in section A according to C1(n), denoted as B1(s, t), s = 1, 2, …, A, t = 1, 2, …, L, and de-mean B1 to obtain B: The covariance matrix C is obtained as follows: Find the eigenvalue λ1 and eigenvector u1 of C: Cu1 = λ1u1 There are L eigenvalues. Let the eigenvector corresponding to the n-th eigenvalue λ1(n) be u1(n), n = 1, 2, …, L. Arrange the eigenvalues from largest to smallest, select the first L1 largest ones, and the typical value of L1 is equal to the integer part of L / 2, and select the corresponding eigenvectors. Calculate the optimized first group of features x1 according to the following formula: x1(p) = u1(p) T [C1(1), …… C1(L)] T p = 1, 2…L1.

5. The method for bird sound recognition based on optimized selection according to claim 1, wherein In the said Step 2, the calculation formula of the filtered signal S2(m) is as follows: Wherein, when f s ·k < N·G -1 (f2(m - 1)) or f s ·k ≥ N·G -1 (f2(m + 1)), then H2 m (k) = 0; When N·G -1 (f2(m - 1)) ≤ f s ·k ≤ N·G -1 (f2(m)), then H2 m (k) is as follows: When N·G -1 (f2(m)) ≤ f s ·k ≤ N·G -1 (f2(m + 1)), H2 m (k) is as follows: In the above three equations, H2 m (k) represents the m-th filter in the second group of filters, and f2(m), f2(m - 1), f2(m + 1) represent the center frequencies of the m-th, (m - 1)-th, and (m + 1)-th filters in the second group of filters. G(*) = 2195 - 2595 * log(1 + (4031 - *) / 700), G -1 (*) = 700(10 * / 2595 - 1); The coefficient C2(n) of the filtered signal S2(m) is calculated by the following formula:

6. The method for bird sound recognition based on optimized selection of sound features according to claim 5, wherein, Calculate the coefficients of the bird sound in section A according to C2(n), denoted as E1(s, t), s = 1, 2, …, A, t = 1, 2, …, L, and de-mean E1 to obtain E: The covariance matrix F is obtained as follows: Find the eigenvalue λ2 and eigenvector u2 of F according to the following formula: Fu2 = λ2u2 There are L eigenvalues. The eigenvector corresponding to the n-th eigenvalue λ2(n) is u2(n), n = 1, 2, …, L. Arrange the eigenvalues from largest to smallest, select the first L1 largest ones, and select the corresponding eigenvectors. Calculate the optimized second group of features x2 according to the following formula: x2(p) = u2(p) T [C2(1), …… C2(L)] T p = 1, 2…L1.

7. The bird sound recognition method based on optimized selection of sound features according to claim 1, characterized in that, The non-linear classification model f(x) constructed in the said Step 3 is as follows: where x is the feature vector of the bird sound, Q is the total number of segments of the bird sound used to train the classification model, and x q is the feature vector of the q-th segment of the bird sound used to train the classification model, and y q is the classification label of the q-th segment of the bird sound, specifically -1 or 1, where -1 indicates that it is not this type of bird and 1 indicates that it is this type of bird, and a q is the weight coefficient, b is the preset bias parameter, and a q and b are parameters to be learned, is the kernel function, and d1 is the kernel parameter.

8. A bird sound recognition method based on optimized selection according to the sound characteristics as claimed in claim 7, characterized in that, The optimization process of the optimal kernel function in the said Step 3 is as follows: First, take the kernel parameter as a crow, and initialize the population number P1, the maximum number of iterations maxi, and the upper and lower bounds of the target space. The initial positions of the crow and the food are set as follows: γ0 = rand(1, g)*(ub - lb)+lb where γ0 is the initial position of the crow and the food, lb is the minimum value of the target space, ub is the maximum value of the target space, g is the number of variables, and rand(1, g) generates a random number matrix of 0 - 1 with one row and g columns; Then update the position as follows: D = |C·wz i+1，j - X(i, t)| A = 2a·r1 C＝2·r2 Wherein, X(i+1, j) is the position of the j-th crow and the food at the (i+1)-th iteration, X(i, j) is the position of the j-th crow and the food at the i-th iteration, wz i，j is the position of the j-th food at the i-th iteration, Z is an adaptive parameter, X(r1, j) is a random position of the j-th crow before the (i+1)-th iteration, a is a variable decreasing arithmetically from 2 to 0, r1 and r2 are random vectors with modulus less than 1, rd, r i and r j are random numbers uniformly distributed in (0, 1), is the flight length of crow j at the (t+1)-th iteration, is the perception probability of crow j at the (i+1)-th iteration, i, j ∈ P; Next, calculate the fitness of the position as follows: shf(γ i，j ) = cur i，j shf(γ i+1，j ) = cur i+1，j Where, shf(γ i，j ) and shf(γ i+1,j ) respectively represent the fitness of positions γ i，j and γ i+1，j , cur i，j and cur i+1，j respectively represent the number of cases where the recognition results of bird species identification using kernel functions with γ i，j and γ i+1，j as kernel parameters are the same as the actual results; Finally, update the food position wz based on the fitness i+1，j : Determine whether the iteration count i is less than maxi. If so, return to execute the position update. If not, then use the γ at this time i+1，j as the optimal kernel parameter for output.

9. The method for bird sound recognition based on optimized selection of sound features according to claim 1, characterized in that, The specific steps of step 4 include the following steps: Step 4.1: Randomly divide the A - segment bird sound feature vector into a training set and a test set according to the ratio of 8:2; Step 4.2: Input the training set into the non-linear classification model f(x) to find the optimal a q and b, where: Solve for a q When q a should satisfy: and When solving for b, arbitrarily select a bird sound feature vector x * and its corresponding label y * , and calculate b according to the following formula: Step 4.3: Use the test set to evaluate the trained non - linear classification model. If the preset evaluation criteria are met, it is determined that the model training is completed.

10. The bird sound recognition method based on optimized selection according to claim 1, characterized in that In step 5, the model is evaluated by the comprehensive index F1score as follows: In the formula, SE is the sensitivity and SP is the specificity.

Images