A fuzzy monotonic correlation image recognition and machine learning method

Abstract

The application discloses a new correlation image recognition and machine learning method based on fuzzy monotony, and belongs to the technical field of artificial intelligence of pattern recognition and machine learning; the method is defined as FMMCA, which evaluates local fuzzy monotone correlation by comparing row vectors and column vectors of an image matrix pair by pair, then the local correlations are weighted and accumulated, and finally the global fuzzy monotone correlation between images is obtained. The application directly uses fuzzy monotone correlation analysis to replace classical correlation analysis for multi-view research, so that the problems existing in classical correlation analysis do not exist, and the fuzzy monotone method feature does not need to be measured statically by distance, but can be measured dynamically by interval change, so that the influence of some noise is offset, the performance is improved, a new fuzzy monotone machine learning method is formed, the optimization of a focus loss function is not needed, the parameters are few, the robustness is good, and the computing power is small.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically pattern recognition and machine learning, and more particularly to a method for fuzzy, monotonic correlation image recognition and machine learning. Background Technology

[0002] Canonical Correlation Analysis (CCA), a classic method for multi-view correlation analysis, has received widespread attention since its introduction by H. Hotelling in 1935. The goal of CCA is to extract the best predictor from each set by maximizing the correlation coefficient between two sets. To achieve this, Hotelling further proposed a symmetric solution, defining the sequences of variable pairs as canonical variables, and the correlation between them as canonical correlation. By projecting complex high-dimensional variable sets into a desired low-dimensional common hidden subspace, CCA simplifies the statistical analysis of variables in two sets while extracting the maximum correlation between them. In recent years, CCA has become a hot research topic and has been widely applied in multiple disciplines, including computer vision, biomedicine, natural language processing, genetics, knowledge, and data engineering. However, despite its crucial role in multi-view correlation analysis and its key applications in research areas such as computer vision and pattern recognition, CCA still has some limitations. In particular, CCA relies on the Pearson correlation coefficient as a measure of correlation, which may not adequately capture nonlinear relationships between data when analyzing correlations. Furthermore, CCA's performance can be significantly impacted when the dataset contains noise, as noisy data may distort or mask the true correlation patterns. Therefore, CCA's effectiveness in handling nonlinear and noisy data may be limited, requiring further research and improvements to overcome these limitations.

[0003] Over the decades, numerous CCA-related algorithms have been proposed. For example, Kernel Canonical Correlation Analysis (KCCA), an extension of CCA, can discover nonlinear projections with maximum correlation, but is limited to reproducing kernel Hilbert spaces with corresponding kernels. Through kernel tricks, KCCA maps nonlinear data to a higher (even infinite) dimensional Hilbert feature space, where the data exhibits strong linearity. In 2014, Randomized Nonlinear Canonical Correlation Analysis (RCCA) was proposed. This research suggests using randomization to construct features that help reveal nonlinear patterns in data. For fundamental tasks such as regression or classification, using nonlinear random features offers almost no performance loss compared to exact kernel methods and effectively reduces computational complexity. In recent years, Deep Neural Networks have become a research hotspot. These networks have attracted widespread attention due to their outstanding performance in various tasks. A significant advantage of deep neural networks is their ability to learn complex nonlinear representations of data, which makes them demonstrate great potential in many fields such as image recognition, speech processing, and natural language processing. Furthermore, deep neural networks do not rely on traditional nonparametric models, meaning they can automatically adjust parameters to adapt to different datasets and task requirements. Given these advantages, many researchers have begun exploring methods combining deep learning with Canonical Correlation Analysis (CCA). This combination leverages the automatic feature extraction capabilities of deep neural networks and the correlation analysis capabilities of CCA to learn deeper, richer representations of data. For example, Andrew et al. proposed Deep Canonical Correlation Analysis (DCCA), which uses a deep neural network to mine more complex nonlinear relationships between two views. Wang et al. proposed Deep Canonical Correlation Autoencoder (DCCAE), a model consisting of two autoencoders that minimizes the reconstruction error of the autoencoders within the CCA framework. Benton et al. proposed Deep Generalized Canonical Correlation Analysis (DGCCA) and applied it to three tasks: speech transcription, articulation measurement, and information recommendation. Chandar et al. proposed CorrNet, which further couples cross-view reconstruction errors, thus providing an accurate reconstruction of one view given another.Yang et al. proposed a Convolutional Canonical Correlation Analysis (ConvCCA) based on convolutional neural networks. This method improves computational efficiency and feature extraction accuracy by optimizing the filter bank learning process through CCA, while significantly reducing information redundancy. Besides research combining CCA with deep networks, many researchers have also conducted a series of innovative studies based on CCA. For example, Lu et al. proposed two canonical correlation learning methods for image representation: robust canonical correlation analysis (robust-CCA) and low-rank representation canonical correlation analysis (LRR-CCA). Robust-CCA removes noise using low-order learning and extracts relevant features from a noise-free data matrix, while LRR-CCA introduces a low-rank representation to ensure the relevance of the extracted features. Subramanian et al. used sparse canonical correlation analysis (SCA) for multimodal data fusion to predict the survival rate of breast cancer patients. Kamlaskar et al. proposed an efficient feature fusion method combining CCA and support vector machine (SVM) classification. This method integrates features from different modalities to generate a more discriminative and compact feature representation, which can then be used in classification tasks.

[0004] In terms of machine learning methods, most current machine learning methods are mainly based on some loss functions for training. Many of these loss functions use distance as a key parameter. The parameters of loss functions are becoming increasingly complex, especially in deep learning methods, which have poor interpretability and increasingly complex parameters. Therefore, it is becoming increasingly important to find machine learning methods with good interpretability and fewer training parameters.

[0005] In summary, CCA-based algorithms play a crucial role in multi-view correlation analysis, but their limitations in handling nonlinear relationships and noisy data cannot be ignored. While CCA performs well in uncovering linear correlations, it is primarily based on the Pearson correlation coefficient, which is not always effective in identifying nonlinear correlations between two variables. Furthermore, CCA relies on Euclidean distance in its calculations, making it difficult to ignore the impact of noise on the original data. For variants of CCA, although they can handle nonlinear data, a nonlinear processing module is typically added before the CCA module to convert the nonlinear data into linear data before correlation analysis, resulting in a complex process. Moreover, the nonlinear processing module's ability to extract nonlinear features is insufficient, which may affect the subsequent image classification performance. Additionally, the widespread use of traditional classifiers such as SVM in image classification tasks may also limit the improvement of image classification performance.

[0006] To address the aforementioned problems, this invention proposes a fuzzy, monotonic correlation image recognition and machine learning method. Summary of the Invention

[0007] The purpose of this invention is to propose a fuzzy, monotonic correlation image recognition and machine learning method to solve the problems mentioned in the background art. This invention can be directly applied to nonlinear correlation analysis, thereby reducing some complex nonlinear transformation operations. Compared with some mainstream machine learning methods, it does not employ loss functions or deep learning machine training methods.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A fuzzy monotone correlation image recognition and machine learning method is defined as FMMCA (fuzzy monotone method correlation analysis). This method evaluates local fuzzy monotone correlation by comparing row and column vectors of the image matrix pairwise. The obtained local fuzzy monotone correlations are then weighted and accumulated to obtain the global fuzzy monotone correlation between images. The method specifically includes the following steps:

[0010] S1, Given two datasets and ,in N Represents the sample size of the image. yes X The i One sample image, yes Y The j1 ≤ 1 sample images i ≤ N ,1≤ j ≤ N ;Will X i and Y j They are respectively denoted as matrices A m×n sum matrix B m×n ;

[0011] S2, Extraction X i and Y j The correlation characteristics between them, from the matrix A m×n sum matrix B m×n The corresponding column vectors and row vectors are extracted separately for correlation calculation, and the local correlation results are integrated by weighted summation.

[0012] S3. Build a new image framework to achieve the image classification task by comparing the correlation strength between sample images.

[0013] Preferably, S2 specifically includes the following:

[0014] S2.1, By traversing the matrix A m×n sum matrix B m×n For all row vectors, calculate the local correlation along the row dimension; its mathematical expression is:

[0015]

[0016] In the formula, A T express A The transpose of the matrix, express A T The i List; B T express B The transpose of the matrix, express B T The i List; m Indicates the height of the sample image; k 1 indicates the number of intervals. p 1 represents the number of elements in the interval. ; k 1 and pThe relationship between 1 and 2 is shown below:

[0017]

[0018] in, n Indicates the width of the sample image, when p When 1=1, Classified as n There are intervals, each containing only one element;

[0019] parameter t 1 is k The optimal upper bound of 1 is determined by iterative optimization, gradually adjusting the parameters. t 1. By evaluating the image classification accuracy after each adjustment, an optimal value is ultimately found. t 1 value;

[0020] S2.2, By traversing the matrix A m×n sum matrix B m×n For all column vectors, calculate the local correlation in another dimension, as shown in the following mathematical expression:

[0021]

[0022] In the formula, A j express A The j List; B j express B The j List; k 2 represents the number of intervals. p 2 represents the number of elements in the interval. , k 2 and p The relationship between 2 is shown below:

[0023]

[0024] S2.3, and The overall correlation between them is obtained by combining the row and column dimensions, as shown in the following expression:

[0025]

[0026] in, The higher the value, the better. and The stronger the fuzzy monotonic relationship between two samples, the stronger the correlation between them.

[0027] S2.4 During machine learning training, an optimal [classification accuracy] is trained. t 1 and t Combinations of 2 are collectively referred to as t ;set up t 1 and t 2. Unification equals a certain t The training process involves identifying the interval divisions that achieve the highest classification accuracy. t As a feature of machine training.

[0028] Preferably, S3 specifically includes the following:

[0029] Assuming sample image Y r The image category is known, and Find the match through iterative updates X i Most relevant Y r Furthermore X i and Y r Images are grouped into the same image category; the mathematical formula for completing the image classification task is as follows:

[0030]

[0031] In the formula, r Indicates the number of iterations.

[0032] The present invention further protects a computer device, the computer device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the instruction, program, code set or instruction set being loaded and executed by the processor to implement the above-mentioned fuzzy monotonic correlation image recognition and machine learning method.

[0033] The present invention further protects a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the instruction, program, code set, or instruction set is loaded and executed by a processor to implement the above-mentioned fuzzy monotonic correlation image recognition and machine learning method.

[0034] Compared with existing technologies, this invention provides a fuzzy and monotonic correlation image recognition and machine learning method, which has the following beneficial effects:

[0035] (1) This invention can directly analyze and process nonlinear data without preprocessing or data conversion. It can directly analyze the correlation between images and identify the image category based on similarity. This invention mainly adopts a fuzzy monotonic correlation analysis method, which can directly perform nonlinear correlation analysis. Then, it performs an overall correlation analysis on the vector details between images through row and column dimensions to obtain the similarity between images. In addition, this fuzzy monotonic method does not require static measurement using distance. It can dynamically measure through interval changes, thereby offsetting and reducing some noise effects and improving performance.

[0036] (2) In terms of machine learning, the present invention is a machine learning method with fewer training parameters and strong interpretability. It does not use loss function as the main training method for machine learning. The reason for fewer training parameters is mainly based on the new features of the fuzzy monotonic correlation analysis method. The interval division directly affects the accuracy of image classification, so only one interval division parameter needs to be set for training. At present, other machine learning training methods based on loss function require more parameters. Deep learning has even more parameters, complex structure, and poor interpretability. Fuzzy monotonic correlation analysis is based on the law of conservation of energy, has good interpretability, and requires less computing power.

[0037] (3) Compared with other image recognition methods such as CCA correlation analysis, which require additional classifiers for image recognition, this invention can directly perform image recognition based on the strength of the correlation between images without the need for additional classifiers, thereby improving performance. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of the fuzzy monotonic correlation image recognition and machine learning method proposed in this invention;

[0040] Figure 2 This is a schematic diagram of the MNIST database sample proposed in Embodiment 2 of the present invention;

[0041] Figure 3 This is a schematic diagram of the Fashion-MNIST database sample proposed in Embodiment 2 of the present invention;

[0042] Figure 4 This is a schematic diagram of a USPS database sample proposed in Embodiment 2 of the present invention;

[0043] Figure 5 This is a schematic diagram of the COIL100 database sample proposed in Embodiment 2 of the present invention;

[0044] Figure 6 This is a schematic diagram of the CMU PIE database sample proposed in Embodiment 2 of the present invention;

[0045] Figure 7 This is a schematic diagram of the Extended Yale B database sample proposed in Embodiment 2 of the present invention;

[0046] Figure 8 This is a schematic diagram of "MNIST samples + independent random noise" proposed in Embodiment 2 of the present invention;

[0047] Figure 9 This is a schematic diagram of "USPS sample + speckle noise" proposed in Embodiment 2 of the present invention;

[0048] Figure 10 This is a schematic diagram of the "COIL100 sample + salt and pepper noise" proposed in Embodiment 2 of the present invention;

[0049] Figure 11 This is a schematic diagram of the "Fashion-MNIST sample + white occlusion noise" proposed in Embodiment 2 of the present invention;

[0050] Figure 12 This is a schematic diagram of the convergence of each sample p and accuracy acc proposed in Embodiment 2 of the present invention; where (a) is MNIST; (b) is COIL100; and (c) is USPS. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0052] Correlation-based algorithms using CCA (Correlation Analysis) have limitations when handling nonlinear relationships and noisy data. While variants of CCA can handle nonlinear data, they typically require a pre-processing module to convert the nonlinear data into linear data before correlation analysis, making the process complex. Furthermore, the nonlinear processing module's ability to extract nonlinear features is insufficient, potentially affecting subsequent image classification performance. Additionally, CCA-based algorithms commonly use traditional classifiers like SVM in image classification tasks, which may also limit performance improvements. Moreover, current machine learning methods generally believe that achieving good prediction results requires continuous optimization of the loss function, which is considered the primary approach to machine learning optimization. Therefore, the optimization of the loss function continuously increases the complexity of various parameters, leading to increasingly numerous and complex parameters in most current machine learning training methods. Meanwhile, mainstream deep learning models are constantly adding structures and parameters, making training increasingly expensive and complex, thus becoming increasingly reliant on computing power. This is mainly because deep learning is a black-box model, lacking interpretable patterns for discovery.

[0053] Because fuzzy monotonicity has the ability to directly perform nonlinear correlation analysis, using fuzzy monotonicity correlation analysis to replace classical correlation analysis for multi-view studies will avoid the problems associated with classical correlation analysis. Moreover, the characteristics of fuzzy monotonicity do not require static measurement using distance; they can be dynamically measured through interval changes, thereby offsetting some noise effects and improving performance. The resulting fuzzy monotonicity machine learning method does not require focusing on the optimization of the loss function, has fewer parameters, better robustness, and requires less computational power.

[0054] The following description, in conjunction with specific accompanying drawings and example data, illustrates the fuzzy monotonic correlation image recognition and machine learning method proposed in this invention, specifically including the following content.

[0055] Example 1:

[0056] This invention proposes a fuzzy, monotonic correlation image recognition and machine learning method. This method is based on the following fundamental technologies, specifically including:

[0057] 1) Initialize the reduced subset to an empty set, and begin traversing the attribute features;

[0058] 2) Sort the two continuous variable attributes and decision classes in ascending or descending order respectively to obtain two new sample sequences. C and D ;

[0059] 3) For the sample sequence C and D Partitioning, number of elements in each interval pThe value range is [2, n / 2], where n It is the number of samples;

[0060] 4) Traversal p The value of is obtained. C ={ C 1, C 2,…, C j}, D ={ D 1, D 2,…, D j},in j The value is n / p , C 1 represents an interval, and so on. The intervals between the partitions of two sequences are calculated according to equation (1):

[0061]

[0062] 5) in different p Under the given value, the fuzzy monotonic membership degree is calculated according to equation (2):

[0063]

[0064] 6) Analyze the changes in fuzzy monotonic membership degrees. { },exist p In the process of approaching 2 from n / 2, p Starting from a certain value, if If the value is consistently greater than or equal to 0.5, then it exhibits a relatively stable fuzzy monotonically increasing or decreasing relationship, where FI represents fuzzy monotonically increasing and FD represents fuzzy monotonically decreasing. A larger value indicates a stronger degree of fuzziness and monotonicity, and the further the value is from 2, the more stable it is. Therefore, this attribute feature and the decision class have a fuzzy monotonically increasing or fuzzy monotonically decreasing relationship, and the attribute feature and the decision class are highly correlated. Add this feature to the reduction subset.

[0065] Based on the above, this invention proposes a fuzzy monotone method correlation analysis (FMMCA) for image recognition and machine learning. From the perspective of correlation in fuzzy monotone method theory, the similarity between images can be characterized by the changing patterns of their original data: images with higher similarity exhibit more consistent data trends and similar energy distribution characteristics. Given that image data is typically stored in matrix form, FMMCA employs an innovative analysis method—assessing local fuzzy monotone correlation by pairwise comparison of row and column vectors of the image matrix, and then weighted summing of these local correlations to obtain the global fuzzy monotone correlation between images. Figure 1 This demonstrates the overall framework of the FMMCA technical solution. The specific implementation steps of the FMMCA technical solution are as follows:

[0066] Given two datasets and ,in N The number of samples representing the image. yes X The i One sample image, yes Y The j 1 ≤ 1 sample images i ≤ N ,1≤ j ≤ N .Will X i and Y j They are respectively denoted as matrices A m×n sum matrix B m×n .

[0067] To effectively extract sample images A and sample images B Correlation characteristics between them, FMMCA from A and B The corresponding column vectors or row vectors are extracted separately for correlation calculation, and then the local correlation results are integrated by weighted summation. Specifically, firstly, the process is traversed... A and B For all row vectors, calculate the local correlation along the row dimension, and its mathematical expression is as follows:

[0068] (3)

[0069] In equation (3), A T yesA The transpose of the matrix, express A T The i List, B T yes B The transpose of the matrix, express B T The i List, m Indicates the height of the sample image. k 1 indicates the number of intervals. p 1 represents the number of elements in the interval. , n Indicates the width of the sample image. k 1 and p The relationship between 1 and 2 is shown in formula (4).

[0070] (4)

[0071] In formula (4), when p When 1=1, Classified as n There are intervals, each containing only one element.

[0072] It can be calculated from the aforementioned equation (2). Parameter t 1 is k The optimal upper bound of 1 can be obtained by training for each database. t 1. To optimize algorithm performance, an iterative optimization method can be used, gradually adjusting the parameters. t 1. By evaluating the image classification accuracy after each adjustment, an optimal value is ultimately found. t 1 value. Specifically, first set an initial value. t A value of 1 is assigned, and then in each iteration, the value is adjusted based on the classification result. t 1. Fine-tune the settings until the image database achieves the highest classification accuracy for that parameter. This ensures... t The optimality of 1 is achieved, thereby maximizing the performance of image classification.

[0073] Secondly, through traversal A and B For all column vectors, calculate the local correlation in another dimension, as shown in the following mathematical expression:

[0074] (5)

[0075] In equation (5), express The List, express The List, Indicates the number of intervals. Indicates the number of elements in the interval. , It can be calculated from equation (2). and The relationship between them is shown in formula (6):

[0076] (6)

[0077] and The overall correlation between them can be obtained by combining both row and column dimensions, as shown in the following expression:

[0078] (7)

[0079] in, The higher the value, the better. and The stronger the fuzzy monotonic relationship between two images, the stronger the correlation between them. Therefore, the key to this machine learning approach is to train an optimal [system / mechanism] based on classification accuracy during training. t 1 and t Combinations of 2 are collectively referred to as t This is completely different from traditional loss function-based training. To facilitate training, we can set... t 1 and t 2. Unification equals a certain t The training process involves identifying the intervals that maximize classification accuracy. t .

[0080] For image classification tasks, FMMCA constructs a novel image framework that performs image classification by comparing the correlation strength between sample images. Assume the sample images... Y r The image categories are known. Now, in order to determine... To determine which image category it belongs to, you can... r The next iteration found and X i With the highest correlation Y r At this time, you can X i and Y r Images are grouped into the same image category. The mathematical formula for FMMCA to perform image classification tasks is as follows:

[0081] (8)

[0082] This eliminates the need for an additional classifier for image classification.

[0083] Example 2:

[0084] Based on Embodiment 1, but with a difference: to fully verify the effectiveness of the proposed technical solution, this invention selected several standard databases widely used in the field of image classification for testing, including MNIST, Fashion-MNIST, COIL100, USPS, CMU PIE, and Extended Yale B. These databases cover various task scenarios such as handwritten digit recognition, object recognition, and face recognition, and most are raw multi-view data, enabling a comprehensive evaluation of the technical solution's performance on different types of data. Examples of images from some of these databases are provided. Figure 2-11 As shown, the image data after adding noise is as follows: Figures 8-11 As shown.

[0085] The results of comparing the FMMCA method with various CCA methods on different datasets are shown in the table below.

[0086] Table 1. Classification results (%) of each experimental method on the COIL100 and Fashion-MNIST databases.

[0087]

[0088] Table 2. Classification results of each experimental method on the USPS and MNIST databases (%)

[0089]

[0090] Table 3. Classification results of each experimental method on the CMU PIE and Extended Yale B databases (%)

[0091]

[0092] Table 4. Classification results (%) of each experimental method on the noisy database

[0093]

[0094] In machine learning, according to formulas (3) to (7), machine training mainly involves finding the optimal upper bound t, which is primarily calculated using k and p. The effectiveness of this machine learning training can be demonstrated by the convergence of p and the accuracy acc. Figure 12 (a)- Figure 12 As shown in (c):

[0095] As shown in the figure above, this machine learning training method can easily converge to a high classification accuracy, requiring only one training parameter p. Compared with other current image recognition machine learning methods, it greatly reduces the number of training parameters and improves efficiency.

[0096] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. For those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for recognizing fuzzy, monotonic correlation images, characterized in that, The method specifically includes the following steps: S1, Given two datasets and ,in N Represents the sample size of the image. yes X The i One sample image, yes Y The j 1 ≤ 1 sample images i ≤ N ,1≤ j ≤ N ;Will X i and Y j They are respectively denoted as matrices A m×n sum matrix B m×n ; S2, Extraction X i and Y j The correlation characteristics between them, from the matrix A m×n sum matrix B m×n The corresponding column vectors and row vectors are extracted separately for correlation calculation, and the local correlation results are integrated by weighted summation; specifically, the following content is included: S2.1, By traversing the matrix A m×n sum matrix B m×n For all row vectors, calculate the local correlation along the row dimension; its mathematical expression is: In the formula, A T express A The transpose of the matrix, express A T The i List; B T express B The transpose of the matrix, express B T The i List; m Indicates the height of the sample image; k 1 indicates the number of intervals. p 1 represents the number of elements in the interval. ; k 1 and p The relationship between 1 and 2 is shown below: in, n Indicates the width of the sample image, when p When 1=1, Classified as n There are intervals, each containing only one element; parameter t 1 is k The optimal upper bound of 1 is determined by iterative optimization, gradually adjusting the parameters. t 1. By evaluating the image classification accuracy after each adjustment, an optimal value is ultimately found. t 1 value; S2.2, By traversing the matrix A m×n sum matrix B m×n For all column vectors, calculate the local correlation in another dimension, as shown in the following mathematical expression: In the formula, the parameters t 2 is k The optimal upper bound of 2; A j express A The j List; B j express B The j List; k 2 represents the number of intervals. p 2 represents the number of elements in the interval. , k 2 and p The relationship between 2 is shown below: S2.3, and The overall correlation between them is obtained by combining the row and column dimensions, as shown in the following expression: in, The higher the value, the better. and The stronger the fuzzy monotonic relationship between two samples, the stronger the correlation between them. S2.4 During machine learning training, an optimal [classification accuracy] is trained. t 1 and t Combinations of 2 are collectively referred to as t ;set up t 1 and t 2. Unification equals a certain t The training process involves identifying the interval divisions that achieve the highest classification accuracy. t As a characteristic of machine training; S3. Construct an image framework to achieve the image classification task, and complete the image classification by comparing the correlation strength between images.

2. The fuzzy monotonic correlation image recognition method according to claim 1, characterized in that, S3 specifically includes the following: Assuming sample image Y r The image category is known, and Find the match through iterative updates X i Most relevant Y r Furthermore X i and Y r Images are grouped into the same image category; the mathematical formula for completing the image classification task is as follows: In the formula, r Indicates the number of iterations.

3. A computer device, characterized in that, The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the fuzzy monotonic correlation image recognition method as described in any one of claims 1-2.

4. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the fuzzy monotonic correlation image recognition method as described in any one of claims 1-2.

Images