CN108681697A - Feature selection approach and device - Google Patents

Abstract

The embodiment of the present invention discloses a feature selection method and device, which can realize feature selection for target object measurement in spectral nondestructive testing, and have good robustness and stability. The method includes: S1, obtaining the spectral data set of the sample; S2, sampling the spectral data set for a first number of times to obtain the first number of sample spaces, and for each sample space, using the sample space to construct a partial The least squares quantitative analysis model, and based on the partial least squares quantitative analysis model, the importance of the features corresponding to the sample space is sorted; S3, according to the importance sorting results of the features corresponding to the first number of sample spaces The features corresponding to the first number of sample spaces are sorted to obtain a feature sorting result, the number of feature selections is determined based on the feature sorting results, and the features of the aforementioned feature selection number are selected as target features according to the feature sorting results.

Description

Feature selection method and device

technical field

Embodiments of the present invention relate to the field of spectral analysis, and in particular to a feature selection method and device.

Background technique

Spectrum-based non-destructive testing technology has the advantages of simple operation, fast, non-destructive, no need for pretreatment and auxiliary reagents, and is widely used in agriculture, chemical industry, environment, medicine and other fields. However, the obtained spectral data usually has the characteristics of large number of characteristic wavelengths, small sample size, and collinearity between features, which makes the direct use of all features to establish a model ineffective. Existing studies have shown that the feature selection method is used to select the wavelength and then build the model, which has the advantages of improving model accuracy, simplifying model complexity, speeding up model operation speed, and enhancing model interpretability.

Common feature selection methods in the field of spectral analysis include: non-informative variable elimination method, competitive adaptive weight method, continuous projection algorithm, interval partial least squares method, and features based on optimization (genetic algorithm, simulated annealing, particle swarm, etc.) selection method etc. These methods are all single-feature selection methods. When a small amount of local variation occurs in the data set, the selected features change greatly and the stability is not good.

Contents of the invention

Aiming at the deficiencies and defects of the prior art, embodiments of the present invention provide a feature selection method and device.

On the one hand, the embodiment of the present invention proposes a feature selection method, including:

S1. Obtain the spectral data set of the sample, wherein the spectral data set of the sample includes spectral data of a specific number of samples, and the content of the target substance in the sample;

S2. Sampling the spectral data set for the first number of times to obtain the first number of sample spaces. For each sample space, use the sample space to construct a partial least squares quantitative analysis model, and based on the partial minimum The quadratic quantitative analysis model ranks the importance of the features corresponding to the sample space;

S3. Sorting the features corresponding to the first number of sample spaces according to the importance ranking results of the features corresponding to the first number of sample spaces, obtaining a feature ranking result, and determining the number of feature selections based on the feature ranking result, And according to the feature sorting result, select the number of features mentioned above as the target features.

On the other hand, an embodiment of the present invention proposes a feature selection device, including:

an acquisition unit, configured to acquire a spectral data set of a sample, wherein the spectral data set of the sample includes spectral data of a specific number of samples, and the content of the target substance in the sample;

The sorting unit is used to sample the spectral data set for a first number of times to obtain the first number of sample spaces, and for each sample space, use the sample space to construct a partial least squares quantitative analysis model, and based on The partial least squares quantitative analysis model ranks the importance of the features corresponding to the sample space;

A selection unit, configured to sort the features corresponding to the first number of sample spaces according to the importance ranking results of the features corresponding to the first number of sample spaces, to obtain a feature ranking result, and determine the feature based on the feature ranking result Select the quantity, and select the above-mentioned features according to the feature sorting result to select the number of features as the target features.

In a third aspect, an embodiment of the present invention provides an electronic device, including: a processor, a memory, a bus, and a computer program stored in the memory and operable on the processor;

Wherein, the processor and the memory complete the mutual communication through the bus;

The above method is implemented when the processor executes the computer program.

In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, where a computer program is stored on the storage medium, and the above method is implemented when the computer program is executed by a processor.

In the feature selection method and device provided in the embodiments of the present invention, the spectral data set of the sample is first obtained; then the spectral data set is sampled a first number of times to obtain the first number of sample spaces, and for each sample space, Using the sample space to construct a partial least squares quantitative analysis model, and based on the partial least squares quantitative analysis model, the importance of the features corresponding to the sample space is sorted; finally according to the features corresponding to the first number of sample spaces Sorting the features corresponding to the first number of sample spaces to obtain the feature ranking results, determining the number of feature selections based on the feature ranking results, and selecting the aforementioned feature selection numbers according to the feature ranking results A feature is used as the target feature. Compared with the single feature selection method, when a small amount of local variation occurs in the data set in this scheme, the selected feature does not change much, and the stability is better, that is, it can realize the detection of target objects in spectral nondestructive testing. The selection of features has good robustness and stability. In addition, using the partial least squares method as a quantitative analysis model, it is found through experiments that on the experimental data set, the accuracy of the model constructed by using this method for feature selection is better than that of using The accuracy of the model constructed by all features.

Description of drawings

Fig. 1 is a schematic flow chart of an embodiment of the feature selection method of the present invention;

2 is a schematic structural view of an embodiment of the feature selection device of the present invention;

FIG. 3 is a schematic diagram of a physical structure of an electronic device provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are the Some, but not all, embodiments are invented. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the embodiments of the present invention.

Referring to Fig. 1, the present embodiment discloses a feature selection method, including:

S1. Obtain the spectral data set of the sample, wherein the spectral data set of the sample includes spectral data of a specific number of samples, and the content of the target substance in the sample;

S2. Sampling the spectral data set for the first number of times to obtain the first number of sample spaces. For each sample space, use the sample space to construct a partial least squares quantitative analysis model, and based on the partial minimum The square quantitative analysis model ranks the importance of the features (ie wavelength) corresponding to the sample space;

In this embodiment, the sample space is used to build a partial least squares quantitative analysis model. The specific process is: select a part of the samples in the sample space as the training set to train the partial least squares quantitative analysis model, and select the partial least squares quantitative analysis model in the sample space. The remaining samples of the partial least squares quantitative analysis model are used as the test set to test the trained partial least squares quantitative analysis model, and the root mean square error of the partial least squares quantitative analysis model is calculated based on the test results; the above process is repeated until the partial least squares quantitative analysis model is obtained. The root mean square error of the analysis model is within a preset range, and the partial least squares quantitative analysis model corresponding to the root mean square error is the partial least squares quantitative analysis model to be constructed.

S3. Sorting the features corresponding to the first number of sample spaces according to the importance ranking results of the features corresponding to the first number of sample spaces, obtaining a feature ranking result, and determining the number of feature selections based on the feature ranking result, And according to the feature sorting result, select the number of features mentioned above as the target features.

In the feature selection method provided by the embodiment of the present invention, the spectral data set of the sample is first obtained; then the spectral data set is sampled a first number of times to obtain the first number of sample spaces, and for each sample space, the Constructing a partial least squares quantitative analysis model in the sample space, and sorting the importance of the features corresponding to the sample space based on the partial least squares quantitative analysis model; finally according to the importance of the features corresponding to the first number of sample spaces sort the features corresponding to the first number of sample spaces to obtain a feature sorting result, determine the number of feature selections based on the feature sorting results, and select the previously described feature selection number of features according to the feature sorting results As the target feature, compared with the single feature selection method, when the data set has a small amount of local variation, the selected feature does not change much and the stability is better, that is, it can realize the feature used for the determination of the target object in the spectral non-destructive testing. selection, has good robustness and stability. In addition, using the partial least squares method as the quantitative analysis model, it is found through experiments that on the experimental data set, the accuracy of the model constructed by using this method for feature selection is better than that of using all features. The accuracy of the built model.

On the basis of the foregoing method embodiments, the sampling may use a bootstrap resampling method, each sampling is a replacement sampling, and the capacity of each sample space is the same as that of the spectral data set.

On the basis of the foregoing method embodiments, the importance ranking of the features corresponding to the sample space based on the partial least squares quantitative analysis model may include:

The coefficient corresponding to each feature in the partial least squares quantitative analysis model is obtained, and the feature ranking vector in the sample space is obtained by sorting the absolute values of the coefficients in descending order.

In this embodiment, the feature ranking vector in the sample space is composed of the importance characterization of each feature in the sample space, and the importance characterization of each feature is the absolute value of the coefficient corresponding to the feature.

On the basis of the foregoing method embodiments, the ranking of the features corresponding to the first number of sample spaces according to the importance ranking results of the features corresponding to the first number of sample spaces to obtain the feature ranking results may include :

A linear weighted sum method is used to integrate the feature ranking vectors in each sample space to obtain the feature ranking result.

In this embodiment, the specific process of using the linear weighted summation method to integrate the feature ranking vectors under each sample space to obtain the feature ranking result is: for each feature, corresponding to the feature in all feature ranking vectors The weighted summation of the values of the features, where the weight of the corresponding value of the feature in each feature sorting vector is obtained based on the partial least squares quantitative analysis model determined by the sample space corresponding to the feature sorting vector. The specific calculation process is: first , using each sample space to construct a partial least squares quantitative analysis model, calculate the root mean square error value of the partial least squares quantitative analysis model constructed by each sample space, and the partial least squares quantitative analysis model obtained by the ith sample space The root mean square error value of the model is recorded as E _i , the smaller the value, the higher the correct rate of the model; secondly, calculate the reciprocal of E _i and record it as O _i ; thirdly, sum these O _i to get SUM; finally, Divide each O _i by SUM to get the weight W _i , which is the weight of the corresponding value of each feature in the feature sorting vector under the i-th sample space; according to the size of the summation result corresponding to each feature obtained in the previous step , sort each feature to obtain the feature sorting result, and the feature with a larger summation result in the feature sorting result is ranked at a higher position.

On the basis of the foregoing method embodiments, the determination of the number of feature selections based on the feature ranking results may include:

By setting the number of feature selections to different values, and based on the values set each time, select the set values and features on the spectral data set according to the feature sorting results, and construct a partial least squares based on the feature selection results. Multiply the quantitative analysis model;

Select the number of feature selections corresponding to the least squares quantitative analysis model with the smallest root mean square error among the least squares quantitative analysis models obtained in the previous step as the target value.

In this embodiment, a cross-validation method may be used for selecting the target value.

Based on the foregoing method embodiments, the difference between every two adjacent values among the different values may be 1.

In this embodiment, the number of different values may be the same as the capacity of the spectral data set.

Referring to Figure 2, this embodiment discloses a feature selection device, including:

An acquisition unit 1, configured to acquire a spectral data set of a sample, wherein the spectral data set of the sample includes spectral data of a specific number of samples, and the content of the target substance in the sample;

A sorting unit 2, configured to sample the spectral data set a first number of times to obtain the first number of sample spaces, and for each sample space, use the sample space to construct a partial least squares quantitative analysis model, and Based on the partial least squares quantitative analysis model, the importance of the features corresponding to the sample space is sorted;

The selection unit 3 is configured to sort the features corresponding to the first number of sample spaces according to the importance ranking results of the features corresponding to the first number of sample spaces, to obtain a feature ranking result, and determine based on the feature ranking result Select the number of features, and select the features of the aforementioned feature selection number as the target features according to the sorting result of the features.

Specifically, the acquisition unit 1 acquires the spectral data set of the sample, wherein the spectral data set of the sample includes the spectral data of a specific number of samples, and the content of the target substance in the sample; the sorting unit 2 sorts the The spectral data set is sampled for a first number of times to obtain the first number of sample spaces. For each sample space, a partial least squares quantitative analysis model is constructed using the sample space, and based on the partial least squares quantitative analysis The model sorts the importance of the features corresponding to the sample space; the selection unit 3 sorts the features corresponding to the first number of sample spaces according to the importance ranking results of the features corresponding to the first number of sample spaces , to obtain the feature ranking result, determine the number of feature selections based on the feature ranking result, and select the features of the aforementioned feature selection number as the target features according to the feature ranking result.

The feature selection device provided by the embodiment of the present invention first acquires the spectral data set of the sample; then performs a first number of samples on the spectral data set to obtain the first number of sample spaces, and for each sample space, use the Constructing a partial least squares quantitative analysis model in the sample space, and sorting the importance of the features corresponding to the sample space based on the partial least squares quantitative analysis model; finally according to the importance of the features corresponding to the first number of sample spaces sort the features corresponding to the first number of sample spaces to obtain a feature sorting result, determine the number of feature selections based on the feature sorting results, and select the previously described feature selection number of features according to the feature sorting results As the target feature, compared with the single feature selection method, when the data set has a small amount of local variation, the selected feature does not change much and the stability is better, that is, it can realize the feature used for the determination of the target object in the spectral non-destructive testing. selection, has good robustness and stability. In addition, using the partial least squares method as the quantitative analysis model, it is found through experiments that on the experimental data set, the accuracy of the model constructed by using this method for feature selection is better than that of using all features. The accuracy of the built model.

On the basis of the foregoing device embodiments, the selection unit may specifically be used for:

By setting the number of feature selections to different values, and based on the values set each time, select the set values and features on the spectral data set according to the feature sorting results, and construct a partial least squares based on the feature selection results. Multiply the quantitative analysis model;

Select the number of feature selections corresponding to the least squares quantitative analysis model with the smallest root mean square error among the least squares quantitative analysis models obtained in the previous step as the target value.

The feature selection device of this embodiment can be used to implement the technical solutions of the foregoing method embodiments, and its implementation principles and technical effects are similar, and will not be repeated here.

The advantages of the present invention are: the idea of integrated learning is introduced into the field of feature selection, and it is applied to the selection of characteristic wavelengths in the quantitative analysis of sample spectra. The established model of the selected wavelength is better than the prediction effect of the whole band, which provides a new method for spectral wavelength selection and has high practical value.

FIG. 3 shows a schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. As shown in FIG. 3 , the electronic device may include: a processor 11, a memory 12, a bus 13, and a A computer program running on the processor 11;

Wherein, the processor 11 and the memory 12 complete mutual communication through the bus 13;

When the processor 11 executes the computer program, it implements the methods provided in the above method embodiments, for example, including: acquiring a spectral data set of a sample; sampling the spectral data set a first number of times to obtain the first number of sample spaces, for each sample space, use the sample space to construct a partial least squares quantitative analysis model, and based on the partial least squares quantitative analysis model, sort the importance of the features corresponding to the sample space; according to the The importance sorting results of the features corresponding to the first number of sample spaces sort the features corresponding to the first number of sample spaces to obtain the feature sorting results, determine the number of feature selections based on the feature sorting results, and according to the described Feature sorting result selection The aforementioned feature selection quantity features are used as target features.

An embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the methods provided in the above-mentioned method embodiments are implemented, for example, including: acquiring spectral data of a sample set; the spectral data set is sampled for the first number of times to obtain the first number of sample spaces, and for each sample space, a partial least squares quantitative analysis model is constructed using the sample space, and based on the partial minimum The quadratic quantitative analysis model sorts the importance of the features corresponding to the sample space; sorts the features corresponding to the first number of sample spaces according to the importance sorting results of the features corresponding to the first number of sample spaces, Obtain the feature sorting result, determine the number of feature selections based on the feature sorting result, and select the features of the aforementioned feature selection number as target features according to the feature sorting result.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is a relationship between these entities or operations. There is no such actual relationship or order between them. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element. The orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must Having a particular orientation, being constructed and operating in a particular orientation, and therefore not to be construed as limiting the invention. Unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be interpreted in a broad sense, for example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection, It can also be an electrical connection; it can be a direct connection, or an indirect connection through an intermediary, or an internal communication between two components. Those of ordinary skill in the art can understand the specific meanings of the above terms in the present invention according to specific situations.

In the description of the invention, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure the understanding of this description. Similarly, it should be appreciated that in the above description of exemplary embodiments of the invention, in order to streamline the present disclosure and to facilitate understanding of one or more of the various inventive aspects, various features of the invention are sometimes grouped together into a single embodiment , figure, or description of it. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of this invention. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present invention is not limited to any single aspect, nor to any single embodiment, nor to any combination and/or permutation of these aspects and/or embodiments. Furthermore, each aspect and/or embodiment of the invention may be used alone or in combination with one or more other aspects and/or embodiments thereof.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: It is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the various embodiments of the present invention. All of them should be covered by the scope of the claims and description of the present invention.

Claims (10)

1. a kind of feature selection approach, which is characterized in that including：

S1, the spectroscopic data collection for obtaining sample, wherein the spectroscopic data collection of the sample includes the spectrum of certain amount of sample The content of object in data and the sample；

S2, the first quantity time sampling is carried out to the spectroscopic data collection, the first quantity sample space is obtained, for each A sample space builds an offset minimum binary Quantitative Analysis Model using the sample space, and fixed based on the offset minimum binary It measures analysis model and importance ranking is carried out to the feature corresponding to the sample space；

S3, according to the importance ranking result of the corresponding feature of the first quantity sample space to the first quantity sample The corresponding feature in this space is ranked up, and obtains feature ordering as a result, determining feature selecting number based on the feature ordering result Amount, and select the preceding feature selecting quantity feature as target signature according to the feature ordering result.

2. according to the method described in claim 1, it is characterized in that, the sampling uses bootstrap method for resampling, every time It is sampled as putting back to sampling, the capacity of each sample space is identical as the capacity of spectroscopic data collection.

3. according to the method described in claim 2, it is characterized in that, the described offset minimum binary Quantitative Analysis Model that is based on is to this Feature corresponding to sample space carries out importance ranking, including：

The coefficient corresponding to each feature in the offset minimum binary Quantitative Analysis Model is obtained, the absolute value to the coefficient is passed through It is ranked up according to sequence from big to small, obtains the feature ordering vector under the sample space.

4. according to the method described in claim 3, it is characterized in that, described corresponding according to the first quantity sample space The importance ranking result of feature is ranked up the corresponding feature of the first quantity sample space, obtains feature ordering knot Fruit, including：

The feature ordering summed under each sample space of method integration using linear weighted function is vectorial, obtains the feature ordering knot Fruit.

5. according to the method described in claim 1, it is characterized in that, described determine feature selecting based on the feature ordering result Quantity, including：

It is different values by the way that the feature selecting quantity is arranged, and based on the value being arranged every time, according to the feature ordering knot Fruit selects the value feature of the secondary setting on the spectroscopic data collection, and it is quantitative that feature based selection result builds offset minimum binary Analysis model；

Choose the least square of corresponding root-mean-square error minimum in each least square Quantitative Analysis Model that back obtains It is worth for the purpose of feature selecting quantity corresponding to Quantitative Analysis Model.

6. according to the method described in claim 1, it is characterized in that, each two adjacent value differs 1 in the different value.

7. a kind of feature selecting device, which is characterized in that including：

Acquiring unit, the spectroscopic data collection for obtaining sample, wherein the spectroscopic data collection of the sample includes certain amount of The content of object in the spectroscopic data of sample and the sample；

It is empty to obtain the first quantity sample for carrying out the first quantity time sampling to the spectroscopic data collection for sequencing unit Between, for each sample space, an offset minimum binary Quantitative Analysis Model is built using the sample space, and partially based on this Least square Quantitative Analysis Model carries out importance ranking to the feature corresponding to the sample space；

Selecting unit, for according to the importance ranking result of the corresponding feature of the first quantity sample space to described the The corresponding feature of one quantity sample space is ranked up, and obtains feature ordering as a result, being determined based on the feature ordering result Feature selecting quantity, and select the preceding feature selecting quantity feature as target signature according to the feature ordering result.

8. device according to claim 7, which is characterized in that the selecting unit is specifically used for：

It is different values by the way that the feature selecting quantity is arranged, and based on the value being arranged every time, according to the feature ordering knot Fruit selects the value feature of the secondary setting on the spectroscopic data collection, and it is quantitative that feature based selection result builds offset minimum binary Analysis model；

Choose the least square of corresponding root-mean-square error minimum in each least square Quantitative Analysis Model that back obtains It is worth for the purpose of feature selecting quantity corresponding to Quantitative Analysis Model.

9. a kind of electronic equipment, which is characterized in that including：Processor, memory, bus and storage on a memory and can located The computer program run on reason device；

Wherein, the processor, memory complete mutual communication by the bus；

The processor realizes the method as described in any one of claim 1-6 when executing the computer program.

10. a kind of non-transient computer readable storage medium, which is characterized in that be stored with computer journey on the storage medium Sequence realizes the method as described in any one of claim 1-6 when the computer program is executed by processor.