Photoacoustic spectrum feature screening method based on multi-strategy sled dog optimization algorithm

Abstract

The invention discloses a photoacoustic spectrum feature screening method based on a multi-strategy sled dog optimization algorithm. The method comprises the following steps: acquiring photoacoustic spectrum data as a training sample; performing feature screening on the photoacoustic spectroscopic data by adopting a multi-strategy improved sled dog optimization algorithm, wherein the multi-strategy improvement comprises performing population initialization by adopting Latin hypercube sampling, performing population individual variation updating by adopting a differential evolution strategy, and introducing a reverse learning strategy to perform population individual learning updating; training a photoacoustic spectrum regression model by using the screened characteristic wavelengths; and inputting to-be-predicted photoacoustic spectrum data into the trained photoacoustic spectrum regression model for feature prediction. According to the method, the initial population diversity of the sled dog optimization algorithm is improved, and premature convergence is avoided; the global search capability of the algorithm is enhanced, and the algorithm adapts to high-dimensional data optimization; and the convergence speed and the feature screening precision are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent spectral analysis technology, and in particular to a photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm. Background Technology

[0002] Currently, research on photoacoustic spectral feature screening has begun to attract attention in academia, but it is still in a relatively new stage compared to the more mature near-infrared or infrared spectroscopy analysis. Photoacoustic spectroscopy technology has gained widespread attention in recent years in fields such as environmental monitoring, biomedical detection, and industrial process control due to its advantages such as high sensitivity, non-invasiveness, and multi-domain adaptability. Especially in gas detection and biomedical imaging, photoacoustic spectroscopy can capture the characteristic absorption information of key molecules in samples, thereby achieving high-precision identification of harmful gas components or precise localization of biological tissue structures and lesions. However, photoacoustic spectral data often exhibits characteristics such as high dimensionality, high noise, and strong nonlinearity, making it difficult for traditional analysis methods to accurately extract key features from large-scale, multi-dimensional photoacoustic spectral data. How to extract the most discriminative and stable features from complex photoacoustic signals has become a key technical challenge for further improving the detection accuracy and application effect of photoacoustic spectroscopy.

[0003] Feature selection is crucial for processing high-dimensional data. By identifying the variables with the most information, it can significantly improve data processing efficiency and enhance the predictive performance and interpretability of models. In photoacoustic spectroscopy, effective feature selection methods not only impact the accuracy and efficiency of analysis but also reveal the intrinsic chemical and physical properties of substances, providing important support for scientific research and industrial applications. Traditional feature selection methods, such as Principal Component Analysis (PCA) and Partial Least Squares (PLS), while performing well in some applications, are not ideal for handling nonlinear and highly cross-correlated spectral data. These methods often rely on linear assumptions, making it difficult to capture complex nonlinear relationships between signals, thus limiting their application in more complex data analysis.

[0004] Swarm optimization algorithms possess strong global search capabilities and are suitable for handling nonlinear problems, and have been widely applied to feature selection problems. These algorithms, such as Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Whale Optimization (WAO), leverage information sharing among individuals in a swarm to effectively explore the solution space and identify potential solutions to the optimization problem. However, these swarm optimization algorithms still have the following drawbacks:

[0005] Traditional random initialization methods result in uneven distribution within the search space and insufficient diversity of the initial population, which affects global search capabilities.

[0006] The algorithm is prone to getting stuck in local optima during the iteration process and lacks an effective escape mechanism.

[0007] These algorithms often require adjusting multiple control parameters, making them complex to operate and prone to getting stuck in local optima. In particular, when faced with high-dimensional photoacoustic spectral data feature spaces, traditional optimization algorithms struggle to efficiently screen key features.

[0008] Therefore, how to provide a high-precision photoacoustic spectral feature screening method that can be adapted to high-dimensional data optimization is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] In response to the aforementioned research status, this invention provides a photoacoustic spectral feature selection method based on a multi-strategy sled dog optimization algorithm. This method improves the initial population diversity of the sled dog optimization algorithm, avoiding premature convergence; enhances the algorithm's global search capability, adapting to high-dimensional data optimization; and improves both convergence speed and feature selection accuracy.

[0010] This invention provides a photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm, comprising the following steps:

[0011] S1: Acquire photoacoustic spectral data as training samples;

[0012] S2: The photoacoustic spectral data is feature-selected using a multi-strategy improved sled dog optimization algorithm, wherein the multi-strategy improvement includes using Latin hypercube sampling for population initialization, using differential evolution strategy for population individual mutation update, and introducing a reverse learning strategy for population individual learning update.

[0013] S3: Train the photoacoustic spectral regression model using the selected characteristic wavelengths;

[0014] S4: Input the photoacoustic spectral data to be predicted into the trained photoacoustic spectral regression model for feature prediction.

[0015] Preferably, the photoacoustic spectral data in S1 is preprocessed photoacoustic spectral data, and the preprocessing step includes preprocessing the original photoacoustic spectral data using median filtering:

[0016]

[0017] In the formula, S ij This is for selecting the center point; f(i,j) is the original photoacoustic spectral data; g(i,j) is the photoacoustic spectral data after median filtering.

[0018] Preferably, the multi-strategy improved sled dog optimization algorithm in S2 includes the following steps:

[0019] S21: Using the Latin hypercube method, the photoacoustic spectral data is divided into equal intervals in each dimension. A value is randomly selected from each interval, and all values ​​are combined in a random order. The resulting initial solution set is used as the initial population.

[0020] S22: During population mutation, use individuals from other populations to perform differential evolution strategy mutations to update the velocity and position of the current population individuals;

[0021] S23: When the algorithm gets stuck in a local optimum or the population diversity is insufficient, a reverse learning strategy is introduced to perform reverse learning replacement of the current population individuals.

[0022] Preferably, step S21 includes the following steps:

[0023] S211: Search space [lb] d ,ub d The d-th dimension is divided into N intervals, lb d ub is the lower bound of the d-th dimension search space. d Let be the upper limit of the search space in the d-th dimension;

[0024] S212: Randomly generate a sample point in each interval, such that each interval is taken only once in each dimension;

[0025] S213: Combine sample points from all dimensions to obtain N sample points as the initial population. The i-th individual in the initial population... Represented as:

[0026]

[0027] In the formula, u i,d Let π be an integer in the range {0, 1, ..., N-1}. i,d It is a random number in the range [0,1).

[0028] Preferably, step S22 includes the following steps:

[0029] S221: Select individuals X from three different populations r X s X t Update the population individual X using the difference vector. i :

[0030] V i =X r +F·(X s -X t );

[0031] In the formula: X r X s X tLet x be the r-th individual in the population. r The position vector in the d-th dimension, and the s-th individual x in the population. s The position vector in the d-th dimension, and the t-th individual x in the population t The position vector in the d-th dimension; F is a scaling factor, V i This is the new vector obtained after mutation;

[0032] S222: Put this V i With individual X in the population i Perform crossover to obtain the test vector U i，d :

[0033]

[0034] In the formula, rand(0,1) is a random number generated from the uniform distribution (0,1); CR∈[0,1] is the preset crossover probability; k is a randomly selected dimension index;

[0035] S223: Calculate the full-dimensional experimental vector U i If the fitness is better than X i Then use U i Replace X i .

[0036] Preferably, step S23 includes the following steps:

[0037] S231: For the location X = (x1,...,x...) D Given the current population individuals, determine their reverse population individuals X′=(x1′,...,x′). D )for:

[0038] x′ d =lb d +ub d -x d ;

[0039] In the formula, lb d ub is the lower bound of the d-th dimension search space; d Let d be the upper limit of the d-th dimension search space, where d∈[1,D] and D is the total dimension;

[0040] S232: Calculate the fitness of X′. If it is better than X, replace X with X′.

[0041] The present invention also provides an electronic device, including a processor and a memory storing a computer program, characterized in that, when the processor executes the computer program, it implements the steps of the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm.

[0042] The present invention also provides a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm.

[0043] Compared with the prior art, the present invention has the following advantages:

[0044] This invention improves the initial population diversity of the sled dog optimization algorithm, avoiding premature convergence; it also enhances the algorithm's global search capability, adapting to high-dimensional data optimization.

[0045] Compared to traditional sled dog optimization algorithms, the convergence speed will be improved, and the feature selection accuracy will also be enhanced.

[0046] In photoacoustic spectral data, an improved sled dog optimization algorithm is used to screen features, which improves the detection accuracy of the photoacoustic spectral regression model. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present invention, and those skilled in the art can obtain other drawings based on the provided drawings without creative effort.

[0048] Figure 1 This is a flowchart of the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm provided in the embodiments of the present invention;

[0049] Figure 2 This is a flowchart of the multi-strategy improved sled dog optimization algorithm provided in the embodiments of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] This invention discloses a photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm (SDO). Building upon the SDO algorithm, it introduces population initialization, mutation, and reverse learning strategies to improve the algorithm's performance. The algorithm's performance is verified using the CEC test function. Finally, it is applied to photoacoustic spectral feature screening. The process is as follows: Figure 1 As shown. Specifically, it includes the following steps:

[0052] S1: Acquire photoacoustic spectral data as training samples;

[0053] S2: The multi-strategy improved sled dog optimization algorithm (ISDO) is used to filter features of photoacoustic spectral data. The multi-strategy improvements include using Latin hypercube sampling for population initialization, using differential evolution strategy for population individual mutation update, and introducing a reverse learning strategy for population individual learning update.

[0054] S3: Train the photoacoustic spectral regression model using the selected characteristic wavelengths;

[0055] S4: Input the photoacoustic spectral data to be predicted into the trained photoacoustic spectral regression model for feature prediction.

[0056] In one embodiment, to verify the accuracy of the photoacoustic spectral regression model prediction after feature screening, in S1 the photoacoustic spectral data is divided into a training set and a test set in a 7:3 ratio. The training set is used to train the photoacoustic spectral regression model, and the test set is used to test the accuracy of the trained photoacoustic spectral regression model.

[0057] In one embodiment, the photoacoustic spectral data in S1 is preprocessed photoacoustic spectral data, and the preprocessing step includes preprocessing the original photoacoustic spectral data using median filtering:

[0058]

[0059] In the formula, S ij This is for selecting the center point; f(i,j) is the original photoacoustic spectral data; g(i,j) is the photoacoustic spectral data after median filtering.

[0060] In one embodiment, the multi-strategy improved sled dog optimization algorithm in S2 includes the following steps:

[0061] S21: Using the Latin Hypercube Sampling (LHS) method, the photoacoustic spectral data is divided into equal intervals along each dimension. Then, a sample value is randomly generated within each interval. The sample values ​​from different dimensions are then combined in a randomly arranged order to obtain a relatively dispersed initial solution set as the initial population. This makes the initial solutions more evenly distributed in the search space, avoiding clustering and greatly improving the diversity and spatial coverage of the initial population. This provides a high-quality search starting point for subsequent optimization processes and significantly enhances the algorithm's global exploration capability.

[0062] S22: During population mutation, differential evolution strategies are used to update the velocity and position of individuals in the current population by mutating individuals from other populations. During algorithm iteration, the mechanism of differential evolution is borrowed to mutate individuals, dynamically updating their velocity and position, increasing population perturbation and diversity, and effectively preventing premature convergence of the algorithm.

[0063] S23: When the algorithm gets stuck in a local optimum or the population diversity is insufficient, a reverse learning strategy is introduced to replace the current population individuals through reverse learning. This mechanism provides the algorithm with the ability to explore backwards, greatly enhancing its ability to escape the local optimum trap and continue searching for the global optimum.

[0064] In this embodiment, traditional random initialization of the sled dog optimization algorithm often results in an uneven distribution in the search space, leading to insufficient initial diversity. To overcome this problem and prevent premature convergence of the algorithm, as well as to address the need to broaden the search scope when the algorithm gets stuck in a local optimum or when the population diversity is insufficient, various strategies are employed to improve it. Therefore, S21 includes the following steps:

[0065] S211: Search space [lb] d ,ub d The d-th dimension is divided into N intervals, lb d ub is the lower bound of the d-th dimension search space. d Let be the upper limit of the search space in the d-th dimension;

[0066] S212: Randomly generate a sample point in each interval, such that each interval is taken only once in each dimension;

[0067] S213: Combine sample points from all dimensions to obtain N sample points as the initial population. The i-th individual in the initial population... Represented as:

[0068]

[0069] In the formula, u i,d Let π be an integer in the range {0, 1, ..., N-1}. i,d It is a random number in the range [0,1).

[0070] In this embodiment, to maintain population diversity and prevent premature convergence of the algorithm, a mutation operation can be added to the sled dog optimization algorithm. This can be directly adapted from strategies in differential evolution (DE). Therefore, S22 includes the following steps:

[0071] S221: Select individuals X from three different populations r X s X t Update the population individual X using the difference vector.i :

[0072] V i =X r +F·(X s -X t );

[0073] In the formula: X r X s X t Let x be the r-th individual in the population. r The position vector in the d-th dimension, and the s-th individual x in the population. s The position vector in the d-th dimension, and the t-th individual x in the population t The position vector in the d-th dimension; F is a scaling factor, typically between 0.5 and 1; V i This is the new vector obtained after mutation;

[0074] S222: Put this V i With individual X in the population i Perform crossover to obtain the test vector U i,d :

[0075]

[0076] In the formula, rand(0,1) is a random number generated from the uniform distribution (0,1); CR∈[0,1] is the crossover probability preset by the user according to their needs; k is a randomly selected dimension index, ensuring that at least one dimension comes from the mutation vector;

[0077] S223: Calculate the full-dimensional experimental vector U i fitness, U i That is, the trial vector of the i-th individual in the total dimension D, if it is better than X i Then use U i Replace X i .

[0078] It should be noted that the fitness calculation uses the sled dog optimization algorithm with the objective function fitness(X) i )=f(X i ), calculate U respectively i The objective function value and X i The objective function values ​​are used to select the best value as the current i-th individual in the population. Generally, the individual with the smallest objective function value is selected as the current i-th individual in the population.

[0079] In this embodiment, when the algorithm gets stuck in a local optimum or the population diversity is insufficient, it is necessary to broaden the search range again, which can be achieved by introducing a reverse learning strategy. Therefore, S23 includes the following steps:

[0080] S231: For the location X = (x1,...,x...) D Given the current population individual, determine its reverse population individual X′=(x′1,...,x′). D )for:

[0081] x′ d =lb d +ub d -x d ;

[0082] In the formula, lb d ub is the lower bound of the d-th dimension search space; d Let d be the upper limit of the d-th dimension search space, where d∈[1,D] and D is the total dimension;

[0083] S232: Calculate the fitness of X′. If it is better than X, replace X with X′.

[0084] In one embodiment, during the training process (S3), various parameters in the model are adjusted; in S4, the photoacoustic spectral regression model with adjusted parameters is used to input the photoacoustic spectral data to be predicted for prediction. In practice, evaluation metrics can be used to test the prediction performance.

[0085] like Figure 2 As shown, the specific execution steps of the sled dog optimization algorithm with multi-strategy improvement in S2 for feature selection are given:

[0086] S210: Population initialization is performed using the Latin hypercube algorithm, with the following parameter settings:

[0087] The maximum number of iterations for the population is T = 500.

[0088] The search space is [lb] d ,ub d ] = [-100, 100];

[0089] Dimension d = 30.

[0090] Determine if the maximum number of iterations (500) has been reached. If it is less than the maximum number of iterations, update all parameter values ​​and execute S230; if the maximum number of iterations has been reached, execute S240.

[0091] S220: Update individual velocity and position using differential mutation;

[0092] S230: Reverse learning replacement of some individuals;

[0093] S240: Input the selected characteristic wavelengths and input them into the regression model for prediction.

[0094] A second aspect of the present invention also discloses an electronic device, including a processor and a memory storing a computer program, characterized in that the processor executes the computer program to implement the steps of the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm of the first aspect of the present invention.

[0095] The third aspect of the present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm of the first aspect of the present invention.

[0096] The above provides a detailed description of the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm provided by the present invention. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

[0097] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm, characterized in that, Includes the following steps: S1: Acquire photoacoustic spectral data as training samples; S2: The photoacoustic spectral data is feature-selected using a multi-strategy improved sled dog optimization algorithm, wherein the multi-strategy improvement includes using Latin hypercube sampling for population initialization, using differential evolution strategy for population individual mutation update, and introducing a reverse learning strategy for population individual learning update. S3: Train the photoacoustic spectral regression model using the selected characteristic wavelengths; S4: Input the photoacoustic spectral data to be predicted into the trained photoacoustic spectral regression model for feature prediction.

2. The photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm according to claim 1, characterized in that, The photoacoustic spectral data in S1 is preprocessed photoacoustic spectral data. The preprocessing steps include using median filtering to preprocess the original photoacoustic spectral data. In the formula, S ij This is for selecting the center point; f(i,j) is the original photoacoustic spectral data; g(i,j) is the photoacoustic spectral data after median filtering.

3. The photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm according to claim 1, characterized in that, The multi-strategy improved sled dog optimization algorithm in S2 includes the following steps: S21: Using the Latin hypercube method, the photoacoustic spectral data is divided into equal intervals in each dimension. A value is randomly selected from each interval, and all values ​​are combined in a random order. The resulting initial solution set is used as the initial population. S22: During population mutation, use individuals from other populations to perform differential evolution strategy mutations to update the velocity and position of the current population individuals; S23: When the algorithm gets stuck in a local optimum or the population diversity is insufficient, a reverse learning strategy is introduced to perform reverse learning replacement of the current population individuals.

4. The photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm according to claim 3, characterized in that, S21 includes the following steps: S211: Search space [lb] d ,ub d The d-th dimension is divided into N intervals, lb d ub is the lower bound of the d-th dimension search space. d Let be the upper limit of the search space in the d-th dimension; S212: Randomly generate a sample point in each interval, such that each interval is taken only once in each dimension; S213: Combine sample points from all dimensions to obtain N sample points as the initial population. The i-th individual in the initial population... Represented as: In the formula, u i,d Let π be an integer in the range {0, 1, ..., N-1}. i,d It is a random number in the range [0,1).

5. The photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm according to claim 3, characterized in that, S22 includes the following steps: S221: Select individuals X from three different populations r X s X t Update the population individual X using the difference vector. i : V i =X r +F·(X s -X t ); In the formula: X r X s X t Let x be the r-th individual in the population. r The position vector in the d-th dimension, and the s-th individual x in the population. s The position vector in the d-th dimension, and the t-th individual x in the population t The position vector in the d-th dimension; F is a scaling factor, V i This is the new vector obtained after mutation; S222: Put this V i With individual X in the population i Perform crossover to obtain the test vector U i,d : In the formula, rand(0,1) is a random number generated from the uniform distribution (0,1); CR∈[0,1] is the preset crossover probability; k is a randomly selected dimension index; S223: Calculate the full-dimensional experimental vector U i If the fitness is better than X i Then use U i Replace X i .

6. The photoacoustic spectral feature screening method based on a multi-strategy sled dog optimization algorithm according to claim 3, characterized in that, S23 includes the following steps: S231: For the location X = (x1,...,x...) D Given the current population individuals, determine their reverse population individuals X′=(x1′,...,x′). D )for: x′ d =lb d +ub d -x d ; In the formula, lb d ub is the lower bound of the d-th dimension search space; d Let d be the upper limit of the d-th dimension search space, where d∈[1,D] and D is the total dimension; S232: Calculate the fitness of X′. If it is better than X, replace X with X′.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the photoacoustic spectral feature screening method based on the multi-strategy sled dog optimization algorithm as described in any one of claims 1 to 6.

Images