Fuzzy reasoning underwater acoustic target classification method adopting integrated learning
By combining ensemble learning and fuzzy reasoning, the accuracy and interpretability issues of underwater acoustic target recognition under limited data conditions were resolved, achieving efficient classification in complex underwater environments and reducing computational resource consumption.
Patent Information
- Application Number
- CN202511531771.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-24
- Publication Date
- 2026-02-03
AI Technical Summary
Existing underwater acoustic target recognition methods struggle to achieve accuracy and interpretability when processing limited data, and they consume significant computational resources, making them unsuitable for complex and ever-changing underwater environments.
An ensemble learning-based fuzzy inference method is adopted. The feature extraction module obtains the Mel-Cepstral Coefficient and Gamma-Tony Cepstral Coefficient features, and combines them with T cascaded fuzzy inference function blocks for underwater acoustic target classification. The training and recognition modules of the fuzzy inference function blocks are used for data classification, thereby improving the robustness and adaptability of the model.
Accurate classification of underwater acoustic targets was achieved with limited data, reducing computational complexity, enhancing generalization performance on unseen data, and improving model interpretability.
Smart Images
Figure CN121456701A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the field of marine underwater target recognition and underwater acoustic target recognition, and relates to a fuzzy reasoning underwater acoustic target classification method using ensemble learning. BACKGROUND
[0002] The field of underwater acoustics has long been a key area of research, especially passive acoustic target recognition technology, which plays a key role in applications such as marine surveillance, underwater exploration, and environmental monitoring. Despite advances in acoustic technology, accurate and efficient recognition of underwater targets remains a challenging task due to the complexity of the underwater acoustic environment, including noise, reverberation, and variability of target characteristics.
[0003] In recent years, there have been numerous studies on automatic recognition methods for underwater acoustic targets. Existing automatic recognition methods can be broadly divided into two categories: statistical-based recognition and learning-based recognition. For statistical-based recognition methods, a large amount of prior knowledge and manually extracted features are required to establish a model. These statistical-based methods have the following characteristics: (1) a large amount of prior knowledge is required to guide feature extraction and recognition method design. (2) When dealing with large-scale data sets, the performance of these methods may decrease and fail to achieve sufficient accuracy, mainly limited by their representation and generalization capabilities. (3) Underwater acoustic target recognition methods rely heavily on manual feature extraction and rule-based classifiers, which are often limited by their inability to adapt to complex and variable underwater environments.
[0004] Learning-based recognition mainly refers to recognition methods using deep learning. Deep learning techniques are highly regarded for their ability to automatically learn hierarchical representations from raw data or specific feature domains. These technological advances have broken the boundaries of acoustic target recognition, achieving new benchmarks in accuracy and robustness. However, deep learning models often require a large amount of labeled data, which is often difficult to obtain in practical applications. In addition, the interpretability of deep learning models remains a challenge, making it difficult to understand and trust their prediction results in critical applications. More importantly, due to the need to find the optimal solution in a vast and complex solution space, these methods require a large amount of computational resources. SUMMARY
[0005] Current research on underwater acoustic target recognition has involved the field of small sample learning, but few research results have been produced on the interpretability of underwater acoustic target classification models. To address the extreme scarcity of existing underwater acoustic target recognition data, there is an urgent need to develop a method that can be trained with a small amount of data while balancing accuracy and interpretability. The technical solution adopted by the present application is: a fuzzy reasoning underwater acoustic target classification method using ensemble learning, comprising the following steps: Obtain underwater acoustic data; Based on underwater acoustic datasets, an underwater acoustic target classification model is constructed. The underwater acoustic target classification model includes a feature extraction module: used to extract and fuse Mel-frequency cepstral coefficient features and gamma-frequency cepstral coefficient features in the underwater acoustic dataset; Integrated classifier: Based on the fused features output by the feature extraction module, it corrects errors in the classification process of underwater acoustic data and achieves the target classification of underwater acoustic data; The underwater acoustic data is input into the underwater acoustic target classification model to classify the underwater acoustic targets.
[0006] Furthermore, the integrated classifier comprises T cascaded fuzzy inference function blocks.
[0007] Furthermore: the fuzzy inference function block includes: Training module: Used to train the sample feature set obtained from feature extraction and build a data prototype for fuzzy inference; Recognition Module: Used to identify and predict the data prototypes generated by the training module for fuzzy inference, and output the category of the samples in the data.
[0008] Furthermore, the process of training the sample feature set obtained from feature extraction to construct the data prototype for fuzzy inference is as follows: Based on dataset D To calculate unique samples for various sample types, define... It is the j-th unique sample of the i-th class; Calculate the data density of the sample; Based on the data density of the samples, establish sample micro-cluster sets; The sample micro-clusters are used to generate the data prototype for fuzzy inference.
[0009] Furthermore, the specific process of establishing the sample micro-cluster set based on the sample data density is as follows: Based on the data density of the samples, the unique samples of each class are sorted, and a sequence is formed for the samples of the i-th class. ; For sequence Calculate the location of local density maxima; Based on sequence Find the local density maxima and create a set of microclusters; Determine the radius of influence of the micro-cluster set; Based on the effective radius of the micro-cluster set, redundant micro-clusters are merged to obtain the center of the merged micro-cluster.
[0010] Further: the pair of sequences The process of calculating the location of local density maxima is as follows: Calculate the sequence using the second derivative. Local maxima of density
[0011] In the formula: Represents a sequence The calculated second derivative at position j The value; It is the first i The unique set of samples corresponding to the local density maxima of a class of samples. express The potential of a set.
[0012] Furthermore, the process of identifying and predicting the categories of samples in the data based on the fuzzy inference data prototype generated by the training module is as follows: When performing identification, for the sample to be identified... The prediction function is as follows:
[0013] In the formula: This represents the score vector of the fuzzy inference function block. For samples The category prediction output, It is a collection of microclusters.
[0014] Furthermore, the process of improving the performance of the ensemble classifier is as follows: The weights of the samples in the T cascaded fuzzy inference function blocks are determined as follows; In the serial structure of the fuzzy inference function block, for the first... t The error rate for each functional block is calculated as follows:
[0015] In the formula: Representing the t A fuzzy inference function block for samples The prediction results Indicates sample The weights of the samples are calculated using the following formula:
[0016] when When, the weights of all samples are expressed as , For dataset D Total number of samples; Indicates the first tThe weight of the fuzzy inference function block, and the weight of the fuzzy inference function block. t Each functional block is represented as , It is a dataset D The new dataset is generated based on weighted resampling; It is the first In the score vector output by the fuzzy inference function block, at the th The score value on the class; It is the first In the score vector output by each fuzzy inference function block, The minimum score, i.e. the minimum distance, for classes other than those in the previous class. The weights of the T cascaded fuzzy inference function blocks are determined as follows: For the t The fuzzy inference function block, the first t Weights of each fuzzy inference function block The calculation expression is as follows:
[0017] For the t A fuzzy inference function block is used to obtain the subvector. The calculation is as follows:
[0018] In the formula: It is a sample With the place The minimum distance between the prototypes of classes other than the one specified.
[0019] Furthermore, the overall prediction process of the ensemble classifier is as follows: For the t A fuzzy inference function block, when processing samples At that time, a representation vector is generated. ,in: It is the first k The predicted representation value of the class is calculated as follows:
[0020] Where: K is the type of the sample set; based on T Each fuzzy inference function block generates the final representation vector. Its calculation is as follows:
[0021] The final prediction output is generated by using the obtained representation vector for the sample to be identified. For vectors KFind the maximum value among the given values, as follows:
[0022] In the formula: For the sample to be tested The final category prediction output.
[0023] This invention provides a fuzzy inference underwater acoustic target classification method using ensemble learning. By combining fuzzy logic with ensemble learning, this invention offers a more robust, adaptive, and data-efficient solution for underwater acoustic target recognition. The fusion of fuzzy inference system and ensemble learning enhances the underwater acoustic target classification model's ability to handle noise, improves its generalization performance on unseen data, and reduces computational complexity. Attached Figure Description
[0024] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0025] Figure 1 This is a flowchart of the method; Figure 2 This is a flowchart of the feature extraction process; Figure 3 This is a structural diagram of the fuzzy inference function block.
[0026] Figure 4 The experimental validation graphs on the dataset include (a) comparisons of the proposed method with other methods on CCR, (b) comparisons of the proposed method with other methods on MWCCR, (c) comparisons of the proposed method with other methods on Gmean, (d) comparisons of the proposed method with other methods on F1, and (e) comparisons of the proposed method with other methods on Kappa. Detailed Implementation
[0027] It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and embodiments.
[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. 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.
[0029] Figure 1 This is a flowchart of the method; A fuzzy inference underwater acoustic target classification method employing ensemble learning includes the following steps: S1: Acquire underwater acoustic data; S2: Construct an underwater acoustic target classification model based on an underwater acoustic dataset; The underwater acoustic target classification model includes a feature extraction module: used to extract and fuse Mel-frequency cepstral coefficient features and gamma-frequency cepstral coefficient features in the underwater acoustic dataset; Integrated classifier: Based on the fused features output by the feature extraction module, it corrects errors in the classification process of underwater acoustic data and achieves the target classification of underwater acoustic data; S3: Input the underwater acoustic data into the underwater acoustic target classification model to classify the underwater acoustic targets.
[0030] When the underwater acoustic dataset used is the type of ship, the underwater ship type can be classified. When the underwater acoustic dataset used is the type of fish, the classification of underwater fish types can be achieved; That is, the classification of underwater acoustic targets is determined by the type of underwater acoustic dataset; Steps S1 / S2 / S3 are executed sequentially; The integrated classifier comprises T cascaded fuzzy inference function blocks.
[0031] The fuzzy inference function block includes: Training module: Used to train the sample feature set obtained from feature extraction and build a data prototype for fuzzy inference; Recognition Module: Used to identify and predict the data prototypes generated by the training module for fuzzy inference, and output the category of the samples in the data.
[0032] Furthermore, the process of training the sample feature set obtained from feature extraction to construct the data prototype for fuzzy inference is as follows: S221: Based on datasetD To calculate unique samples for various sample types, define... It is the j-th unique sample of the i-th class; S222: Calculate the data density of the sample; S223: Based on the data density of the samples, establish a set of sample micro-clusters; S224: Sample micro-cluster set, generating data prototypes for fuzzy inference.
[0033] Furthermore, the specific process of establishing the sample micro-cluster set based on the sample data density is as follows: S2231: Based on the data density of the samples, sort the unique samples of each class and form a sequence for the samples of the i-th class. ; S2232: For the sequence Calculate the location of local density maxima; S2233: Sequence-based Find the local density maxima and create a set of microclusters; S2234: Determine the radius of influence of the micro-cluster set; S2235: Based on the effective radius of the micro-cluster set, the redundant micro-clusters are merged to obtain the center of the merged micro-cluster.
[0034] Furthermore, the sequence The process of calculating the location of local density maxima is as follows: Calculate the sequence using the second derivative. Local maxima of density
[0035] In the formula: Represents a sequence The calculated second derivative at position j The value; It is the first i The unique set of samples corresponding to the local density maxima of a class of samples. express The potential of a set.
[0036] Furthermore, the process of identifying and predicting the categories of samples in the data based on the fuzzy inference data transmitted by the training module is as follows: When performing identification, for the sample to be identified... The prediction function is as follows:
[0037] In the formula: This represents the score vector of the fuzzy inference function block. For samples The category prediction output, It is a collection of microclusters.
[0038] Furthermore, the process of improving the performance of the ensemble classifier is as follows: S2241: Determine the weights of the samples in the T cascaded fuzzy inference function blocks, the specific process is as follows; In the serial structure of the fuzzy inference function block, for the first... t The error rate for each functional block is calculated as follows:
[0039] In the formula: Representing the t A fuzzy inference function block for samples The prediction results Indicates sample The weights of the samples are calculated using the following formula:
[0040] when When, the weights of all samples are expressed as , For dataset D Total number of samples; Indicates the first t The weight of the fuzzy inference function block, and the weight of the fuzzy inference function block. t Each functional block is represented as , It is a dataset D The new dataset is generated based on the weighted resampling. It is the first In the score vector output by the fuzzy inference function block, at the th The score value on the class; It is the first In the score vector output by each fuzzy inference function block, Minimum score in classes other than those in the above categories S2242: Determine the weights of the T cascaded fuzzy inference function blocks, the specific process of which is as follows: For the t The fuzzy inference function block, the first t Weights of each fuzzy inference function block The calculation expression is as follows:
[0041] For the t A fuzzy inference function block is used to obtain the subvector. The calculation is as follows:
[0042] In the formula: It is a sample With the place The minimum distance between the prototypes of classes other than the one specified.
[0043] Furthermore, the overall prediction process of the ensemble classifier is as follows: For the t A fuzzy inference function block, when processing samples At that time, a representation vector is generated. ,in: It is the first k The predicted representation value of the class is calculated as follows:
[0044] Where: K is the type of the sample set; based on T Each fuzzy inference function block generates the final representation vector. Its calculation is as follows:
[0045] The final prediction output is generated by using the obtained representation vector for the sample to be identified. Find the maximum value among the K values in the vector, as follows:
[0046] In the formula: For the sample to be tested The final category prediction output.
[0047] Example 1 A fuzzy inference underwater acoustic target classification method employing ensemble learning includes the following steps: S1: Acquire underwater acoustic data; S2: Construct an underwater acoustic target classification model based on an underwater acoustic dataset; The underwater acoustic target classification model includes a feature extraction module: used to extract and fuse Mel-frequency cepstral coefficient features and gamma-frequency cepstral coefficient features in the underwater acoustic dataset; Integrated classifier: Based on the fused features output by the feature extraction module, it corrects errors in the classification process of underwater acoustic data and achieves the target classification of underwater acoustic data; S3: Input the underwater acoustic data into the underwater acoustic target classification model to classify the underwater acoustic targets.
[0048] Figure 2This is a flowchart of the feature extraction process; Furthermore, the process of extracting and fusing the Mel-frequency cepstral coefficient features and the gamma-frequency cepstral coefficient features from the underwater acoustic dataset is as follows: S2.1: Divide the underwater acoustic data into 3-second segments; S2.2 For each data segment Perform short-time discrete Fourier transform;
[0049] In the formula: yes Data after windowing; It is a Hanning window; It's a long window. .
[0050] S2.3: Calculate the Mel-Cepstral Coefficients after Short-Time Discrete Fourier Transform ;
[0051] In the formula: This represents the Mel cepstral notation; It is the short-time Fourier transform of the data segment; This indicates the number of filters in a Mel filter bank; Indicates to Performed discrete cosine transform; It is the length of the output result of the discrete cosine transform. The Mel filter is calculated as follows:
[0052] In the formula This represents the center frequency of the m-th filter. ; In this invention The filter's center frequency range is 100. Hz up to 6 kHz ; Therefore, for each frame of data, It is a 30-dimensional feature vector.
[0053] S2.4: Calculate the Gammatone Frequency Cepstral Coefficients (GTCC) after short-time discrete Fourier transform;
[0054] In the formula: These are the Gammatoni cepstral coefficients; Indicates to The discrete cosine transform was performed. It is the length of the output result of the discrete cosine transform; This is a reverse spectral of Gammatoni. The time-domain representation of a gamma-toni filter (frequency domain) is as follows:
[0055] In the formula: Represents output gain; represent The bandwidth can be calculated from the equivalent rectangular bandwidth (ERB); represent The order of, here =3; Represents the total number of gamma-toni filters; express The center frequency; when The center frequency range is 50. Hz Up to 4500 Hz, and Therefore, for each frame of data, It is also a 30-dimensional feature vector. S2.5 combines the two features, Mel-Cepstral Coefficient and Gammatoni Cepstral Coefficient; Data of each frame and Then, the features are fused (connected front and back ends) to form a 60-dimensional feature vector, as follows:
[0056] In the formula: Represents a vector connection.
[0057] Furthermore, a fuzzy inference function block (Block of ZOFIS) is constructed. The fuzzy inference function block includes: Training module: Used to train the sample feature set obtained from feature extraction and build a data prototype for fuzzy inference; Recognition Module: Used to identify and predict the data prototypes for fuzzy inference based on the training module, and output the category of the samples in the data.
[0058] The fuzzy inference function block is a basic functional component of the subsequent data processing logic, used... or The principle is explained in the following text. Figure 3 ; Figure 3 middle DThis refers to the set of sample features obtained from the feature extraction in the previous step (hereinafter referred to as the sample set). ,in Indicates the first i One sample, yes Category labels; or they can also be represented as Include K Class of samples, where Indicates belonging to the first i The set of all samples of the class ; Indicates the first i The class of j One sample; Indicates the first i The class of j A prototype; for those with K A collection of class samples, Indicates the first i The total number of class prototypes, ; Indicates the need to find samples With prototype The similarity between them; It is an operation to find the category with the maximum similarity.
[0059] Furthermore, the process of training the sample feature set obtained from feature extraction to construct the data prototype for fuzzy inference is as follows: S221: For the training phase dataset D To calculate a unique sample for each type of sample, define: For the first i The class of j One unique sample; S222: Calculate the density of class sample data ;
[0060] In the formula: for The number of times it appears in the dataset. It is the first i A unique set of samples for each class; Number of samples in the middle.
[0061] S222: Establish a micro-cluster set, the specific process is as follows: S2221: Sort the unique samples of each category; Using the density calculated above, sort the unique samples of each class. For the i-th i Classes form sequences , The process is as follows:
[0062] S2222: For the sequence Calculate the location of local maxima of density; Calculate the sequence using the second derivative. The location of the local maximum density;
[0063] In the formula: Represents a sequence The calculated second derivative at position j The numerical value. It is the first i The unique set of samples corresponding to the local density maxima of a class of samples. express The potential of a set.
[0064] S2223: Creating a collection of micro-clusters ; For dataset D The Middle i Class Sample ,and By comparing samples of local density maxima, micro-clusters can be formed. The calculation process is as follows:
[0065] In the formula: To be consistent with the sample Recent local density maxima samples The serial number; Representing the i The first in the class j Microclusters; Represents a set of micro-clusters The potential (number of elements); It is a microcluster The center.
[0066] S2224: Determine the radius of influence of the micro-cluster set After each sample in D is bound to a micro-cluster in the previous step, the effective range (or effective radius) of each cluster can be determined by the granularity parameter. The following iterative method was used to calculate:
[0067] In this invention ; S2225: Fusion of redundant microclusters; To eliminate overlapping microclusters, the microcluster fusion method can be calculated using the following formula:
[0068] S223: Generate the final prototype; The merged micro-cluster centers are used as the prototype output of the fuzzy system:
[0069] Furthermore, the process of improving the performance of the ensemble classifier is as follows: S2241: Sample weighting, determining the weights of samples in the T cascaded fuzzy inference function blocks, the specific process is as follows; Serial training is required T A fuzzy inference function block These functional blocks are connected sequentially as a whole. This serial working structure gradually improves the overall recognition rate by correcting errors in the previous module through subsequent functional blocks (focusing on learning). In the serial structure of fuzzy inference functional blocks, for the first... t The error rate for each functional block is calculated as follows:
[0070] In the formula: Representing the t A fuzzy inference function block for samples The prediction results Indicates sample The weights of the samples are calculated using the following formula:
[0071] For the sake of simplicity, when When, the weights of all samples are expressed as , For dataset D Total number of samples; Indicates the first t The weight of the fuzzy inference function block, and the weight of the fuzzy inference function block. t Each functional block is represented as , It is a dataset D The new dataset is generated based on the weighted resampling. It is the first In the score vector output by the fuzzy inference function block, at the th The score value on the class; It is the first In the score vector output by each fuzzy inference function block, Minimum score in classes other than those in the above categories S2242: Determine the weights of the T cascaded fuzzy inference function blocks, the specific process of which is as follows: For thet The fuzzy inference function block, the first t Weights of each fuzzy inference function block The calculation expression is as follows:
[0072] For the t A fuzzy inference function block is used to obtain the subvector. The calculation is as follows:
[0073] In the formula: It is a sample With the place The minimum distance between the prototypes of classes other than the one specified.
[0074] Furthermore, the overall prediction process of the ensemble classifier is as follows: For the t A fuzzy inference function block, when processing samples At that time, a representation vector is generated. ,in: It is the first k The predicted representation value of the class is calculated as follows:
[0075] Where: K is the type of the sample set; based on T Each fuzzy inference function block generates the final representation vector. Its calculation is as follows:
[0076] The final prediction output is generated by using the obtained representation vector for the sample to be identified. For vectors K Find the maximum value among the given values, as follows: (32) In the formula: For the sample to be tested The final category prediction output.
[0077] This method integrates learning-based fuzzy inference underwater acoustic target classification and was tested on two underwater target datasets: DeepShip and ShipsEar.
[0078] The performance measures used in the experiment are as follows: (1)CCR: correct classification rate (33) (2)MWCCR: mean within-group correct classification rate (34) in: In the category Correct recognition rate (CCR) on the screen; It represents the total number of categories in the dataset.
[0079] (3) G-mean (35) (4) Macro-F1 (36) (37) in: This refers to the category On the precision; Refers to category Recall rate. (5) Kappa fraction , (38) (39) (40) Split the ShipsEar dataset The experimental data was divided into two non-overlapping parts: a training set and a test set. The ratio of the training set data to the total data was calculated for each test. The dataset is configured according to the elements in the set {1 / 20, 1 / 15, 1 / 10, 1 / 5, 1 / 4, 1 / 3, 1 / 2, 2 / 3, 4 / 5, 9 / 10}; the corresponding test set consists of data from the entire dataset excluding the training set. Experimental results on the dataset are shown below. Figure 4 Among them, (a) the comparison of the method of this application with other methods on CCR, (b) the comparison of the method of this application with other methods on MWCCR, (c) the comparison of the method of this application with other methods on Gmean, (d) the comparison of the method of this application with other methods on F1, and (e) the comparison of the method of this application with other methods on Kappa.
[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A fuzzy inference underwater acoustic target classification method employing ensemble learning, characterized in that: Includes the following steps: Acquire underwater acoustic data; Based on underwater acoustic datasets, an underwater acoustic target classification model is constructed. The underwater acoustic target classification model includes a feature extraction module: used to extract and fuse Mel-frequency cepstral coefficient features and gamma-frequency cepstral coefficient features in the underwater acoustic dataset; Integrated classifier: Based on the fused features output by the feature extraction module, it corrects errors in the classification process of underwater acoustic data and achieves the target classification of underwater acoustic data; The underwater acoustic data is input into the underwater acoustic target classification model to classify the underwater acoustic targets.
2. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 1, characterized in that: The integrated classifier comprises T cascaded fuzzy inference function blocks.
3. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 2, characterized in that: The fuzzy inference function block includes: Training module: Used to train the sample feature set obtained from feature extraction and build a data prototype for fuzzy inference; Recognition Module: Used to identify and predict the data prototypes generated by the training module for fuzzy inference, and output the category of the samples in the data.
4. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 3, characterized in that: The process of training the sample feature set obtained from feature extraction to construct the data prototype for fuzzy inference is as follows: Based on dataset D To calculate a unique sample for each type of sample, define: It is the j-th unique sample of the i-th class; Calculate the data density of the sample; Based on the data density of the samples, establish sample micro-cluster sets; The sample micro-clusters are used to generate the data prototype for fuzzy inference.
5. The underwater acoustic target classification method using fuzzy inference with ensemble learning according to claim 4, characterized in that: The specific process of establishing a sample micro-cluster set based on sample data density is as follows: Based on the data density of the samples, the unique samples of each class are sorted, and a sequence is formed for the samples of the i-th class. ; For sequence Calculate the location of local density maxima; Based on sequence Find the local density maxima and create a set of microclusters; Determine the radius of influence of the micro-cluster set; Based on the effective radius of the micro-cluster set, redundant micro-clusters are merged to obtain the center of the merged micro-cluster.
6. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 5, characterized in that: The pair of sequences The process of calculating the location of local density maxima is as follows: Calculate the sequence using the second derivative. Local maxima of density In the formula: Represents a sequence The calculated second derivative at position j The value; It is the first i The unique set of samples corresponding to the local density maxima of a class of samples. express The potential of a set.
7. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 3, characterized in that: The process of identifying and predicting the categories of samples in the data based on the fuzzy inference data prototype generated by the training module is as follows: When performing identification, for the sample to be identified... The prediction function is as follows: In the formula: This represents the score vector of the fuzzy inference function block. For samples The category prediction output, It is a collection of microclusters.
8. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 1, characterized in that: The process of improving the performance of the ensemble classifier is as follows: The weights of the samples in the T cascaded fuzzy inference function blocks are determined as follows; In the serial structure of the fuzzy inference function block, for the first... t The error rate for each functional block is calculated as follows: In the formula: Representing the t A fuzzy inference function block for samples The prediction results Indicates sample The weights of the samples are calculated using the following formula: when When, the weights of all samples are expressed as , For dataset D Total number of samples; Indicates the first t The weight of the fuzzy inference function block, and the weight of the fuzzy inference function block. t Each functional block is represented as , It is a dataset D The new dataset is generated based on the weighted resampling. It is the first In the score vector output by the fuzzy inference function block, at the th The score value on the class; It is the first In the score vector output by each fuzzy inference function block, The minimum score, i.e. the minimum distance, for classes other than those in the previous class. The weights of the T cascaded fuzzy inference function blocks are determined as follows: For the t The fuzzy inference function block, the first t Weights of each fuzzy inference function block The calculation expression is as follows: For the t A fuzzy inference function block is used to obtain the subvector. The calculation is as follows: In the formula: It is a sample With the place The minimum distance between the prototypes of classes other than the one specified.
9. The underwater acoustic target classification method using fuzzy reasoning with ensemble learning according to claim 1, characterized in that: The overall prediction process of the ensemble classifier is as follows: For the t A fuzzy inference function block, when processing samples At that time, a representation vector is generated. ,in: It is the first k The predicted representation value of the class is calculated as follows: Where: K is the type of the sample set; based on T Each fuzzy inference function block generates the final representation vector. Its calculation is as follows: The final prediction output is generated by using the obtained representation vector for the sample to be identified. For vectors K Find the maximum value among the given values, as follows: （32） In the formula: For the sample to be tested The final category prediction output.