A contraband detection device risk early warning method and system using a cloud network end

Abstract

The application provides a contraband detection equipment risk early warning method and system using a cloud network end, wherein Raman spectrum data of a sample to be detected is obtained and wavelet transform is performed to generate a feature spectrum to be matched; a cloud server gathers non-alarm spectrum data uploaded by multiple terminals, and generates an environment background feature spectrum set by clustering; the feature spectrum to be matched is matched with a standard spectrum library and a background spectrum set, and based on spectrum cross-correlation, peak shift and peak shape correlation weighted operation, a comprehensive matching score is obtained; when the comprehensive matching score of the feature spectrum to be matched and any background spectrum is higher than a background suppression threshold, an adjusted contraband comprehensive matching score is generated; based on the adjusted contraband comprehensive matching score, a ratio of the highest score and the second highest score is calculated to determine a matching ambiguity coefficient, and in combination with the highest score, a preset threat level of the corresponding contraband and the matching ambiguity coefficient, a risk early warning level is output.

Description

Technical Field

[0001] This application belongs to the field of equipment early warning, and in particular relates to a risk early warning method and system for contraband detection equipment using cloud network terminals. Background Technology

[0002] Raman spectroscopy is a rapid, non-destructive, and fingerprint-based molecular spectral analysis technique. By irradiating a substance with a laser and analyzing its scattering spectrum, molecular structural information can be quickly obtained, enabling accurate identification of contraband such as explosives, drugs, and hazardous chemicals. However, actual detection environments are complex. Samples are often present in various packaging materials, containers, or mixtures. These coexisting background substances, such as plastics, textiles, beverages, and cosmetics, can severely interfere with the characteristic peaks of the target contraband, even completely masking the target signal. Existing techniques typically use static background spectral libraries to subtract background interference, but this method struggles to cope with changes in the on-site environment. For example, different batches of packaging materials, variations in ambient temperature and humidity, and detergent residues can introduce new and unknown background signals, causing the static spectral library to become invalid. This results in numerous false alarms and missed alarms, severely impacting the accuracy and reliability of the detection.

[0003] Most systems rely on single similarity calculation methods such as spectral cross-correlation or Euclidean distance when performing spectral matching, comparing the calculated score with a fixed threshold to determine the presence of contraband. When the spectrum of the sample shows a certain degree of similarity to multiple standard substances, the simple "highest score match" principle cannot reliably handle this ambiguity in identification. More importantly, most existing early warning mechanisms are binary, outputting only "alarm" or "safe" results without comprehensively considering the reliability of the identification results or the inherent differences in the hazard levels of different contraband. For example, the risks of high-concentration explosives and low-toxicity chemicals are obviously different, but traditional early warning systems cannot distinguish between them, lacking multi-level risk assessment and early warning capabilities. Especially when the identification results are ambiguous, such as being similar to the spectra of multiple substances, this uncertainty cannot be quantified and incorporated into the risk assessment, failing to meet the needs of rapid and tiered risk assessment in modern advanced security scenarios. Summary of the Invention

[0004] This invention proposes a risk warning method for contraband detection equipment using cloud-based networks. This method addresses the shortcomings of existing methods, such as their inability to cope with changes in the field environment and their failure to comprehensively consider the reliability of identification results and the inherent differences in the hazard levels of different contraband items. The method includes the following steps:

[0005] The Raman spectral data of the sample to be tested is acquired and wavelet transform is performed to generate the characteristic spectrum to be matched; the cloud server aggregates the non-alarm spectral data uploaded by multiple terminals in the warning area within a variable time period, and clusters them to generate a set of environmental background characteristic spectra;

[0006] The feature spectrum to be matched is matched with each standard spectrum in the cloud-based contraband standard spectrum library and each background spectrum in the environmental background feature spectrum set. Based on spectral cross-correlation, peak position shift and peak shape correlation, a weighted calculation is performed to obtain a comprehensive matching score corresponding to each of the standard spectrum and background spectrum.

[0007] When the combined matching score of the feature spectrum to be matched and any of the background spectra is higher than the background suppression threshold, nonlinear suppression is applied to the combined matching score of the feature spectrum to be matched and each of the standard spectra to generate a set of adjusted combined matching scores for contraband.

[0008] Based on the adjusted comprehensive matching score of prohibited items, the ratio of the highest score to the second highest score is calculated to determine the matching ambiguity coefficient. Then, the risk warning level is output by combining the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching ambiguity coefficient.

[0009] Optionally, the step of acquiring the Raman spectral data of the sample to be tested and performing wavelet transform to generate the characteristic spectrum to be matched includes:

[0010] The Raman spectral data is decomposed into a predetermined number of layers using a wavelet basis of a predetermined type, and the low-frequency approximation coefficients obtained from the specified layer decomposition are extracted as the feature spectra to be matched.

[0011] Optionally, the cloud server aggregates non-alarm spectral data uploaded by multiple terminals within the warning area over a variable time period, and clusters it to generate a set of environmental background feature spectra, including:

[0012] A density-based spatial clustering algorithm is used to perform cluster analysis on the aggregated non-alarm spectral data. The neighborhood radius of the algorithm is a preset value Eps, the minimum number of samples for the core object is a preset value MinPts, and the arithmetic mean of all spectral data in each cluster is used as the environmental background feature spectrum of the cluster.

[0013] Optionally, the duration of the time period Through formula Calculation, where As the base time period, The value is the average of the risk level values ​​of all warning events in the previous period, and k is the risk sensitivity coefficient. It is a positive number.

[0014] Optionally, the overall matching score S is calculated using the formula... calculate;

[0015] in This represents the normalized cross-correlation value between the feature spectrum to be matched and the standard or background spectrum. This is the normalized peak position matching score calculated based on the peak position deviation of the two main characteristic peaks. The correlation coefficient value of the peak profiles of the two is given. , and These are the corresponding preset weighting coefficients.

[0016] Optionally, the background suppression threshold Through formula

[0017]

[0018] Calculation, where Based on the inhibition threshold, The number of non-alarm spectral samples contained in the cluster to which the background spectrum belongs.

[0019] Optionally, the step of nonlinearly suppressing the comprehensive matching score between the feature spectrum to be matched and each of the standard spectra to generate a set of adjusted comprehensive matching scores for contraband includes:

[0020] Adjusted Comprehensive Matching Score for Prohibited Items Through formula calculate;

[0021] in The original comprehensive matching score with the i-th standard spectrum, The highest comprehensive matching score is obtained by matching the feature spectrum to be matched with each background spectrum in the set of environmental background feature spectra. The background suppression threshold is the background spectrum corresponding to the highest comprehensive matching score. This is the inhibition coefficient.

[0022] Optionally, the step of combining the highest score, the preset threat level of the contraband corresponding to the highest score, and the matching fuzziness coefficient to output a risk warning level includes:

[0023] Through formula Calculate the Risk score;

[0024] The risk score is then divided into multiple warning levels based on multi-level preset thresholds.

[0025] in This is the highest score after adjustment. The highest score corresponds to the preset threat level of the contraband. To match the ambiguity coefficients, , and The preset weighting coefficients, and .

[0026] Furthermore, this invention also relates to a risk warning system for contraband detection equipment utilizing a cloud network, comprising the following modules:

[0027] The first generation module is used to acquire the Raman spectral data of the sample to be tested and perform wavelet transform to generate the feature spectrum to be matched; the cloud server aggregates the non-alarm spectral data uploaded by multiple terminals in the warning area within a variable time period, and clusters them to generate a set of environmental background feature spectra;

[0028] The calculation module is used to match the feature spectrum to be matched with each standard spectrum in the cloud-based contraband standard spectrum library and each background spectrum in the environmental background feature spectrum set. Based on spectral cross-correlation, peak position shift and peak shape correlation, a weighted calculation is performed to obtain a comprehensive matching score corresponding to each of the standard spectrum and background spectrum.

[0029] The second generation module is used to perform nonlinear suppression on the comprehensive matching score of the feature spectrum to be matched and each of the standard spectra when the comprehensive matching score of the feature spectrum to be matched and any of the background spectra is higher than the background suppression threshold, and generate a set of adjusted comprehensive matching scores of contraband.

[0030] The adjustment module is used to calculate the ratio of the highest score to the second highest score based on the adjusted comprehensive matching score of prohibited items to determine the matching ambiguity coefficient, and output the risk warning level by combining the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching ambiguity coefficient.

[0031] Preferably, the step of acquiring the Raman spectral data of the sample to be tested and performing wavelet transform to generate the characteristic spectrum to be matched includes:

[0032] The Raman spectral data is decomposed into a predetermined number of layers using a wavelet basis of a predetermined type, and the low-frequency approximation coefficients obtained from the specified layer decomposition are extracted as the feature spectra to be matched.

[0033] Preferably, the cloud server aggregates non-alarm spectral data uploaded by multiple terminals within the warning area over a variable time period, and clusters it to generate an environmental background feature spectral set, including:

[0034] A density-based spatial clustering algorithm is used to perform cluster analysis on the aggregated non-alarm spectral data. The neighborhood radius of the algorithm is a preset value Eps, the minimum number of samples for the core object is a preset value MinPts, and the arithmetic mean of all spectral data in each cluster is used as the environmental background feature spectrum of the cluster.

[0035] Preferably, the duration of the time period Through formula Calculation, where As the base time period, The value is the average of the risk level values ​​of all warning events in the previous period, and k is the risk sensitivity coefficient. It is a positive number.

[0036] Preferably, the comprehensive matching score S is calculated using the formula... calculate;

[0037] in This represents the normalized cross-correlation value between the feature spectrum to be matched and the standard or background spectrum. This is the normalized peak position matching score calculated based on the peak position deviation of the two main characteristic peaks. The correlation coefficient value of the peak profiles of the two is given. , and These are the corresponding preset weighting coefficients.

[0038] Preferably, the background suppression threshold Through formula

[0039]

[0040] Calculation, where Based on the inhibition threshold, The number of non-alarm spectral samples contained in the cluster to which the background spectrum belongs.

[0041] Preferably, the step of nonlinearly suppressing the comprehensive matching score between the feature spectrum to be matched and each of the standard spectra to generate a set of adjusted comprehensive matching scores for contraband includes:

[0042] Adjusted Comprehensive Matching Score for Prohibited Items Through formula calculate;

[0043] in The original comprehensive matching score with the i-th standard spectrum, The highest comprehensive matching score is obtained by matching the feature spectrum to be matched with each background spectrum in the set of environmental background feature spectra. The background suppression threshold is the background spectrum corresponding to the highest comprehensive matching score. This is the inhibition coefficient.

[0044] Preferably, the step of combining the highest score, the preset threat level of the contraband corresponding to the highest score, and the matching fuzziness coefficient to output a risk warning level includes:

[0045] Through formula Calculate the Risk score;

[0046] The risk score is then divided into multiple warning levels based on multi-level preset thresholds.

[0047] in This is the highest score after adjustment. The highest score corresponds to the preset threat level of the contraband. To match the ambiguity coefficients, , and The preset weighting coefficients, and .

[0048] This invention constructs a background spectral library that reflects the environmental composition in real time by aggregating and clustering the spectra collected by numerous front-end devices in the cloud. This solves the problem of fixed spectral libraries being unable to cope with complex and ever-changing on-site interference, thus improving the accuracy of background subtraction. For spectral identification, a weighted matching method integrating multi-dimensional features such as spectral cross-correlation, peak position, and peak shape is employed to enhance the ability to identify mixtures and weak signals. Furthermore, a mechanism for suppressing high-frequency background signals is incorporated. When the analyte is highly similar to common background substances, the matching score between the analyte and the standard spectrum of prohibited items is reduced, decreasing false alarms caused by common packaging, solvents, etc. By fusing matching scores, the ambiguity of matching results, and the threat level of the substance itself, a risk assessment model is established to output refined warning levels, providing security personnel with a more reliable basis for decision-making and improving the overall performance of the detection system. Attached Figure Description

[0049] Figure 1 A flowchart of the first embodiment;

[0050] Figure 2 This is a schematic diagram illustrating the relationship between the threshold and sample density.

[0051] Figure 3 This is a schematic diagram of the matching score after suppression;

[0052] Figure 4 This is a schematic diagram of risk assessment weight adjustment based on ambiguity. Detailed Implementation

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] In the first embodiment, the present invention proposes a risk warning method for contraband detection equipment using cloud network terminals, such as... Figure 1 This includes the following steps:

[0055] S1. Acquire the Raman spectral data of the sample to be tested and perform wavelet transform to generate the characteristic spectrum to be matched; The cloud server aggregates the non-alarm spectral data uploaded by multiple terminals in the warning area within a variable time period, and clusters them to generate a set of environmental background characteristic spectra.

[0056] Specifically, the operator uses a handheld Raman spectrometer to collect the raw Raman spectrum of the sample to be tested. This raw spectral data includes the target signal, background fluorescence, cosmic ray noise, and thermal noise. The raw spectral data is decomposed and reconstructed in eight layers using, for example, the Daubechies 4 wavelet basis. During the decomposition process, high-frequency noise coefficients are removed to smooth the spectrum, and low-frequency approximation coefficients are subtracted after polynomial fitting. The baseline is corrected to generate a matching characteristic spectrum with clearer characteristic peaks, free from noise and fluorescence background interference.

[0057] Multiple Raman spectroscopy detection terminals deployed at airport security checkpoints upload the spectral data of samples deemed safe to a cloud server in real time via a 5G network. The server collects all non-alarm spectra within a set time period, such as one hour. After the period ends, the server calculates the average risk warning level of all alarm events within that period. If the average risk level is high, the next collection period is shortened to 30 minutes; otherwise, it is extended to two hours. The server uses the DBSCAN density clustering algorithm to process the collected non-alarm spectra, grouping similar spectra into clusters and calculating the centroid spectrum of each cluster. All centroid spectra together constitute the environmental background characteristic spectrum set. To prevent the non-alarm spectral data from containing contraband characteristics, all non-alarm spectra are filtered, for example, through manual verification or by retaining only non-alarm spectra that differ significantly from all alarm spectra.

[0058] In an optional embodiment, acquiring the Raman spectral data of the sample to be tested and performing wavelet transform to generate the characteristic spectrum to be matched includes:

[0059] The Raman spectral data is decomposed into a predetermined number of layers using a wavelet basis of a predetermined type, and the low-frequency approximation coefficients obtained from the specified layer decomposition are extracted as the feature spectra to be matched.

[0060] Raw Raman spectra are acquired, typically as a one-dimensional data array containing noise and baseline drift, such as a sequence of 1024 data points. A specific wavelet basis function is selected, such as the Daubechies4 (db4 wavelet), and the decomposition level is set to 5. The wavelet transform model decomposes the signal at different scales by convolving the raw spectral signal with the wavelet basis function, equivalent to passing the signal through a series of high-pass and low-pass filters.

[0061] The aforementioned 1024-point spectral data is input into the 5-layer db4 wavelet decomposition model. Each layer of decomposition generates a set of high-frequency detail coefficients D and a set of low-frequency approximation coefficients A. After 5 layers of decomposition, the following is obtained: , , , , and Six sets of coefficients. Extract the low-frequency approximation coefficients of the fifth layer. This set of coefficients represents the core contour features of the original spectrum, while filtering out high-frequency noise and reducing the data dimensionality, for example, from 1024 points to approximately 37 points. This set of low-frequency approximation coefficients... That is, the spectrum of features to be matched is used for subsequent matching and identification.

[0062] In an optional embodiment, the cloud server aggregates non-alarm spectral data uploaded by multiple terminals within a variable time period in the early warning area, and clusters it to generate an environmental background feature spectral set, including:

[0063] A density-based spatial clustering algorithm is used to perform cluster analysis on the aggregated non-alarm spectral data. The neighborhood radius of the algorithm is a preset value Eps, the minimum number of samples for the core object is a preset value MinPts, and the arithmetic mean of all spectral data in each cluster is used as the environmental background feature spectrum of the cluster.

[0064] The density-based spatial clustering algorithm model employed, such as DBSCAN, automatically identifies categories representing common environmental backgrounds from a large amount of unlabeled, non-alarm spectral data. Before the algorithm executes, two key parameters need to be set: the neighborhood radius Eps and the minimum number of samples (MinPts) for core objects. For example, setting Eps to 0.5 and MinPts to 20 means that if a spectral data point has at least 20 other spectral data points within a 0.5-distance range, it is defined as a core object, thus forming the starting point of a high-density region.

[0065] Assume a cloud server has collected 100,000 non-alarm spectral data points from different detection points. The DBSCAN algorithm traverses all data points and, based on preset Eps and MinPts values, divides densely connected data points into different clusters. For example, the algorithm might identify three main clusters, corresponding to three background types: common plastic bottles, textiles, and leather. For one cluster representing plastic bottles, assuming it contains 5,000 spectral data points, the intensity values ​​at each Raman shift point are arithmetically averaged to generate a smooth and representative average spectrum. This spectrum is defined as the environmental background characteristic spectrum of the cluster.

[0066] In an optional embodiment, the duration of the time period Through formula Calculation, where As the base time period, The value is the average of the risk level values ​​of all warning events in the previous period, and k is the risk sensitivity coefficient. It is a positive number.

[0067] A detection frequency adjustment model was constructed. The core idea of ​​this model is to adjust the future detection rhythm based on historical risks. A baseline time period is pre-set. For example, 60 seconds is used as the standard detection interval when no risk events occur. A risk sensitivity coefficient k, such as 0.8, is also set to amplify the impact of the risk level on the cycle, along with a normalization factor. For example, 1 ensures that the denominator is not zero and serves as the basic adjustment quantity.

[0068] After a monitoring period ends, all warning events that occurred within that period are statistically analyzed. Assuming that two warning events occurred in the previous period (n-1 periods), with risk levels of 0.5 and 0.7 respectively, then the average risk level for that period is... The value is 0.6. Substitute this value into the formula to calculate the duration of the next cycle. The interval was approximately 40.5 seconds. This indicates that the system automatically shortened the detection interval and increased the detection frequency due to sensing a higher risk. If there were no warning events in the previous cycle, If it is 0, then It will be restored to the baseline period of 60 seconds.

[0069] S2, Match the feature spectrum to be matched with each standard spectrum in the cloud-based contraband standard spectrum library and each background spectrum in the environmental background feature spectrum set, and perform weighted calculations based on spectral cross-correlation, peak position shift and peak shape correlation to obtain a comprehensive matching score corresponding to each of the standard spectrum and background spectrum;

[0070] Specifically, the spectral cross-correlation score is obtained by calculating the Pearson correlation coefficient between the feature spectrum to be matched and the standard spectrum A. Peak finding processing is performed on both to identify their respective main characteristic peaks. The average difference between the corresponding peak positions is calculated, and the average value is converted into a peak position offset score using a preset function. The smaller the offset, the higher the score; a small window of 20 data points around each corresponding characteristic peak is extracted, and the average correlation of the spectral segments within the window is calculated as the peak shape correlation score. ; through formula 0.5× +0.3× +0.2× Perform a weighted summation to obtain the comprehensive matching score between the feature spectrum to be matched and the standard spectrum A; repeat this process for all standard and background spectra in the spectral library.

[0071] In an optional embodiment, the comprehensive matching score S is determined by the formula... calculate;

[0072] in This represents the normalized cross-correlation value between the feature spectrum to be matched and the standard or background spectrum. This is the normalized peak position matching score calculated based on the peak position deviation of the two main characteristic peaks. The correlation coefficient value of the peak profiles of the two is given. , and These are the corresponding preset weighting coefficients.

[0073] Furthermore, the normalized peak position matching score The calculation methods include:

[0074] and .

[0075] Calculate the mean absolute deviation of the corresponding characteristic peak: , ;

[0076] By setting the maximum allowable peak position deviation Normalization is performed to obtain the peak position matching score:

[0077]

[0078] in, It can be set according to the spectrometer resolution and the properties of the substance, for example . The range of values ​​is A higher value indicates a higher peak position matching degree. A multi-dimensional spectral similarity evaluation model is implemented, which improves the accuracy of identification by weighted fusion of feature matching degrees from three different levels. The normalized cross-correlation values ​​between two spectra are calculated. This represents the overall linear correlation between two spectral curves. By identifying and comparing the Raman shift positions of their main characteristic peaks, the peak position deviation is mapped to a normalized peak position matching score. The smaller the deviation, the higher the score. The correlation coefficient value of the peak profile is calculated. To assess the similarity of the local shapes of characteristic peaks.

[0079] Assuming the spectrum to be measured is matched with a standard spectrum, and a preset weight is used... It is 0.5. It is 0.3. The weighting coefficient is 0.2. The weighting coefficients can be determined experimentally based on the actual application scenario and the characteristics of the spectral database; the sum of the three is usually 1. The overall correlation is then calculated. The value is 0.95, indicating that the peak position deviation of the main characteristic peak is extremely small, thus yielding the peak position score. The value is 0.98, and the peak shape profile is also very similar, resulting in... If the score is 0.90, then the overall matching score S = 0.949. This high score integrates the high consistency of the overall structure, key locations, and local shapes, providing a more reliable matching conclusion than a single indicator.

[0080] S3, when the combined matching score of the feature spectrum to be matched and any of the background spectra is higher than the background suppression threshold, nonlinear suppression is applied to the combined matching score of the feature spectrum to be matched and each of the standard spectra to generate a set of adjusted combined matching scores for contraband.

[0081] Assuming the combined matching score between the measured spectrum and the PET spectra of mineral water bottles in the background spectrum set is 0.95; server statistics show that the cluster to which the PET spectra belong contains 2000 spectra, indicating a high sample density. The background suppression threshold corresponding to this cluster is calculated to be a relatively low 0.9. Figure 2 Since 0.95 is higher than 0.9, the suppression mechanism is triggered. At this point, the original comprehensive matching score between the measured spectrum and ammonium nitrate in the standard spectral library is 0.8, so the suppression function is applied. The matching score was processed to obtain an adjusted ammonium nitrate matching score of 0.8 × exp(-2 × 0.95). This reduced score avoids false alarms caused by interference from mineral water bottles. Figure 3 .

[0082] In an optional embodiment, the nonlinear suppression of the comprehensive matching scores between the feature spectrum to be matched and each of the standard spectra to generate a set of adjusted comprehensive matching scores for contraband includes:

[0083] Adjusted Comprehensive Matching Score for Prohibited Items Through formula calculate;

[0084] in The original comprehensive matching score with the i-th standard spectrum, The highest comprehensive matching score is obtained by matching the feature spectrum to be matched with each background spectrum in the set of environmental background feature spectra. The background suppression threshold is the background spectrum corresponding to the highest comprehensive matching score. This is the inhibition coefficient.

[0085] Background interference suppression intelligently corrects the match score of contraband in the presence of background substances with similar spectra to the contraband, reducing false alarms. The original match score between the test sample and the i-th contraband is calculated. Simultaneously, it matches the sample to be tested against all known environmental background spectral characteristics to identify the highest matching score. And obtain the suppression threshold corresponding to this background. .

[0086] Assume the original matching score between the sample to be tested and the prohibited item A. It has a score of 0.85. This is the highest background matching score obtained after matching with all background spectra. The threshold value is 0.80, which corresponds to this background value. The value is 0.70. Let the inhibition coefficient be... The score is 2.0. At this point, the background matching score exceeds the threshold by 0.1. The adjusted score... The score is approximately 0.696. The original score has decreased, indicating strong background interference. If... If the value is 0.65, which is below the threshold of 0.70, then the excess amount is 0, the exponential portion is 0, and the adjusted score is... It remains at 0.85, unaffected. When the background matching score just exceeds the threshold, the suppression effect is weak, avoiding missing potential real signals; however, when the background matching score far exceeds the threshold, indicating that the sample is very likely to be background material, the suppression effect will be sharply enhanced, more effectively lowering the matching score of the prohibited item.

[0087] In an optional embodiment, the background suppression threshold Through formula

[0088]

[0089] Calculation, where Based on the inhibition threshold, The number of non-alarm spectral samples contained in the cluster to which the background spectrum belongs.

[0090] A threshold model based on statistical confidence levels was established to adjust the suppression intensity for different background environments. A global baseline suppression threshold was set. For example, 0.9 represents the initial suppression standard for any background. It is a key parameter that represents the size of the cluster in which a specific background spectrum belongs, i.e., the frequency of this type of background in historical data.

[0091] Suppose the system identifies that the current environment matches background A and background B. Background A comes from a context containing... The clusters of samples represent a very common packaging material. Suppression threshold calculation. It is approximately 0.088. Background B comes from a source containing only... A cluster of samples represents an uncommon material, threshold The value is approximately 0.183. Therefore, for background A, which occurs very frequently, a very low suppression threshold is used. Only when the match between the analyte and the background A is less than 0.088 is it considered to be without interference; otherwise, strong suppression is performed.

[0092] S4. Based on the adjusted comprehensive matching score of prohibited items, calculate the ratio of the highest score to the second highest score to determine the matching ambiguity coefficient, and combine the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching ambiguity coefficient to output the risk warning level.

[0093] Specifically, after background suppression, the adjusted matching score between the analyte and triacetone triperoxide (TATP) is calculated to be 0.88, the highest score, and the matching score with RDX (rhexane) is 0.7, the second highest score. The calculated matching ambiguity coefficient is approximately 0.795. According to the preset information database, the threat level of TATP is 5, representing extremely high danger. The risk assessment model adjusts the weights according to the ambiguity coefficient, and the risk score is calculated as (1-0.795)×0.88+0.795×5 / 5, where 5 / 5 is the normalized threat level. If the calculated risk score falls within the preset risk range, for example, greater than 0.8, a level one red warning is output.

[0094] In an optional embodiment, the step of combining the highest score, the preset threat level of the contraband corresponding to the highest score, and the matching fuzziness coefficient to output a risk warning level includes:

[0095] Through formula Calculate the Risk score;

[0096] The risk score is then divided into multiple warning levels based on multi-level preset thresholds.

[0097] in This is the highest score after adjustment. The second highest score, The highest score corresponds to the preset threat level of the contraband. To match the ambiguity coefficients, , and The preset weighting coefficients, and .

[0098] The model integrates the confidence level of the match, the ambiguity of the identification, and the inherent danger of the substance to output a comprehensive risk assessment. The model is evaluated based on the highest matching score. and the second highest score The ratio is used to calculate the matching ambiguity coefficient. The closer this coefficient is to 1, the more difficult it is to distinguish between the two most likely substances, and the lower the certainty of the identification result.

[0099] Assuming pre-defined weights , , The highest score after adjustment in a certain test. The value is 0.92, corresponding to substance A, with a preset threat level. It is 0.9; the second highest score. The value is 0.83. The calculated ambiguity is... Risk Score The risk score is approximately 0.859. Due to the high ambiguity, the model automatically reduces the weight of the matching score while increasing the weight of the threat level of the substance itself, reflecting the principle of conservative early warning under uncertain conditions. The risk score of 0.859 is compared with preset multi-level thresholds such as 0.5 and 0.8 to determine that the risk score belongs to a high-level risk warning. Figure 4 .

[0100] In a second embodiment, the present invention also provides a risk warning system for contraband detection equipment utilizing a cloud network, comprising the following modules:

[0101] The first generation module is used to acquire the Raman spectral data of the sample to be tested and perform wavelet transform to generate the feature spectrum to be matched; the cloud server aggregates the non-alarm spectral data uploaded by multiple terminals in the warning area within a variable time period, and clusters them to generate a set of environmental background feature spectra;

[0102] The calculation module is used to match the feature spectrum to be matched with each standard spectrum in the cloud-based contraband standard spectrum library and each background spectrum in the environmental background feature spectrum set. Based on spectral cross-correlation, peak position shift and peak shape correlation, a weighted calculation is performed to obtain a comprehensive matching score corresponding to each of the standard spectrum and background spectrum.

[0103] The second generation module is used to perform nonlinear suppression on the comprehensive matching score of the feature spectrum to be matched and each of the standard spectra when the comprehensive matching score of the feature spectrum to be matched and any of the background spectra is higher than the background suppression threshold, and generate a set of adjusted comprehensive matching scores of contraband.

[0104] The adjustment module is used to calculate the ratio of the highest score to the second highest score based on the adjusted comprehensive matching score of prohibited items to determine the matching ambiguity coefficient, and output the risk warning level by combining the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching ambiguity coefficient.

[0105] In this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0106] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0107] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A risk warning method for contraband detection equipment utilizing cloud network terminals, characterized in that, Includes the following steps: The Raman spectral data of the sample to be tested are acquired and wavelet transform is performed to generate the characteristic spectrum to be matched; The cloud server aggregates non-alarm spectral data uploaded by multiple terminals within the warning area during a variable time period, and clusters them to generate a set of environmental background feature spectra. The feature spectrum to be matched is matched with each standard spectrum in the cloud-based contraband standard spectrum library and each background spectrum in the environmental background feature spectrum set. Based on spectral cross-correlation, peak position shift and peak shape correlation, a weighted calculation is performed to obtain a comprehensive matching score corresponding to each of the standard spectrum and background spectrum. When the combined matching score of the feature spectrum to be matched and any of the background spectra is higher than the background suppression threshold, nonlinear suppression is applied to the combined matching score of the feature spectrum to be matched and each of the standard spectra to generate a set of adjusted combined matching scores for contraband. Based on the adjusted comprehensive matching score of prohibited items, the ratio of the highest score to the second highest score is calculated to determine the matching ambiguity coefficient. Then, the risk warning level is output by combining the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching ambiguity coefficient.

2. The method according to claim 1, characterized in that, The process of acquiring the Raman spectral data of the sample to be tested and performing wavelet transform to generate the characteristic spectrum to be matched includes: The Raman spectral data is decomposed into a predetermined number of layers using a wavelet basis of a predetermined type, and the low-frequency approximation coefficients obtained from the specified layer decomposition are extracted as the feature spectra to be matched.

3. The method according to claim 1, characterized in that, The cloud server aggregates non-alarm spectral data uploaded by multiple terminals within the warning area over a variable time period, and clusters it to generate a set of environmental background feature spectra, including: A density-based spatial clustering algorithm is used to perform cluster analysis on the aggregated non-alarm spectral data. The neighborhood radius of the algorithm is a preset value Eps, the minimum number of samples for the core object is a preset value MinPts, and the arithmetic mean of all spectral data in each cluster is used as the environmental background feature spectrum of the cluster.

4. The method according to claim 1, characterized in that, The duration of the time period Through formula Calculation, where As the base time period, The value is the average of the risk level values ​​of all warning events in the previous period, and k is the risk sensitivity coefficient. It is a positive number.

5. The method according to claim 1, characterized in that, The overall matching score S is obtained through the formula calculate; in This represents the normalized cross-correlation value between the feature spectrum to be matched and the standard or background spectrum. This is the normalized peak position matching score calculated based on the peak position deviation of the two main characteristic peaks. The correlation coefficient value of the peak profiles of the two is given. , and These are the corresponding preset weighting coefficients.

6. The method according to claim 1, characterized in that, The background suppression threshold Through formula Calculation, where Based on the inhibition threshold, The number of non-alarm spectral samples contained in the cluster to which the background spectrum belongs.

7. The method according to claim 1, characterized in that, The process involves nonlinearly suppressing the comprehensive matching scores of the feature spectra to be matched and each of the standard spectra to generate a set of adjusted comprehensive matching scores for prohibited items, including: Adjusted Comprehensive Matching Score for Prohibited Items Through formula calculate; in The original comprehensive matching score with the i-th standard spectrum, The highest comprehensive matching score is obtained by matching the feature spectrum to be matched with each background spectrum in the set of environmental background feature spectra. The background suppression threshold is the background spectrum corresponding to the highest comprehensive matching score. This is the inhibition coefficient.

8. The method according to claim 1, characterized in that, The risk warning level is output by combining the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching fuzziness coefficient, including: Through formula Calculate the Risk score; The risk score is then divided into multiple warning levels based on multi-level preset thresholds. in This is the highest score after adjustment. The highest score corresponds to the preset threat level of the contraband. To match the ambiguity coefficients, , and The preset weighting coefficients, and .

9. A risk warning system for contraband detection equipment utilizing cloud network terminals, characterized in that, Includes the following modules: The first generation module is used to acquire the Raman spectral data of the sample to be tested and perform wavelet transform to generate the characteristic spectrum to be matched; The cloud server aggregates non-alarm spectral data uploaded by multiple terminals within the warning area during a variable time period, and clusters them to generate a set of environmental background feature spectra. The calculation module is used to match the feature spectrum to be matched with each standard spectrum in the cloud-based contraband standard spectrum library and each background spectrum in the environmental background feature spectrum set. Based on spectral cross-correlation, peak position shift and peak shape correlation, a weighted calculation is performed to obtain a comprehensive matching score corresponding to each of the standard spectrum and background spectrum. The second generation module is used to perform nonlinear suppression on the comprehensive matching score of the feature spectrum to be matched and each of the standard spectra when the comprehensive matching score of the feature spectrum to be matched and any of the background spectra is higher than the background suppression threshold, and generate a set of adjusted comprehensive matching scores of contraband. The adjustment module is used to calculate the ratio of the highest score to the second highest score based on the adjusted comprehensive matching score of prohibited items to determine the matching ambiguity coefficient, and output the risk warning level by combining the highest score, the preset threat level of the prohibited item corresponding to the highest score, and the matching ambiguity coefficient.

10. The system according to claim 9, characterized in that, The process of acquiring the Raman spectral data of the sample to be tested and performing wavelet transform to generate the characteristic spectrum to be matched includes: The Raman spectral data is decomposed into a predetermined number of layers using a wavelet basis of a predetermined type, and the low-frequency approximation coefficients obtained from the specified layer decomposition are extracted as the feature spectra to be matched.

Images