WO2018121082A1 - Self-learning-type qualitative analysis method based on raman spectrum - Google Patents

Abstract

Disclosed is a method for carrying out self-learning-type qualitative analysis based on a Raman spectrum. The method comprises: a Raman spectrum acquisition step for acquiring a Raman spectrum; a feature extraction and comparison step for extracting spectrum data and comparing same with a spectrum feature library of a spectrogram library so as to obtain an original recognition substance ID list; a similarity comparison step for obtaining, by means of calculation, the similarity between substance IDs in the original recognition substance ID list for the Raman spectrum so as to generate a similarity list, and comparing same with a similarity threshold value library in the spectrogram library; and a substance ID selection step for verifying, based on a self-learning library, a list, obtained after comparison with a similarity threshold value, of similarity recognition substance IDs, the similarity thereof exceeding a threshold value, comprising when there is, in the similarity list, a substance ID exceeding a threshold value corresponding to a substance ID in the similarity threshold value library, executing false positive detection; and when there is not, in the similarity list, a substance ID exceeding a threshold value corresponding to a substance ID in the similarity threshold value library, executing false negative detection.

Description

Self-learning qualitative analysis method based on Raman spectroscopy

Cross-reference to related applications

The present application claims the benefit of the Chinese Patent Application No. 201611220308.2, filed on Dec. 26, 2016, the entire content of which is hereby incorporated by reference.

Technical field

The present disclosure relates to the field of Raman spectroscopy, and in particular to a self-learning qualitative analysis method based on Raman spectroscopy.

Background technique

Raman spectroscopy is a non-contact spectroscopy technique based on the Raman scattering effect of excitation light. It can qualitatively and quantitatively analyze the composition of a substance. Raman spectroscopy is a molecular vibrational spectroscopy that reflects the fingerprint characteristics of a molecule, and the Raman spectrum of each substance is unique. The Raman spectrum obtained by comparison with the known Raman spectrum database of various substances is used to identify the composition of the substance to be tested, and thus can be used for detecting substances, and has been widely used for liquid security. , jewelry testing, explosives testing, drug testing, drug testing and other fields.

In the prior art, a conventional Raman spectroscopy detecting device generally performs a qualitative analysis based on a spectral database to perform a qualitative measurement, and finally displays a measurement result, and the approximate workflow can be summarized as: collecting spectral data; preprocessing the acquired spectral image The pre-processed acquired spectra are compared with the spectral library; the qualitative analysis results are obtained; and the qualitative analysis results are displayed.

The Raman spectral similarity of the two species can be quantitatively represented, for example, by a "similarity" parameter, such as the similarity function commonly used to calculate similarity.

However, this conventional Raman spectroscopy method for qualitative analysis generally has a high false alarm rate and a false negative rate for substances of low purity, and is merely exhaustive and mechanically performed with the exhaustivity of the spectral database. Contrast until a consistent alignment result is obtained to complete the qualitative analysis, so that the analysis process takes a long time; and the sample with a small difference between the two components adopts a global simple repeated Raman spectral similarity alignment analysis, which is difficult to be similar Degree calculation result Therefore, the current conventional similarity calculation method and the similarity discrimination threshold also encounter certain difficulties.

Therefore, there is an urgent need for an improved method for qualitative analysis of Raman spectroscopy, which has self-learning ability and can make full use of similarity method, self-learning method and combination with optional artificial identification method to achieve efficient and rapid screening. The spectral processing of the fractions achieves fast convergence and accurate material detection.

Summary of the invention

The purpose of the present disclosure is to address at least one aspect of the above problems and deficiencies existing in the prior art. The embodiments of the present disclosure provide a self-learning qualitative analysis method based on Raman spectroscopy, which combines self-learning and manual comparison to complete Raman spectroscopy, which can reduce false positives due to insufficient material purity in qualitative analysis. And the incidence of underreporting, improve the accuracy of qualitative analysis; shorten the analysis processing time; and shorten the system startup time.

In order to achieve the above object, according to a first aspect of the present disclosure, an embodiment of the present disclosure provides a method for self-learning qualitative analysis based on Raman spectroscopy, comprising: a Raman spectroscopy acquisition step: collecting Raman of an item to be measured Spectral; feature extraction and comparison steps: extracting the Raman spectral data and comparing the spectral signature database in the spectral library to obtain a list of original identification substance IDs; similarity comparison step: obtaining each substance in the list of original identification substance IDs for Raman spectroscopy calculation The similarity of the IDs to generate a similarity list, and compared with the similarity threshold library in the spectral library; and the substance ID selection step: the similarity super similarity threshold obtained after comparing with the similarity threshold based on the self-learning library The similarity identification substance ID list is subjected to verification detection, including false alarm detection and false negative detection. When there is a substance ID in the similarity list exceeding the similarity threshold of the substance ID stored in the similarity threshold library, the false alarm detection is performed. ; when there is no substance ID in the similarity list that exceeds the similarity threshold of the substance ID stored in the similarity threshold library Executive false negative test.

According to an embodiment of the present disclosure, when there is a substance ID exceeding the similarity threshold of the substance ID stored in the similarity threshold library in the similarity list, the false negative detection is additionally performed after the false alarm detection is performed.

According to an embodiment of the present disclosure, any one of the false positive detection and the false negative detection is set to selectively perform three parallel material ID selection methods, including: statistical selection Method: statistically select all false positives or missing material IDs in the self-learning library; feature recognition method: for the false alarms or missing material IDs of the “self-learning type” in the self-learning library The selection of the feature recognition method; and the secondary recognition method: the secondary recognition mode is selected for the false alarm or the missing material ID of the "self-learning type" in the self-learning library.

According to an embodiment of the present disclosure, any one of the false positive detection and the false negative detection is set to include a pre-processing step and a post-processing step, the pre-processing step comprising: by listing the identified substance IDs ID and self-learning library for all false positives or missing material IDs, false positives or missing material IDs for "self-learning types" in the self-learning library, and "for self-learning libraries" The self-learning type "false identification or missing material ID" whose value is "secondary recognition" is respectively compared to generate the highest correct substance ID of the statistical selection method, the feature recognition method, and the secondary recognition method, respectively. And the post-processing step selectively performing the three substance IDs based on a comparison of the highest correct substance ID number of the statistical selection method, the feature identification method, and the secondary identification method with respective number thresholds Method of choosing.

According to an embodiment of the present disclosure, the list of identified substance IDs in the pre-processing step of the false positive detection is selected as the similarity identifying substance ID list.

According to an embodiment of the present disclosure, the list of identified substance IDs in the pre-processing step of the missing report detection is selected as the original identification substance ID list.

According to an embodiment of the present disclosure, a threshold number of times of the highest correct substance ID number obtained for all false positive or missing material IDs in the self-learning library is set to be larger than "self-learning" for the self-learning library The type "value" is a threshold value of the number of times the highest correct substance ID number obtained by the false positive or missing material ID of one of "feature recognition" and "secondary recognition".

According to an embodiment of the present disclosure, when the number of the highest correct substance IDs of the statistical selection method, the feature recognition method, and the secondary recognition method is compared with the respective corresponding number of times thresholds, the condition "the highest correct substance ID number is greater than In the case where the number threshold is established at least twice, the method of satisfying the condition in the three parallel substance ID selection methods is continuously performed selectively to generate the corresponding at least two identification substance ID lists.

According to an embodiment of the present disclosure, if the generated at least two identification substance ID lists are equal, it is confirmed as the list of the identified substance IDs after the verification detection.

According to an embodiment of the present disclosure, if there is an intersection of the generated at least two identification substance ID lists, the intersection is confirmed as the list of the identified substance IDs after the verification detection.

According to an embodiment of the present disclosure, the substance ID selection step is performed again for a portion other than the intersection in the generated at least two identification substance ID lists.

According to an embodiment of the present disclosure, the substance ID selection step performed again includes enhanced detection by using an additive-to-measurement article and an enhancer to obtain an enhanced Raman spectrum.

According to an embodiment of the present disclosure, in the pre-processing step of the false positive detection, the post-processing step of the false positive detection is performed only when the number of statistical false positives is greater than the false positive number threshold.

According to an embodiment of the present disclosure, the method further includes adding the obtained false positive substance ID list and the missing material ID list to the self-learning library according to the “self-learning type” field after performing the qualitative analysis on the item to be measured.

According to one embodiment of the present disclosure, prior to performing qualitative analysis on the item to be measured, the method further includes creating a self-learning library using one of an initial self-learning library that performs initial learning and input presets on the self-learning library.

According to an embodiment of the present disclosure, the method further comprises selectively identifying the substance using a manual comparison method.

According to another aspect of the present disclosure, an embodiment of the present disclosure further provides an electronic device, including: a memory for storing executable instructions; and a processor for executing executable instructions stored in the memory to perform as before Said method.

DRAWINGS

Embodiments of the present disclosure will now be described by way of example only, with reference to the accompanying drawings, in which FIG. A brief description of the drawings is as follows:

1 shows a schematic diagram of a basic process according to an embodiment of the present disclosure, the illustrated components being two phases of a learning phase and an actual detection phase;

2 is a schematic diagram showing the overall flow of an actual detection stage according to an embodiment of the present disclosure as shown in FIG. 1;

3(a) and 3(b) respectively show schematic diagrams of Raman spectra before and after the pretreatment step in the overall flow of the actual detection phase shown in FIG. 2;

4(a) shows an exemplary similarity list obtained in step S31 in the overall flow diagram shown in FIG. 2; FIG. 4(b) shows step S32 in the overall flow diagram shown in FIG. An exemplary threshold library for threshold comparison included in the Raman spectral spectrum library; FIG. 4(c) shows an exemplary excess generated after threshold comparison in step S32 in the overall flow diagram shown in FIG. 2. a list of threshold substances; FIG. 4(d) shows schematic content of an exemplary self-learning library generated by step S10 in the overall flow diagram shown in FIG. 2;

FIG. 5 shows a basic schematic flow chart of false alarm detection in the actual detection phase as shown in FIG. 2;

Figure 6 is a schematic flow chart showing an extension of the "three method election" implementation of false alarm detection in the actual detection phase as shown in Figure 2;

Figure 7 is a schematic flow diagram of an extended exemplary embodiment of false alarm detection as shown in Figure 6;

Figure 8 is a schematic flow chart of another extended exemplary embodiment of false alarm detection as shown in Figure 6;

Figure 9 is a sub-flow diagram of re-false alarm detection performed using enhanced Raman spectroscopy in another extended exemplary embodiment of false alarm detection as shown in Figure 8, showing a re-error as shown in Figure 8. An exemplary decomposition step of reporting detection;

Figure 10 shows a basic schematic flow chart of the false negative detection in the actual detection phase as shown in Figure 2;

Figure 11 is a schematic flow chart showing an extension of the "three method election" implementation of the false negative detection in the actual detection phase as shown in Figure 2;

Figure 12 is a schematic flow chart of an extended exemplary embodiment of the false negative detection shown in Figure 11;

Figure 13 is a schematic flow chart of another extended exemplary embodiment of the false negative detection shown in Figure 11;

Figure 14 is a sub-flow diagram of re-missing detection performed using enhanced Raman spectroscopy in another extended exemplary embodiment of the false negative detection shown in Figure 13, showing the re-leakage as shown in Figure 13 An exemplary decomposition step of reporting detection;

Figure 15 shows an operational schematic of the method of the embodiment of Figure 1 in accordance with the present disclosure;

16 shows still another flow diagram according to an embodiment of the present disclosure, which is illustrated as being divided into Two phases, the learning phase and the actual testing phase, which show the possible detection methods for simultaneous false positives and false negatives;

FIG. 17 is a block diagram showing an example hardware arrangement of an electronic device in accordance with still another embodiment of the present invention.

detailed description

The above described objects, features and advantages of the present disclosure will be more apparent from the following description of the embodiments of the invention. In the specification, the same or similar reference numerals indicate the same or similar parts. The description of the embodiments of the present disclosure is intended to be illustrative of the present invention, and is not to be construed as limiting

In the following detailed description, numerous specific details are set forth Obviously, however, one or more embodiments may be practiced without these specific details. In other instances, well-known structures and devices are shown in the drawings in the drawings.

According to the general idea of the present disclosure, a self-learning qualitative analysis method based on Raman spectroscopy is provided, comprising: a Raman spectroscopy acquisition step: acquiring a Raman spectrum of an item to be measured; a feature extraction and comparison step: extracting Raman spectroscopy data Comparing with the spectral feature library in the spectral library, obtaining a list of original identification substance IDs; similarity comparison step: obtaining the similarity degree of each substance ID in the original identification substance ID list for the Raman spectrum calculation to generate a similarity list, and the spectrum Comparing the similarity threshold library in the gallery; and the substance ID selection step: verifying, based on the self-learning library, the similarity identification substance ID list obtained by comparing the similarity super-similarity threshold with the similarity threshold, including the error Report detection and false negative detection. When there is a substance ID in the similarity list that exceeds the similarity threshold of the substance ID stored in the similarity threshold database, false alarm detection is performed; when there is no similarity threshold library in the similarity list When the substance ID stored in the substance corresponds to the substance ID of the similarity threshold, the false negative detection is performed.

In the following detailed description, numerous specific details are set forth Obviously, however, one or more embodiments may be practiced without these specific details.

FIG. 1 shows a schematic diagram of a basic process according to an embodiment of the present disclosure. Two stages of the learning phase and the actual testing phase.

In the learning phase, the main purpose is to establish a Raman spectral self-learning library for the samples used for actual testing. In the actual detection stage, the self-learning library is used, and the actual sample to be tested is detected by combining artificial contrast Raman spectroscopy to obtain the result of the qualitative analysis.

The above learning phase can also be equivalently considered as a pre-set or calibration phase of the self-learning library, for example typically comprising the steps of measuring the Raman spectrum of the learning sample, such as by extracting its spectral features and comparing it to the spectral feature library; For example, by comparing the spectral features, the similarity list is obtained and compared with the similarity threshold library; whether there is a substance exceeding the threshold exists, and based on the judgment result, (1) if there is more than the listed in the comparison with the threshold library The substance ID of the similarity threshold performs false positive detection (ie, whether there is a substance that is detected by the current similarity threshold exceeding the similarity threshold and is not substantially included in the current learning sample due to a false alarm), the error The report detection selects the false positive substance ID, for example, by comparing with the false positive substance ID or name in the existing self-learning library, and further selectively adopting different self-learning type methods, and (2) if there is no more than The material ID of the similarity threshold is listed, and the false negative detection is performed (ie, it is judged whether there is currently a false report and is substantially included in the current learning). Within the sample but measured as a substance that does not exceed the similarity threshold, the missing detection is compared, for example, by comparison with a missing substance ID or name in an existing self-learning library, and further selectively using different self-learning types The method selects the missing material ID; then optionally determines whether to perform manual comparison and selectively performs manual comparison based on the judgment result; finally, the substance ID such as the correct identification and the correction identification type (ie, false alarm, false negative) The information is entered into the self-learning library as part of the initial preset value of the self-learning library. The above process can be performed separately for one or more learning samples until the Raman spectra of the new learning samples that are no longer needed require acquisition and qualitative detection.

The actual detection phase described above can also be equivalently considered as a stage for qualitative analysis of a test sample based on a generated self-learning library, for example typically comprising the steps of measuring the Raman spectrum of the sample to be measured, such as by extracting its spectral characteristics. And comparing with the spectral feature library; and obtaining the similarity list based on the comparison of the spectral features and comparing with the similarity threshold library; determining whether there is a substance exceeding the threshold, and based on the judgment result, (1) if If there is a substance ID that exceeds the similarity threshold listed in the threshold database, then a false positive detection is performed (ie, it is determined whether there is a false positive in the currently detected substance exceeding the similarity threshold. While not substantially included in the current sample to be measured, the false positive detection is compared, for example, by comparison with a false positive substance ID or name in an existing self-learning library, and further selectively adopting different self-learning types. The method of selecting the false positive substance ID, and (2) if there is no substance ID exceeding the similarity threshold listed therein, performing the false negative detection (ie, determining whether there is currently a false negative report and substantially including the current to be measured a sample within the sample but measured as not exceeding a similarity threshold, the missing detection being compared, for example, by comparison with a missing substance ID or name in an existing self-learning library, and further selectively employing different self-learning types The method selects the missing material ID; then optionally determines whether to perform manual comparison and selectively performs manual comparison based on the determination result; finally displays the identification result of the qualitative analysis; and then the substance ID such as the correct identification and the correction identification type thereof The information (ie, false positives, false negatives) is entered into the self-learning library as part of the initial preset value of the self-learning library.

For conventional Raman spectroscopy methods, if only the measured samples are directly tested and judged based on the original Raman spectroscopy data, in some cases the accuracy of the detection is difficult to guarantee for certain samples such as samples of insufficient purity; And if only the artificial comparison method is used, usually based on the experience of the tester, objective and accurate test results cannot be obtained; and the conventional Raman spectroscopy detection method generates at most the initial calibration sample database for direct comparison, and there is no self. Learning ability to adapt to lack of flexibility when performing qualitative analysis, for example, on substances in mixtures of different components. Moreover, the conventional Raman spectroscopy method generally has a problem that the analysis processing time is long.

It can be seen from the above-described schematic basic flowchart of the embodiment of the present disclosure as shown in FIG. 1 that the self-learning qualitative analysis method based on Raman spectroscopy according to an embodiment of the present disclosure utilizes a combination of self-learning and manual comparison. The measured sample is tested. In this way, the self-learning library is continuously supplemented and perfected, such as learning by using learning samples in a pre-staged learning phase, and the results of qualitative analysis of different samples of the substance to be tested in actual use. Learning to improve the accuracy and efficiency of the recognition results, so that the detection efficiency and detection accuracy of qualitative analysis based on Raman spectroscopy can be optimally optimized, especially in the case of insufficient material purity, which cannot be used by conventional Raman detection methods. In case of direct identification.

As an example, in order to prevent an error in the calibration work due to the learning sample itself, the learning sample may, for example, be selected such that the characteristic peaks in the generated spectrum are clear, the peak position is uniform, A sample of a substance that interferes with small substances. Moreover, it is desirable that the learning samples are selected to have a more uniform peak interval and a certain interval to facilitate more accurate pre-learning. In an embodiment of the present disclosure, the learning sample is, for example, a liquid or solid sample. And, for example, considering that the sample to be actually tested is a mixture of a plurality of substances, the learning sample is selected, for example, as a mixture of a plurality of components whose single component purity is not absolutely superior to be adapted for comparison in a later measurement. Claim.

As an example, the Raman spectrum of the learning sample has, for example, at least four characteristic peaks. The greater number of characteristic peaks is beneficial to the accuracy of the initial learning to improve the accuracy of subsequent qualitative detection operations based on the self-learning library. However, this is not essential, and the learning sample can also have, for example, two or three characteristic peaks.

In the Raman spectroscopy-based self-learning qualitative analysis method according to the present disclosure, on the one hand, an initial self-learning library may be established using representative learning sample items; on the other hand, the above learning phase is not necessary. For example, the operator can perform a qualitative analysis of the sample material to be measured using a self-learning library that is input in advance rather than a newly generated self-learning library. On the other hand, the above-mentioned pre-self-learning phase does not have to be performed for a long time before the actual detection, for example, instead of self-learning while detecting the measured sample substance at the inspection site, the newly added test sample is accumulated during use. Add substances to the self-learning library.

2 shows a general flow diagram of an actual detection phase in accordance with an embodiment of the present disclosure as shown in FIG. In the qualitative analysis of the embodiments of the present disclosure, in order to shorten the analysis processing time and the system startup time, the overall spectral library in the usual Raman detection is subdivided into a plurality of sub-libraries: a spectral feature library, such as by Some basic features such as peak number, peak position, and peak intensity of the graph are extracted to generate the spectral feature library for use in algorithm comparison and identification, and are loaded at software startup; (similarity) threshold library, including recognition spectrum The similarity threshold, material ID, library number and other information are used for display processing and loaded at software startup; the substance name library includes material ID, name, alias and other information for use in software display processing. Thus, the sub-libraries of the respective subdivisions are respectively loaded for comparison at the respective detection steps, and it is not necessary to always load the complete spectral library as a whole or multiple times, thereby shortening the response time of each step and improving the response time. Detection speed.

As shown in FIG. 2, the actual detection phase includes, for example:

Step S0: start;

Step S1: generating a Raman spectrum to be detected and extracting Raman spectrum data;

Step S2: comparing the extracted Raman spectral data with a spectral feature library;

Step S3: using a similarity calculation and a similarity threshold comparison to generate a preliminary determined substance list;

Step S4: Determine whether there is a substance exceeding the threshold?

Step S5: further performing false alarm detection for the case where it is determined that there is a substance exceeding the threshold;

Step S6: further performing false negative detection for the case where it is determined that there is no substance exceeding the threshold;

Step S7: generating a list of substances confirmed by a false positive (or missing report) test;

(Optional) Step S8: Manually comparing the detection of Raman spectroscopy;

Step S9: generating a list of substances for final detection confirmation, and finding a substance name from the substance library;

Step S10: all the test results of the current time are written into the self-learning library;

And step S11: displaying the detection result of the qualitative analysis, and the current detection process is terminated.

As an exemplary embodiment, specifically, the foregoing step S1 specifically includes:

Step S11: collecting a Raman spectrum, which can be obtained, for example, by a known process such as beam emission, collection, and splitting;

Step S12: pre-processing the collected Raman spectrum to obtain a raw Raman spectrum to be tested;

Step S13: extracting spectral data from the original Raman spectrum to be tested.

Since the original spectrum of the sample collected by the Raman spectrometer contains interference information such as fluorescent background, detector (CCD) noise, and transmitter power fluctuations, subsequent alignment and signal processing are affected. Therefore, the measured raw spectral data needs to be preprocessed as shown in step S12 above to facilitate the extraction of subsequent valid information. The pre-processed spectral pre-processing of the above step S12 generally includes interpolation, de-noising, baseline correction, normalization processing, etc., in particular, the main purpose is to perform smooth denoising processing on the input spectrogram signal. Spectral signals before and after pre-processing are shown in Figures 3(a) and 3(b), respectively. In the embodiment of the present disclosure, the collected original spectrum generally needs to be pre-processed, and for brevity, it will not be described below.

Moreover, the above step S3 includes, for example, specifically:

Step S31: calculating a list of acquired similarities;

Step S32: The similarity list is compared with the similarity threshold library, and a substance list exceeding the threshold is acquired.

In the above embodiment, for the sake of clarity, FIG. 4(a) shows an exemplary similarity list obtained in step S31 in the overall flow diagram shown in FIG. 2; FIG. 4(b) shows FIG. An exemplary threshold library for threshold comparison included in the Raman spectral spectrum library in step S32 is shown in the overall flow diagram; FIG. 4(c) shows step S32 in the overall flow diagram shown in FIG. An exemplary over-threshold substance list generated after threshold comparison; FIG. 4(d) shows schematic content of an exemplary self-learning library generated in step S10 in the overall flow diagram shown in FIG. 2.

The qualitative analysis of the Raman spectrum of the sample to be measured is still based on the typical idea of Raman spectroscopy, that is, the comparison with the reference Raman spectrum, that is, the measured Raman spectrum and the reference pull of the sample to be measured Whether the error of the spectroscopy is within a predetermined range, for example, by calculating the similarity between the two. As an example, the calculation of the similarity in the above step S31 is, for example, a plurality of methods, for example, calculating the similarity based on the Euclidean distance algorithm as an industry standard algorithm for spectral search; more specifically, as an example, assuming The reference Raman spectrum curve of the sample that has been studied is A(x), and the measured Raman spectrum curve of the sample to be measured is B(x). In an example, the maximum likelihood algorithm is used, based on the Euclidean distance algorithm. The similarity between the two can be calculated by equation (1):

Corr represents the similarity between the reference Raman spectrum of the sample that has been studied and the measured Raman spectrum of the sample to be measured, and "·" indicates the dot product operation.

In another alternative example, the similarity is calculated in an algorithm similar to that described above, but the average of the spectra is subtracted prior to execution of the algorithm. Specifically, A(x) and B(x) may be sampled separately to obtain n sampling points, respectively denoted as A ₁ , A ₂ , . . . , A _n and B ₁ , B ₂ , . . . , B _{n .} , the similarity of the learned reference Raman spectrum and the measured Raman spectrum of the sample to be measured Corr can be calculated according to formula (2):

Among them, "·" also represents a dot product operation.

In still another alternative example, A(x) and B(x) may also be sampled separately to obtain n sample points, denoted as A ₁ , A ₂ , . . . , A _n and B ₁ , B ₂ , respectively. ..., B _n , the similarity of the learned reference Raman spectrum and the measured Raman spectrum of the sample to be measured Corr can be calculated according to formula (3):

The above similarity calculation may be performed for the entire Raman spectrum, or may be performed only for the portion having the characteristic portion in the Raman spectrum. The closer the similarity value is to 1, the higher the degree of similarity. The above is merely an example of some similarity calculations, and some other similarity calculation methods known to those skilled in the art are also feasible. It is determined whether the error of the measured Raman spectrum of the sample to be measured and the learned reference Raman spectrum is within a predetermined range, and may be determined by the above similarity being greater than a certain threshold. As an example, the threshold of the similarity may be set to 0.9, 0.8, and the like. The similarity threshold is given, for example, by more actual detection sensitivity, accuracy of the detection instrument, and the like.

In the present disclosure, the term "characteristic portion" refers to a key portion of a Raman spectrum curve of a sample to be tested that differs from other samples in a Raman spectrum curve. For example, the feature portion may be one or more feature peaks, feature valleys, phase inflection points, and the like. And, for example, in the case where the Raman spectrum curve of the sample to be measured includes a characteristic peak, the above similarity may be weighted based on the peak position, the peak width, and/or the peak height of the characteristic peak. In an example, the feature peaks may also be searched and sorted prior to calculating the similarity.

The above is merely an example of some similarity calculations, and some other similarity calculation methods known to those skilled in the art are also feasible. For example, in contrast to the above similarity algorithm based on the Euclidean distance of an industry standard algorithm for spectral search, for example, a base is also used instead. The similarity is calculated from the distance of other p values other than the Euclidean distance of p=2 in the formula of the distance. The Euclidean distance formula is as shown in the following formula (4), and when p=2, it is the Euclidean distance.

As a further embodiment, since the Raman spectrum of each substance is a reflection of the molecular structure of the substance, it has unique structural and mode characteristics. By considering the number of spectral data points as the dimension of the model space, a Raman spectral spectrum can be expressed as a pattern vector in the pattern space, and the analysis of the similarity between the N maps is transformed into the computational pattern space. The similarity of N pattern vectors. Correspondingly, the similarity calculation such as the angle cosine method or the Jakedian similarity coefficient method based on the Jachard distance is used, so that the method for calculating the HQI value is simple and fast, and the calculated value is also based on the above-mentioned Euclidean The similarity calculation of the distance algorithm similarly has a fixed interval range between 0 and 1, which is easy to measure. Further, an adjusted cosine similarity algorithm can also be selectively employed.

As an example, supplementally or alternatively, for example, determining whether the error of the Raman spectrum and the reference Raman spectrum of the sample to be measured is within a predetermined range, or directly passing peak intensity detection (amplitude detection) and peak position detection (Phase detection or inflection detection) to extract the information of the characteristic peaks, thereby directly comparing the measured Raman spectrum with the information of the characteristic peaks in the reference Raman spectrum.

In Raman spectroscopy, the Raman spectrum is biased due to the difference in sample uniformity, instrument noise, fluorescence background, etc., and in the spectral processing process, denoising, baseline correction, etc. will also produce errors. The accuracy of the substance recognition using only the similarity in the recognition process is not high. Therefore, in the embodiment of the present disclosure, the object to be inspected is further qualitatively analyzed, for example, by introducing a combination of the self-learning recognition method and the manual contrast recognition method.

Fig. 5 shows a basic schematic flow chart of the false alarm detecting step S5 in the actual detecting phase as shown in Fig. 2. As shown in the figure, in the exemplary example of the present disclosure, in the case where it is determined that there is a substance exceeding the threshold after the comparison with the threshold library, the false alarm detection step S5 is further performed, which is performed. The report detecting step S5 includes two stages: a false positive check pre-processing step S50, S50' and S50"; and a false positive detection post-processing step S51.

In one aspect, as an illustrative example of the present disclosure, for example, as shown in FIG. 5, the false positive detection pre-processing steps S50, S50', and S50" are three logically parallel sub-flows, respectively corresponding to subsequent post-processing steps. The nth (n=1, 2, 3) substance ID selector to be used in S51 Method: S50 corresponds to the first substance ID selection method, that is, the statistical method is used to verify one by one, which is also called “statistical selection” method; S50′ corresponds to the second substance ID selection method, that is, the preset “features are called” Corresponding algorithm for identifying the interface to select the verified substance ID, also referred to as the "feature recognition" method; and S50" corresponds to the third substance ID selection method, that is, calling the corresponding algorithm of the preset "secondary identification interface" The verified substance ID is also referred to as a "secondary recognition" method. Accordingly, based on the characteristics of the respective material ID selection methods to be used hereinafter, S50 is also referred to as a pre-processing step of "statistical selection", and S50' is also called For the pre-processing step of "feature recognition", S50" is also referred to as the pre-processing step of "secondary recognition". The above-described three pre-processing steps S50, S50' and S50" are logically parallel to mean performing independently of each other, for example, substantially simultaneously, or sequentially, or temporally independent of each other.

Specifically, as shown in FIG. 5, the pre-reporting pre-processing step, that is, the pre-processing step S50 of "statistical selection", the pre-processing step S50' of "feature recognition", and the pre-processing step S50 of "secondary recognition" "For example:

Step S500, S500', S500": The false positive check subroutine starts.

Step S501, S501', S501": the substance IDs in the list of identification substance IDs (hereinafter referred to as "threshold identification list") of the similarity super-threshold acquired after the threshold comparison are sequentially and (in the corresponding/or corresponding) in the self-learning library A single) "false positive substance ID" field is compared.

Here, specifically, as shown in FIG. 5, for example, step S501 is to sequentially compare the IDs in the threshold-valued threshold identification list with the "false positive substance ID" field in the entire self-learning library; step S501 'Comparing the IDs in the threshold-identified threshold identification list with the "false positive substance ID" field in the case where the "self-learning type" field in the self-learning library takes the value of "feature identification"; and the steps S501" is to compare the IDs in the threshold-identified threshold identification list with the "false positive substance ID" field in the case where the "self-learning type" field in the self-learning library takes the value of "secondary recognition";

Steps S502, S502', S502": determine whether the same ID is matched (ie, is it recognized that the false positive substance ID exists?).

Steps S503, S503', S503": If the same substance ID is matched, it is equivalent to finding a false positive substance ID, and the false alarm count counter is incremented by one.

Steps S504, S504', S504": If the same substance ID is not matched, the current substance ID is not a false alarm but is actually considered to exist, and the correct substance ID number is counted. Add 1 to the device.

Steps S505, S505', S505": determining whether the comparison of the identification substance ID list is completed. If the comparison is not completed, the process proceeds to step S501, S501', S501" is executed cyclically; if the comparison is completed, the process proceeds to the next step S506, S506', S506. ".

Step S506, S506', S506": determining whether the number of false positives is greater than 10. If the number of false positives is less than or equal to 10, it is considered that the number of false positives is insufficient to ensure the smooth progress of the self-learning detection, thereby jumping to manual contrast recognition; If the number is greater than 10, the assignment step of the "Maximum correct substance ID times" field is entered.

Here, the number of false positives is set to 10 is an empirical value. When it is confirmed that the number of false positives exceeds the value, it is determined that the number of false positives generated is sufficient to generate a sufficiently large set of substance IDs to be verified, Subsequent post-processing step S51 performs material ID selection. Specifically, the above three types of false positive check processing steps, that is, the "statistical selection" step S50, the "feature recognition" step S50', and the "secondary recognition" step S50" respectively correspond to the nth type used in the post-processing step ( n=1, 2, 3) material ID selection method: the first substance ID selection method is selected by statistical verification one by one; the second substance ID selection method is to call a corresponding algorithm of the preset “feature recognition interface” The verified substance ID is selected; and the third substance ID selection method is to select the verified substance ID by calling a corresponding algorithm of the preset "secondary identification interface".

Steps S507, S507', S507": assigning respective current "correct substance ID times counters" to the corresponding "highest correct substance ID times" field MaxRightIDNum(n), respectively, as a post-processing step S51 to determine whether or not to perform subsequent correspondence. The criterion for the nth substance ID selection method.

On the other hand, as an illustrative example of the present disclosure, as shown in FIG. 5, the post-false alarm detection post-processing step S51 includes, for example:

S511: It is judged that for the above three component flows S50, S50' and S50", the comparison formula "field MaxRightIDNum(n)> corresponding threshold THR(n)? Whether it is established for at least two groups. This judgment is a criterion for dividing whether the highest correct substance ID number is sufficient to ensure the execution of the corresponding substance ID selection method, and if satisfied, at least two substance ID selection methods are available for acquiring at least two The group material ID list is used to jointly verify the existence of the substance ID that can be identified in a program-controlled manner. Conversely, if for the above three-component processes S50, S50' and S50", the comparison formula is not true or only one group If it is established, it means that it is impossible to pass at least two of the above. The list of substance IDs identified by the substance ID selection method is subjected to election for qualitative analysis, so that substantially the self-learning process continues to be meaningless, the operation is terminated and the manual comparison is recognized.

S512: In the case where the formula "field MaxRightIDNum(n)>the corresponding threshold value THR(n)?" is established, the respective corresponding substance list IDn (for example, ID1 or ID2 or ID3) is acquired by the nth method, respectively.

S513: It is judged whether at least two of the generated substance lists IDn (that is, confirmed by the false positive test) are identical. If YES, the subsequent step S514 is continued. If "None", then the self-learning process continues to be meaningless, the operation is terminated and the manual comparison is recognized.

S514: The same at least two substance lists are used as a list of identification substances that are respectively recognized and jointly confirmed by the corresponding at least two substance ID selection methods.

Wherein, for the above step S511, the respective thresholds THR(n) of the field MaxRightIDNum(n) are respectively set, for example, as thresholds for the "statistical selection" method, the "feature recognition" method, and the "secondary recognition" method, respectively. The first threshold value THR(1), the second threshold value THR(2), and the third threshold value THR(3), respectively. Since the “feature recognition” method is a dimensionality reduction method used in pattern recognition to remove uncorrelated or redundant features from the original feature set, the “secondary recognition” method is used to estimate the mean and covariance matrix, for example, after feature extraction. The classifier is trained to be classified and identified, so that the two can achieve the purpose of reducing the number of features, improving the detection accuracy, and reducing the running time; and the "statistical selection" method is inconsistently comparing and confirming one by one, thereby "statistics" The reliability of the selection method is smaller than the "feature recognition" method or the "secondary recognition" method using pattern recognition, and accordingly, the first threshold THR(1) is set to be compared to the second threshold THR (2). And the third threshold THR(3) is larger. For example, in the embodiment of the present disclosure, THR(n) is set to THR(1)=10, THR(2)=5, and THR(3)=6, respectively.

For the above step S512, on the one hand, in an exemplary example of the present disclosure, for example, the "feature recognition" method is a dimensionality reduction method in pattern recognition for rejecting irrelevant or redundant features from the original feature set, for example in The embodiment of the present disclosure is implemented by calling a plurality of feature recognition interfaces preset in the “feature identification interface” field of the self-learning library, and may be selected as at least one of the following:

Filter/Filter, which characterizes the importance of each feature by selecting an indicator, and then sorts the features based on the index values of the features, such as by setting thresholds and removing them Feature selection is not performed by the characteristics of the threshold, or by setting the number of features to be selected and selecting the top N or sorting to a certain percentage of the top. In other words, by weighting the features of each dimension, the weights represent the importance of the dimension features and are then sorted by weight. The usual filtering method uses the characteristics of the training set to screen out the feature subsets. Generally, the independence of the features or the relationship with the dependent variables, such as chi-square test, information gain, correlation coefficient, etc., are considered.

Wrapper, which selects several grouping features for a training set each time, or excludes several grouping features, based on an objective function (usually an evaluation of the predicted effect). In other words, the parcel/encapsulation method essentially considers the selection of feature subsets as a search optimization problem, and generates different combinations (feature subsets) by packaging, and evaluates the combinations and compares them with other combinations, for example. The accuracy of the classification is used as a measure of how good or bad the feature subset is. Therefore, the selection of subsets is regarded as an optimization problem, for example, it can be solved by many optimization algorithms, especially heuristic optimization algorithms, such as genetic algorithm, particle swarm optimization algorithm, differential evolution algorithm, artificial bee colony algorithm and so on. Parcel/encapsulation methods such as recursive feature elimination algorithms.

Embedded: It uses some machine learning algorithms and models to train, obtains the weight coefficients of each feature, and then selects features according to the weight coefficients from large to small. Similar to the Filter method, but through training to determine the pros and cons of the feature, that is, to learn the best attributes to improve the accuracy of the model in the case of the model. Specifically, in the process of establishing the model, it is important to select the characteristics that are important for the training of the model (for example, the greatest contribution to improving the accuracy). The most common Embedded methods are the regularization methods.

On the other hand, in an exemplary example of the present disclosure, for example, the "secondary recognition" method is implemented, for example, by calling a plurality of secondary recognition interfaces preset in a "feature recognition interface" field of the self-learning library, and For example, it is constructed in such a manner as to use a quadratic discriminant equation QDF classifier commonly used in pattern recognition, an MQDF improved quadratic discriminant equation classifier, etc., and the classifier is trained by estimating the mean and covariance matrix, and the covariance matrix reflects the feature. The spread between the two, the greater the covariance, the more information is included, the more accurate the final classification.

Thus, in an exemplary embodiment of the present disclosure, as described above, for example, the values of the respective thresholds are respectively set to THR(1)=10, THR(2)=5, and THR(3)=6. On the premise, if the field MaxRightIDNum(1)>10 is established, the obtained statistical selection substance list ID1 is selected statistically from the "false positive substance ID" field of the entire self-learning library; if the field MaxRightIDNum(2)>5 Once established, the feature recognition interface is called to obtain the feature recognition substance. List ID2; if the field MaxRightIDNum(3)>6 is established, the secondary recognition interface is called to obtain the secondary identification substance list ID3.

For the above step S514, the substance identification verification is performed independently by using at least two sets of substance ID selection methods, and then the confirmed substance ID list is compared. Once the same, it means that based on the similarity judgment, further A list of identified substance IDs is co-confirmed using at least two independent methods, thereby obtaining a more accurate self-learning substance identification ID than conventional Raman spectroscopy based only on similarity judgment and manually performed Raman spectroscopy. List.

In the exemplary embodiment of the present disclosure, after the above-described false alarm detection post-processing step S51, the jump to S7 generates a substance list confirmed by the false positive check.

Similarly, the missing report detection S6 is discussed. Fig. 10 shows a basic schematic flow chart of the false negative detection in the actual detection phase as shown in Fig. 2. As shown in the figure, in the exemplary example of the present disclosure, for the case where it is determined that there is no substance exceeding the threshold after the comparison with the threshold library, the missing report detection step 71 is further performed. The missing report detecting step S6 includes two stages: a missing pre-test pre-processing step S60, S60' and S60"; and a missing-post detection post-processing step S61.

In one aspect, as an illustrative example of the present disclosure, for example, as shown in FIG. 10, the pre-missing detection pre-processing steps S60, S60', and S60" are three logically parallel sub-flows, respectively corresponding to subsequent post-processing steps. The nth (n=1, 2, 3) substance ID selection method to be used in S61: S60 corresponds to the first type, that is, the aforementioned "statistical selection" method; and S60' corresponds to the second type, that is, the aforementioned "feature recognition" The method; and S60" corresponds to the third, ie, the aforementioned "secondary recognition" method. Accordingly, based on the characteristics of the material ID selection method to be used in each of the following, S60 is also referred to as a pre-processing step of "statistical selection", and S60' is also referred to as a pre-processing step of "feature recognition", and S60" is also referred to as " Pre-processing steps of secondary recognition. The above three pre-processing steps S60, S60' and S60" are logically parallel to mean that they are executed independently of each other, for example, substantially simultaneously, or sequentially, or temporally independent of each other in time. Execution.

Specifically, as shown in FIG. 10, it is substantially similar to the foregoing pre-false positive check processing steps S50, S50' and S50", the pre-reporting pre-processing step, that is, the pre-processing step S60 of "statistical selection", "feature recognition" The pre-processing step S60' and the "secondary recognition" pre-processing step S60" include, for example:

Steps S600, S600', S600": The underreporting test subroutine begins.

Steps S601, S601', S601": The substance IDs in the original identification substance ID list are sequentially compared with the (whole/or corresponding single) "missing substance ID" field in the self-learning library.

Here, specifically, as shown in FIG. 10, for example, step S601 is to sequentially compare the IDs in the original identification substance ID list with the "false negative substance ID" field in the entire self-learning library; step S601' is The IDs in the original identification substance ID list are sequentially compared with the "false negative substance ID" field in the case where the "self-learning type" field in the self-learning library takes the value of "feature recognition"; and step S601" is the original identification substance The IDs in the ID list are sequentially compared with the "Reporting Substance ID" field in the case where the "Self-learning Type" field in the self-learning library takes the value of "Secondary Recognition";

Step S602, S602', S602": It is judged whether or not the same ID is matched (ie, is it recognized that the missing substance ID exists?).

Steps S603, S603', S603": If the same substance ID is matched, it is equivalent to finding the missing substance ID once, and the counter of the correct substance ID number (here, equivalent to the number of missing substance IDs) is incremented by one.

Steps S604, S604', S604": determining whether the comparison of the identification substance ID list is completed. If the comparison is not completed, the process proceeds to step S601, S601', S601" is cyclically executed; if the comparison is completed, the process proceeds to the next step S605, S605', S605. ".

Steps S605, S605', S605": assigning respective current "correct substance ID times counters" to the corresponding "highest correct substance ID times" field MaxRightIDNum(n), respectively, as a post-processing step S61 to determine whether or not to perform subsequent correspondence. The criterion for the nth substance ID selection method.

On the other hand, as an exemplary example of the present disclosure, as shown in FIG. 10, substantially similar to the aforementioned false positive detection post-processing step S51 shown in FIG. 5, the missing report detection post-processing step S61 includes, for example:

S611: It is judged that for the above three component flows S60, S60' and S60", the comparison formula "field MaxRightIDNum(n)> corresponding threshold value THR(n)'? Whether it is established for at least two groups. This judgment is a criterion for dividing whether the highest correct substance ID number is sufficient to ensure the execution of the corresponding substance ID selection method, and if satisfied, at least two substance ID selection methods are available for acquiring at least two Group material ID list to jointly verify substances that can be identified by program control The existence of the ID. On the other hand, if the comparison formula is not true for the above three-component processes S60, S60' and S60", or only one group is established, it means that the substance cannot be identified by the above at least two substance ID selection methods. The ID list is elected for qualitative analysis, so that essentially the self-learning process continues to be meaningless, then the operation is terminated and jumped to manual comparison recognition.

S612: In the case where the formula "field MaxRightIDNum(n)> corresponding threshold THR(n)'?" is established, respectively obtain the respective substance list IDn' (for example, ID1' or ID2' or ID3') by the nth method. .

S613: It is judged whether at least two of the generated substance lists IDn' (that is, confirmed by the false negative test) are identical. If "Yes", the subsequent step S614 is continued. If "None", then the self-learning process continues to be meaningless, the operation is terminated and the manual comparison is recognized.

S614: The same at least two substance lists are used as a list of identification substances that are respectively recognized and jointly confirmed by the corresponding at least two substance ID selection methods.

Wherein, for the above step S611, the selection and setting of the corresponding threshold THR(n)' of the field MaxRightIDNum(n) is the same as or similar to the false positive detection. For example, the first threshold THR(1)' is set to be larger than the second threshold THR(2) and the third threshold THR(3). For example, in the embodiment of the present disclosure, THR(n)' is set to THR(1)'=10, THR(2)'=5, and THR(3)'=6, respectively. And the "feature recognition" method and the "secondary recognition" method are also the same or similar, and are respectively executed by calling a plurality of different "feature recognition interfaces" and a plurality of "secondary recognition interfaces".

Thus, in an exemplary embodiment of the present disclosure, as described above, for example, the values of the respective thresholds are respectively set to THR(1)'=10, THR(2)'=5, THR(3), respectively. Under the premise of '=6, if the field MaxRightIDNum(1)'>10 is established, the obtained statistical selection substance list ID1' is selected statistically from the "missing substance ID" field of the entire self-learning library; if the field MaxRightIDNum (2) When '5' is established, the feature recognition interface is called to obtain the feature recognition substance list ID2'; if the field MaxRightIDNum(3)'>6 is established, the secondary recognition interface is called to obtain the secondary identification substance list ID3'.

For the above step S614, at least two sets of substance ID selection methods are used to independently perform substance identification verification, and then the confirmed substance ID list is compared. Once the same, it means that based on the similarity judgment, further Use at least two separate methods The list of identification substance IDs is collectively confirmed, thereby obtaining a more accurate list of self-learning substance identification IDs than conventional Raman spectroscopy tests based on similarity judgment and manually performed Raman spectroscopy.

In the exemplary embodiment of the present disclosure, after the above-described missing report detection processing step S61, the jump to S7 generates a substance list confirmed by the false negative check.

For the sake of illustration, FIG. 15 shows a schematic diagram of an operation for detecting a Raman spectrum of a test sample using a method according to an embodiment of the present disclosure. The main processes in this example include:

1) After preparing the sample, collect the data;

2) calling the algorithm interface to perform spectral preprocessing and extracting spectral feature data;

3) compared with the spectral feature library;

4) Obtain a similarity list, as shown in Figure 2;

5) compared with the threshold library, as shown in Figure 3;

6) Obtain a list of substances exceeding the threshold, as shown in Figure 4;

7) Is there a substance exceeding the threshold? If it is "No", jump to 14);

8) If it is "Yes", find in the self-learning library whether there is a false positive substance ID, if it is "No", jump to 21);

9) If it is "Yes", call the "statistical selection" algorithm to select the false positive substance ID;

10) Calling the "feature selection" algorithm to select the false positive substance ID;

11) Calling the "secondary recognition" algorithm to select the false positive substance ID;

12) Calling the "three scheme election determination results" algorithm to select the final possible correct material ID;

13) Find the substance name from the spectrum library according to the substance ID and jump to 21);

14) Is there a missing material ID in the self-learning library? If it is "No", jump to 21);

15) If it is "Yes", call the "statistical selection" algorithm to select the missing material ID;

16) Calling the "feature selection" algorithm to select the missing material ID;

17) Calling the "secondary recognition" algorithm to select the missing material ID;

18) Calling the "three scheme election determination results" algorithm to select the final possible correct material ID;

19) "Is there a missing material ID?" If it is "No", jump to 21);

20) If it is "Yes", find the name of the substance from the spectrum library;

21) Display the measurement results;

22) Is there a “manual comparison”? If you select "No", jump to 26);

23) If it is “Yes”, list the results of all spectra of the sample spectrum and the spectrum library, including similarity, number of peaks, peak position, peak intensity, etc., for manual analysis, screening and judgment;

24) "Is there a false positive or missing report?", if it is "No", jump to 26);

25) If it is "Yes", select the correct substance, selection type and other information to be written into the self-learning library for analysis and self-learning;

26) End.

Similarly, in other embodiments, numerous modifications and variations are possible based on the preferred embodiments described above.

Figure 6 shows a schematic flow chart of an extension of the "three method elections" implementation of false positive detection in the actual detection phase as shown in Figure 2. The difference between the false positive detection flow S5 in the example of FIG. 6 and the false positive detection flow S5 in the example of FIG. 5 mainly lies in, as shown in FIG. 6, for example, based on "at least two (by various substance IDs) After the selection method separately identifies the identified substance ID list, the false positive detection post-processing step S51 additionally includes an optional step S515, that is, a further "three-method election" based on "intersection". For the sake of brevity, the remaining sub-steps will not be described again.

Further, FIG. 7 is a schematic flow chart of a substantially extended exemplary embodiment of false alarm detection as shown in FIG. 6. The difference between the false positive detection flow S5 in the example of FIG. 7 and the false positive detection flow S5 in the example of FIG. 15 is mainly that, as shown in FIG. 7, for example, the optional step of the post-false positive detection processing step S51 S515 specifically includes:

Step S5150: It is judged that there is an intersection of at least two of the generated substance lists ID1, ID2, and ID3. If yes, proceed to step S5150, that is, there is an overlap portion of the list of substance IDs respectively selected by using at least two independent methods, and the overlapping portion can be used to generate a list of commonly recognized identification substance IDs; otherwise, jump Go to manual contrast recognition.

Step S5151: In the case where step S5150 is established, the intersection is assigned to the first identification list.

Thereafter, the first identification list is directly used as a list of substances confirmed after the false positive check in the subsequent step S7.

The extended flowchart of the false positive detection S5 shown in FIG. 7 further utilizes at least two independent methods after jointly identifying the identification substance ID list based on the similarity recognition and the determination of the same result using at least two independent methods. The overlapping portion of the results, that is, the judgment of the intersection, together confirms the list of identification substance IDs, ensuring that the recognition accuracy is further improved.

Figure 8 is a schematic flow diagram of another further expanded exemplary embodiment of false alarm detection as shown in Figure 6. The difference between the false positive detection flow S5 in the example of FIG. 8 and the false positive detection flow S5 in the example of FIG. 7 is mainly that, as shown in FIG. 8, for example, the optional step of the false positive detection post-processing step S51 S515 additionally includes a list of substance IDs selected for each of at least two independent methods, in addition to confirming the intersection portion, further verifying the non-intersection portion. For example, after steps S5150 and S5151, the optional step S515 of the post-false positive detection processing step S51 additionally includes:

S5152: Subtract the intersection of the at least two substance lists ID1, ID2, and ID3 to obtain a list of substances to be rechecked.

S5153: The list of substances to be rechecked is again subjected to enhanced false positive detection.

S5154: It is judged whether there is a newly confirmed substance list generation after re-incrementing the false alarm detection. If yes, proceed to step S5155, otherwise, go to step S5156.

S5155: Generate a re-recognition list.

S5156: The re-recognition list is assigned the value NONE.

S5157: Assign the re-identification list to the second recognition list.

S5158: The first and second identification lists are combined to generate a list of identification substances.

Wherein, as described in the above sub-steps, the false positive detection S5 of FIG. 8 is substantially based on the example shown in FIG. 7, and is substantially a portion of the "additions other than the intersection" that cannot be confirmed after the "intersection judgment". Perform further analysis and verification. The specific steps are explained in detail below. For example, FIG. 9 is a sub-flow diagram of re-false alarm detection S5153 performed using enhanced Raman spectroscopy in another extended exemplary embodiment of false alarm detection as shown in FIG. The re-false alarm detects an exemplary decomposition step of S5153.

In an exemplary embodiment of the present disclosure, as shown in FIG. 9, for the complement portion other than the intersection, the re-false alarm detection S5153 includes, for example:

S51531: Acquire enhanced Raman spectroscopy by mixing the sample to be tested and the enhancer.

S51532: Perform false alarm detection. Specifically, for example, based on enhanced Raman spectroscopy, embedded The sleeve utilizes the aforementioned step S5.

S51533: (for example, human confirmation) to determine whether to jump to manual comparison.

S51534: Jump to manual comparison for false positive detection.

S51535: Generate a list of substances that are confirmed to be present by re-execution of false alarm detection using enhanced Raman spectroscopy.

As an example, in the above step S51531, when detecting by using the enhanced Raman spectroscopy data of the sample of the measured substance, the mixture of the sample to be tested and the enhancer may be directly mixed by the sample to be tested and the enhancer or by the sample to be tested. The aqueous solution or organic solution is mixed with the reinforcing agent. Similarly, the mixture of the measured substance sample and the enhancer is formed by directly mixing the sample of the measured substance with the enhancer or by mixing an aqueous solution of the sample of the test substance or an organic solution with the enhancer. As an example, the enhancer may comprise any one of metal nanoparticle materials, metal nanowires, metal nanoclusters, carbon nanotubes, and carbon nanoparticles, or a combination thereof. In another example, the enhancer may comprise a metal nanomaterial, or may also contain a chloride nanoparticle, a bromide ion, a sodium ion, a potassium ion, or a sulfate ion. The metal may include, for example, any one of gold, silver, copper, magnesium, aluminum, iron, cobalt, nickel, palladium, or platinum, or a combination thereof. In the mixture of the measured substance sample and the enhancer, the molecules of the measured substance sample adhere to the surface of the enhancer material, and the electromagnetic field on the surface of the enhancer material enhances the Raman spectrum signal of the sample of the measured substance.

Similarly, in other exemplary embodiments of the present disclosure, a variation regarding the false negative detection S6 is also obtained.

Figure 11 shows a schematic flow chart of an extension of the "three method elections" implementation of false negative detection in the actual detection phase as shown in Figure 2. The difference between the missing report detection flow S6 in the example of FIG. 11 and the missing report detection flow S6 in the preferred embodiment of FIG. 10 is mainly as shown in FIG. 11, for example, based on "at least two (by various After the substance ID selection method separately identifies the identified substance ID list, the missing report detection processing step S61 additionally includes an optional step S615, that is, a further "three method election" based on "intersection". For the sake of brevity, the remaining sub-steps will not be described again.

Further, FIG. 12 is a schematic flow chart of a substantially expanded exemplary embodiment of the false negative detection shown in FIG. The difference between the missing report detection flow S6 in the example of FIG. 12 and the missing report detection flow S6 in the example of FIG. 15 mainly lies in, as shown in FIG. 12, for example, The optional step S615 of the post-report detection post-processing step S61 specifically includes:

Step S6150: It is judged that there is an intersection of at least two of the generated substance lists ID1', ID2', ID3'? If yes, proceed to step S6150, that is, there is an overlapped portion of the list of substance IDs respectively selected by using at least two independent methods, and the overlapping portion can be used to generate a list of commonly recognized identification substance IDs; otherwise, jump Go to manual contrast recognition.

Step S6151: In the case where step S6150 is established, the intersection is assigned to the first identification list.

Thereafter, the first identification list is directly used as a list of substances confirmed by the missing report test in the subsequent step S7.

The extended flowchart of the false negative detection S6 shown in FIG. 12 further utilizes at least two independent methods after jointly identifying the identification substance ID list based on the similarity recognition and the determination of the same result using at least two independent methods. The overlapping portion of the results, that is, the judgment of the intersection, together confirms the list of identification substance IDs, ensuring that the recognition accuracy is further improved.

FIG. 13 is a schematic flow chart of another further extended exemplary embodiment of the false negative detection shown in FIG. The difference between the missing report detection flow S6 in the example of FIG. 13 and the missing report detection flow S6 in the example of FIG. 12 is mainly that, as shown in FIG. 13, for example, the optional step of the missing report detection post-processing step S61 S615 additionally includes a list of substance IDs selected for each of the at least two independent methods, and further verifying the non-intersection portion in addition to confirming the intersection portion. For example, after steps S6150 and S6151, the optional step S615 of the post-report detection post-processing step S61 additionally includes:

S6152: Subtract the intersection of the at least two substance lists ID1, ID2, and ID3 to obtain a list of substances to be re-examined.

S6153: The list of substances to be rechecked is again subjected to enhanced false negative detection.

S6154: It is judged whether or not a newly confirmed substance list is generated after the enhanced false negative detection is performed again. If yes, proceed to step S6155, otherwise, go to step S6156.

S6155: Generate a re-recognition list.

S6156: The re-recognition list is assigned the value NONE.

S6157: Assign the re-identification list to the second recognition list.

S6158: The first and second identification lists are combined to generate a list of identification substances.

Wherein, as described in the above sub-steps, the missing report detection S6 of FIG. 13 is substantially on the basis of the example shown in FIG. 12, and is substantially a portion of the "additions other than the intersection" that cannot be confirmed after the "intersection judgment". Perform further analysis and verification. The specific steps are explained in detail below. For example, FIG. 14 is a sub-flowchart of re-false negative detection S6153 performed using enhanced Raman spectroscopy in another extended exemplary embodiment of the false negative detection shown in FIG. 13, which is shown in FIG. The re-missing detection detects an exemplary decomposition step of S6153.

In an exemplary embodiment of the present disclosure, as shown in FIG. 14, for the complement portion other than the intersection, the re-missing detection S6153 includes, for example:

S61531: Using the sample to be tested and the enhancer to obtain an enhanced Raman spectrum.

S61532: Perform a false negative detection. Specifically, for example, based on the enhanced Raman spectrum, the above-described step S6 is nested.

S61533: (for example, human confirmation) to determine whether to jump to manual comparison.

S61534: Jump to manual comparison for false negative detection.

S61535: Generate a list of substances that are confirmed to be present by re-executing the false negative detection using the enhanced Raman spectrum.

The above specific operation flow has strict logic and can avoid the abnormal operation of the user. For the purposes of the present disclosure, the self-learning described above is also replaced, for example, by using a self-learning mixture analysis method.

In another embodiment of the present disclosure, for example, in reality, in step S4, that is, whether or not there is a substance exceeding a threshold value, there is a possibility that although it is confirmed that there is a substance exceeding a threshold value, It is not excluded that there is still a possibility of underreporting the actual substance. In particular, FIG. 16 shows a further flow diagram in accordance with an embodiment of the present disclosure, illustrated as being divided into two phases, a learning phase and an actual detection phase, in which a detection manner regarding the simultaneous presence of false positives and false negatives is shown.

As schematically shown in Fig. 16, if it is judged that there is no substance exceeding the threshold, as described in the foregoing embodiment, only the false negative detection is performed. However, if it is judged that there is a substance exceeding the threshold, for example, false positive detection and false negative detection are sequentially performed. This allows a more comprehensive qualitative identification of the substance ID.

In accordance with yet another embodiment of the present disclosure, an electronic device is also provided, and FIG. 17 is a block diagram showing an example hardware arrangement 100 of the electronic device. The hardware arrangement 100 includes a processor 106 (eg, a microprocessor (μP), a digital signal processor (DSP), etc.). Processor 106 may be a single processing unit or a plurality of processing units for performing different acts of the method steps described herein. The arrangement 100 may also include an input unit 102 for receiving signals from other entities, and an output unit 104 for providing signals to other entities. Input unit 102 and output unit 104 may be arranged as a single entity or as separate entities.

Moreover, arrangement 100 can include at least one readable storage medium 108 in the form of a non-volatile or volatile memory, such as an electrically erasable programmable read only memory (EEPROM), flash memory, and/or a hard drive. The readable storage medium 108 includes a computer program 110 that includes code/computer readable instructions that, when executed by the processor 106 in the arrangement 100, cause the hardware arrangement 100 and/or the device including the hardware arrangement 100 to The flow described above in connection with the above embodiments and any variations thereof are performed.

Computer program 110 can be configured as computer program code having a computer program module 110A-110C architecture, for example. Thus, in an example embodiment when a hardware arrangement 100 is used, for example, in a device, the code in the computer program of arrangement 100 includes a plurality of modules, including but not limited to, for example, illustrated modules 110A, 110B, and 110C, the plurality of modules Respectively configured to perform different determinations or operational steps, such as any of the processes, sub-processes, sub-processes, and/or steps performed in the previous Figures 1-2, and 5-16 .

The computer program module can substantially perform the various actions in the flow described in the above embodiments to simulate the device. In other words, when different computer program modules are executed in processor 106, they may correspond to the different units described above in the device.

Although the code means in the embodiment disclosed above in connection with FIG. 17 is implemented as a computer program module that, when executed in processor 106, causes hardware arrangement 100 to perform the actions described above in connection with the above-described embodiments, in alternative embodiments At least one of the code means can be implemented at least partially as a hardware circuit.

The processor may be a single CPU (Central Processing Unit), but may also include two or more processing units. For example, a processor can include a general purpose microprocessor, an instruction set processor, and/or a related chipset and/or a special purpose microprocessor (eg, an application specific integrated circuit (ASIC)). At The processor can also include an onboard memory for caching purposes. The computer program can be carried by a computer program product connected to the processor. The computer program product can comprise a computer readable medium having stored thereon a computer program. For example, the computer program product can be flash memory, random access memory (RAM), read only memory (ROM), EEPROM, and the computer program modules described above can be distributed to different computers in the form of memory within the UE in alternative embodiments. In the program product.

The present disclosure has at least the following advantages: it can make full use of the similarity method, the self-learning method, and the combination with the optional manual recognition method to achieve efficient and rapid spectral processing of substance recognition.

The various embodiments in the present specification are described in a progressive manner, and each embodiment focuses on differences from other embodiments, and the same similar parts between the embodiments can be referred to each other. I will not repeat them here.

It will be understood by those skilled in the art that the embodiments described above are exemplary and can be modified by those skilled in the art, and the structures described in the various embodiments do not conflict in structure or principle. In the case of free combination.

The present disclosure has been described with reference to the accompanying drawings, which are intended to be illustrative of the preferred embodiments of the present invention and are not to be construed as limiting. The size ratios in the drawings are merely illustrative and are not to be construed as limiting the disclosure.

While some embodiments of the present general inventive concept have been shown and described, it will be understood by those of ordinary skill in the art The scope is defined by the claims and their equivalents.

Claims (17)

1. A method for self-learning qualitative analysis based on Raman spectroscopy, comprising:

   Raman spectroscopy acquisition step: collecting Raman spectra of the items to be tested;

   Feature extraction and comparison steps: extracting Raman spectral data and comparing the spectral feature database in the spectral library to obtain a list of original identification substance IDs;

   Similarity comparison step: obtaining a similarity list for each substance ID in the original identification substance ID list for Raman spectroscopy calculation, and generating a similarity list, and comparing with the similarity threshold library in the spectral library;

   Substance ID selection step: performing verification detection on the similarity identification substance ID list obtained by comparing the similarity super-similarity threshold with the similarity threshold based on the self-learning library, including false positive detection and false negative detection, when the similarity list When there is a substance ID exceeding the similarity threshold of the substance ID stored in the similarity threshold library, the false positive detection is performed; when there is no similarity threshold in the similarity list that exceeds the substance ID stored in the similarity threshold library When the substance ID is used, the false negative detection is performed.
2. The method according to claim 1, wherein when there is a substance ID in the similarity list that exceeds the similarity threshold of the substance ID stored in the similarity threshold library, the false negative detection is performed after the false alarm detection is performed first. .
3. The method of claim 1 wherein any one of said false positive detection and said false negative detection is arranged to selectively perform three parallel material ID selection methods, comprising:

   Statistical selection method: statistical selection of all false positives or missing material IDs in the self-learning library;

   Feature recognition method: selecting a feature recognition method for a false positive or missing material ID of a "self-learning type" in the self-learning library; and

   Secondary recognition method: For the self-learning library, the "self-learning type" takes the value of "secondary recognition" and the false positive or missing material ID is selected twice.
4. The method of claim 3 wherein said false positive detection and said false negative Any one of the tests is set to include a pre-processing step and a post-processing step,

   The pre-processing step includes: by using the ID in the list of identified substance IDs and the self-learning library for all false positives or missing material IDs, and for the "self-learning type" in the self-learning library, the value is "feature identification" The false positive or false negative substance ID, and the false positive or false negative substance IDs whose "self-learning type" values in the self-learning library are "secondary identification" are respectively compared to generate the statistical selection method, the The feature identification method and the highest correct substance ID number of the secondary recognition method;

   The post-processing step selectively performs the three substance ID selection methods based on a comparison of the highest correct substance ID number of the statistical selection method, the feature identification method, and the secondary recognition method with respective number thresholds.
5. The method according to claim 4, wherein the list of identified substance IDs in the pre-processing step of the false positive detection is selected as the similarity identifying substance ID list.
6. The method according to claim 4, wherein the list of identified substance IDs in the pre-processing step of the missing report detection is selected as the original identification substance ID list.
7. The method of claim 4, wherein a threshold number of times of the highest correct substance ID number obtained for all false positive or missing material IDs in the self-learning library is set to be larger than for the self-learning library The "self-learning type" takes the value of the number of times of the highest correct substance ID number obtained by the false positive or missing material ID of one of "feature recognition" and "secondary recognition".
8. The method according to claim 4 or 7, wherein when the number of highest correct substance IDs of said statistical selection method, said feature recognition method, and said secondary recognition method is compared with respective respective number of times thresholds, the condition is "highest" In the case where the correct substance ID number is greater than the number of times threshold "established at least twice, the method of satisfying the condition in the three parallel substance ID selection methods is continuously performed selectively to generate the corresponding at least two identification substance ID lists.
9. The method of claim 8 wherein the generated at least two identifiers If the quality ID list is equal, it is confirmed as the list of the identification substance ID after the verification detection.
10. The method according to claim 8, wherein if there is an intersection of the generated at least two identification substance ID lists, the intersection is confirmed as the list of the identified substance IDs after the verification detection.
11. The method according to claim 10, wherein said substance ID selecting step is performed again for a portion other than the intersection in the generated at least two identification substance ID lists.
12. The method according to claim 11, wherein said substance ID selecting step performed again comprises enhanced detection by using an article to be measured and an enhancer to obtain an enhanced Raman spectrum.
13. The method according to claim 4, wherein in the pre-processing step of the false positive detection, the post-processing step of the false positive detection is performed only when the number of statistical false positives is greater than the false alarm count threshold.
14. A method according to any one of claims 3 to 13, further comprising:

    After the qualitative analysis is performed on the measured object, the obtained false positive substance ID list and the missing material ID list are added to the self-learning library according to the "self-learning type" field.
15. The method according to claim 1, before performing qualitative analysis on the item to be tested, further comprising:

    The self-learning library is created by using the learning sample material to perform initial learning on the self-learning library and inputting a preset initial self-learning library.
16. The method of claim 1 further comprising:

    Manually comparing methods to identify substances.
17. An electronic device comprising:

    a memory for storing executable instructions;

    A processor for executing executable instructions stored in a memory to perform the method of any of claims 1-16.

Images