Residual intelligent detection method and device based on low-rank transformation

Abstract

The invention discloses a residual error intelligent detection method and device based on low-rank transformation, and the method comprises the steps: carrying out the preprocessing of training data and test data, and carrying out the decomposition processing of the preprocessed training data through robustness principal component analysis, and obtaining a low-rank matrix L and an error matrix; performing inverse solution on the low-rank matrix L to obtain a transformation matrix T, and calculating residual errors of the training data and the test data outside the transformation matrix T for the feature vector space; and in combination with the projection of the training data in the residual space, calculating a threshold value of a projection error corresponding to the training features, and outputting an identification result according to the projection of the test data in the residual space and the threshold value. According to the method, the low-rank feature part of the data space is extracted, the transformation matrix is solved through inverse solution, main features of the data are reserved while a large amount of communication data redundant information is screened out, and reliability and effectiveness are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data detection and classification, and in particular to a residual intelligent detection method and device based on low-rank transformation. Background Art

[0002] In the modern industrial field, data from all walks of life are diverse. Their complexity and engineering implementation requirements have brought great difficulties to the detection and classification of data. For example, in the process links such as vibration data detection and abnormal classification during mechanical operation, most of the existing detection and classification methods are aimed at a single feature of the data such as spectrum and time domain features, without taking into account the high-dimensional features contained in the data itself, and the effect has not been well solved. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a residual intelligent detection method and device based on low-rank transformation, which extracts the low-rank components of the data matrix and further inversely solves the low-rank transformation feature matrix and residuals to judge the residuals of the data falling outside the transformation feature space, thereby obtaining better recognition effects with as little cost as possible.

[0004] The technical solution adopted by the present invention is as follows:

[0005] A residual intelligent detection method based on low-rank transformation, comprising:

[0006] Preprocess the training data and test data, and decompose the preprocessed training data through robust principal component analysis to obtain a low-rank matrix L and an error matrix;

[0007] The low-rank matrix L is inversely solved to obtain the transformation matrix T, and the residuals of the preprocessed training data and test data outside the transformation matrix T are calculated;

[0008] Combined with the projection of the training data in the residual space, the threshold of the projection error corresponding to the training features is calculated, and the recognition result is output based on the projection of the test data in the residual space and the threshold.

[0009] Furthermore, the inverse solution of the low-rank matrix L to obtain the transformation matrix T includes:

[0010] Based on the low-rank matrix L and the preprocessed training data, the least squares method is used to inversely solve the transformation matrix T:

[0011] T*X=L

[0012] Among them, X is the preprocessed training data.

[0013] Furthermore, the projection of the training data in the residual space is combined to calculate the threshold of the projection error corresponding to the training feature, including: combining the transformation matrix T, calculating the projection T*X_test of the preprocessed test data under the transformation of the transformation matrix T, and calculating the error err_test; calculating the threshold tol according to the error matrix obtained by decomposing the preprocessed training data.

[0014] Furthermore, the calculating the threshold value tol according to the error matrix obtained by decomposing the preprocessed training data includes: using the mean of the training data error plus or minus three times the standard deviation as the upper and lower limits of the threshold value tol.

[0015] Furthermore, the outputting of the recognition result based on the projection of the test data in the residual space and the threshold value includes: judging whether the error err_test is greater than the allowable error, and if so, judging that the recognition result is that the test data and the training data do not belong to the same category; otherwise, judging that the test data and the training data belong to the same category.

[0016] A residual intelligent detection device based on low-rank transformation, comprising:

[0017] A preprocessing module, configured to preprocess the training data and the test data;

[0018] A decomposition module is configured to decompose the preprocessed training data through robust principal component analysis to obtain a low-rank matrix L and an error matrix;

[0019] A feature module is configured to inversely solve the low-rank matrix L to obtain a transformation matrix T;

[0020] A residual module, configured to calculate the residual of the preprocessed training data and test data outside the transformation matrix T;

[0021] A threshold module is configured to calculate a threshold of a projection error corresponding to a training feature in combination with a projection of the training data in the residual space;

[0022] The judgment module is configured to output a recognition result according to the projection of the test data in the residual space and the threshold.

[0023] Furthermore, in the feature module, the low-rank matrix L is inversely solved to obtain the transformation matrix T, including:

[0024] Based on the low-rank matrix L and the preprocessed training data, the least squares method is used to inversely solve the transformation matrix T:

[0025] T*X=L

[0026] Among them, X is the preprocessed training data.

[0027] Furthermore, in the threshold module, the projection of the training data in the residual space is combined to calculate the threshold of the projection error corresponding to the training feature, including: combining the transformation matrix T, calculating the projection T*X_test of the preprocessed test data under the transformation of the transformation matrix T, and calculating the error err_test; calculating the threshold tol based on the error matrix obtained by decomposing the preprocessed training data.

[0028] Furthermore, in the threshold module, the threshold tol is calculated according to the error matrix obtained by decomposing the preprocessed training data, including: using the mean of the training data error plus or minus three times the standard deviation as the upper and lower limits of the threshold tol.

[0029] Furthermore, in the judgment module, the recognition result is output according to the projection of the test data in the residual space and the threshold, including: judging whether the error err_test is greater than the allowable error, if so, judging that the recognition result is that the test data and the training data do not belong to the same category; otherwise, judging that the test data and the training data belong to the same category.

[0030] The beneficial effects of the present invention are:

[0031] (1) Reliability of feature selection. For general matrices, the present invention uses robust principal component analysis to solve the low-rank part. On this basis, by further digging into the low-rank part and the original matrix, the transformation characteristics of the low-rank spatial part of the data can be cleverly grasped, thereby reducing the redundant information of the data and using it to characterize the feature space with high accuracy.

[0032] (2) Contains continuous information of data. The present invention selects to collect continuous data as samples, avoiding breaking up the data and destroying the continuous relationship of the data. The continuous relationship corresponding to the image is not destroyed, so that more complete information can be extracted, and the effect will be better.

[0033] (3) Due to the large amount of redundant data, the low-rank method of the present invention is not only for extracting features from high-dimensional data, but also for easily capturing the low-rank transformation matrix of the data, making it simple and easy to operate, and saving computing resources.

[0034] (4) The present invention can be applied to the field of data detection and classification, and can also be applied to other fields, such as packaging detection and anomaly classification of electronic devices such as chips. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of a residual intelligent detection method based on low-rank transformation in Example 1 of the present invention.

[0036] Figure 2This is a flow chart of a residual intelligent detection method based on low-rank transformation in Example 2 of the present invention.

[0037] Figure 3 This is a block diagram of a residual intelligent detection device based on low-rank transformation according to embodiment 3 of the present invention. DETAILED DESCRIPTION

[0038] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific implementation methods of the present invention are now described. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the embodiments described are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0039] Example 1

[0040] This embodiment provides a residual intelligent detection method based on low-rank transformation, including:

[0041] Preprocess the training data and test data, and decompose the preprocessed training data through robust principal component analysis to obtain a low-rank matrix L and an error matrix;

[0042] The low-rank matrix L is inversely solved to obtain the transformation matrix T, and the residuals of the preprocessed training data and test data outside the transformation matrix T are calculated;

[0043] Combined with the projection of the training data in the residual space, the threshold of the projection error corresponding to the training features is calculated, and the recognition result is output based on the projection of the test data in the residual space and the threshold.

[0044] Specifically, Figure 1 As shown, the residual intelligent detection method of this embodiment can be implemented by the following steps:

[0045] S100. Preprocess the training data, and preprocess the test data based on the mean and standard deviation of the training data;

[0046] S200. Decompose the preprocessed training data X through robust principal component analysis to obtain the corresponding low-rank matrix L and error matrix err_train;

[0047] S300. Using the pre-processed training data X and the low-rank matrix L, the least square method is used to inversely solve the transformation matrix T to satisfy T*X=L;

[0048] S400. Combine the transformation matrix T, calculate the projection T*X_test of the preprocessed test data X_test under the transformation matrix T, calculate the error err_test, and calculate the threshold tol according to the error of step S200;

[0049] S500. Determine whether the error err_test is greater than the allowable error. If the error err_test is greater than the allowable error, the recognition result is that the test data and the training data do not belong to the same category; if the error is less than the allowable error, the recognition result is that the test data and the training data belong to the same category, and the task ends.

[0050] Preferably, in order to ensure data completeness, when selecting data, it is necessary to ensure that the number of columns of the data matrix is ​​much larger than the dimension of a data sample.

[0051] Preferably, in order to ensure the generality of recognition, in step 400, the mean of the training data error plus or minus three times the standard deviation is used as the upper and lower limits of the threshold, thereby ensuring the accuracy of recognition.

[0052] Preferably, when faced with multiple data for training classification, k transformation matrices T1, T2, ..., Tk are obtained for each type of data in combination with the above steps, and corresponding errors err_test_1, err_test_2, ..., err_test_k can be obtained for a set of test data X_test. If err_test_i is the smallest, X_test belongs to the i-th category.

[0053] Example 2

[0054] This embodiment is based on embodiment 1:

[0055] like Figure 2 As shown, this embodiment provides a residual intelligent detection method based on low-rank transformation, comprising the following steps:

[0056] S100-1. Obtain the corresponding transformation matrix T1 by training the training data X1;

[0057] S100-2. Obtain the corresponding transformation matrix T2 by training the training data X2;

[0058] S200-1. Combine the transformation matrix T1, calculate the projection T1*X_test of the preprocessed test data X_test under the transformation of T1, and calculate the error err_test_1;

[0059] S200-2. Combine the transformation matrix T2, calculate the projection T2*X_test of the preprocessed test data X_test under the transformation of T1, and calculate the error err_test_2;

[0060] S300. Perform a modulo comparison on the errors err_test_1 and err_test_2. If err_test_1 < err_test_2 and err_test_1 < tol1, the test data belongs to the first category. If err_test_2 < err_test_1 and err_test_2 < tol2, the test data belongs to the second category. Otherwise, it belongs to other categories.

[0061] Example 3

[0062] As Figure 3 shown, this embodiment provides a residual intelligent detection device based on low-rank transformation, including:

[0063] A preprocessing module configured to preprocess training data and test data;

[0064] A decomposition module configured to perform decomposition processing on the preprocessed training data through robust principal component analysis to obtain a low-rank matrix L and an error matrix;

[0065] A feature module configured to inversely solve the low-rank matrix L to obtain a transformation matrix T;

[0066] A residual module configured to calculate the residuals of the preprocessed training data and test data outside the transformation matrix T;

[0067] A threshold module configured to calculate the threshold of the projection error corresponding to the training features in combination with the projection of the training data in the residual space;

[0068] A judgment module configured to output an identification result based on the projection of the test data in the residual space and the threshold.

[0069] Preferably, in the feature module, inversely solving the low-rank matrix L to obtain the transformation matrix T includes:

[0070] Based on the low-rank matrix L and the preprocessed training data, use the least squares method to inversely solve and obtain the transformation matrix T:

[0071] T*X = L

[0072] where X is the preprocessed training data.

[0073] Preferably, in the threshold module, calculating the threshold of the projection error corresponding to the training features in combination with the projection of the training data in the residual space includes: combining the transformation matrix T, calculating the projection T*X_test under the transformation of the transformation matrix T for the preprocessed test data, and calculating the error err_test; calculating the threshold tol according to the error matrix obtained by decomposing the preprocessed training data.

[0074] Preferably, in the threshold module, the threshold tol is calculated according to the error matrix obtained by decomposing the preprocessed training data, including: using the mean of the training data error plus or minus three times the standard deviation as the upper and lower limits of the threshold tol.

[0075] Preferably, in the judgment module, the recognition result is output according to the projection of the test data in the residual space and the threshold, including: judging whether the error err_test is greater than the allowable error, if so, judging that the recognition result is that the test data and the training data do not belong to the same category; otherwise, judging that the test data and the training data belong to the same category.

[0076] Example 4

[0077] This embodiment is based on embodiment 1:

[0078] This embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a residual intelligent detection method based on low-rank transformation of embodiment 1 is implemented. The computer program may be in source code form, object code form, executable file, or some intermediate form.

[0079] Example 5

[0080] This embodiment is based on embodiment 1:

[0081] The present embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a residual intelligent detection method based on low-rank transformation of embodiment 1. The computer program may be in source code form, object code form, executable file, or some intermediate form, etc. The storage medium includes: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content contained in the storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electric carrier signals and telecommunication signals.

[0082] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

Claims

1. A residual intelligent detection method based on low-rank transformation, characterized in that: include: Preprocess the training data and test data, and decompose the preprocessed training data through robust principal component analysis to obtain a low-rank matrix L and an error matrix; The low-rank matrix L is inversely solved to obtain the transformation matrix T, and the residuals of the preprocessed training data and test data outside the transformation matrix T are calculated; Combined with the projection of the training data in the residual space, the threshold of the projection error corresponding to the training features is calculated, and the recognition result is output based on the projection of the test data in the residual space and the threshold.

2. According to claim 1, a residual intelligent detection method based on low-rank transformation is characterized in that: The inverse solution of the low-rank matrix L to obtain the transformation matrix T includes: Based on the low-rank matrix L and the preprocessed training data, the least squares method is used to inversely solve the transformation matrix T: T*X=L Among them, X is the preprocessed training data.

3. The method for intelligent detection of residual errors based on low-rank transformation according to claim 2, characterized in that: The projection of the training data in the residual space is combined to calculate the threshold of the projection error corresponding to the training feature, including: combining the transformation matrix T, calculating the projection T*X_test of the preprocessed test data under the transformation of the transformation matrix T, and calculating the error err_test; calculating the threshold tol according to the error matrix obtained by decomposing the preprocessed training data.

4. The method for intelligent residual detection based on low-rank transformation according to claim 3, characterized in that: The calculating the threshold value tol according to the error matrix obtained by decomposing the preprocessed training data includes: using the mean value of the training data error plus or minus three times the standard deviation as the upper and lower limits of the threshold value tol.

5. The method for intelligent detection of residual errors based on low-rank transformation according to claim 3, characterized in that: The recognition result is output according to the projection of the test data in the residual space and the threshold, including: judging whether the error err_test is greater than the allowable error, if so, judging that the recognition result is that the test data and the training data do not belong to the same category; otherwise, judging that the test data and the training data belong to the same category.

6. A residual intelligent detection device based on low-rank transformation, characterized in that: include: A preprocessing module, configured to preprocess the training data and the test data; A decomposition module is configured to decompose the preprocessed training data through robust principal component analysis to obtain a low-rank matrix L and an error matrix; A feature module is configured to inversely solve the low-rank matrix L to obtain a transformation matrix T; A residual module, configured to calculate the residual of the preprocessed training data and test data outside the transformation matrix T; A threshold module is configured to calculate a threshold of a projection error corresponding to a training feature in combination with a projection of the training data in the residual space; The judgment module is configured to output a recognition result according to the projection of the test data in the residual space and the threshold.

7. The low-rank transform-based residual intelligent detection device according to claim 6, characterized in that: In the feature module, the low-rank matrix L is inversely solved to obtain the transformation matrix T, including: Based on the low-rank matrix L and the preprocessed training data, the least squares method is used to inversely solve the transformation matrix T: T*X=L Among them, X is the preprocessed training data.

8. The low-rank transform-based residual intelligent detection device according to claim 7, characterized in that: In the threshold module, the projection error threshold corresponding to the training feature is calculated in combination with the projection of the training data in the residual space, including: combining the transformation matrix T, calculating the projection T*X_test of the preprocessed test data under the transformation of the transformation matrix T, and calculating the error err_test; calculating the threshold tol according to the error matrix obtained by decomposing the preprocessed training data.

9. The low-rank transform-based residual intelligent detection device according to claim 8, characterized in that: In the threshold module, the threshold tol is calculated according to the error matrix obtained by decomposing the preprocessed training data, including: using the mean of the training data error plus or minus three times the standard deviation as the upper and lower limits of the threshold tol.

10. The low-rank transform-based residual intelligent detection device according to claim 8, characterized in that: In the judgment module, the recognition result is output according to the projection of the test data in the residual space and the threshold value, including: judging whether the error err_test is greater than the allowable error, if so, judging that the recognition result is that the test data and the training data do not belong to the same category; otherwise, judging that the test data and the training data belong to the same category.

Images