Hazardous chemical substance identification method and device based on multispectral fusion

Abstract

The invention provides a hazardous chemical substance identification method and device based on multispectral fusion, and relates to the technical field of hazardous chemical substance identification. The method comprises the following steps: respectively carrying out feature extraction on a first spectrum and a second spectrum of a to-be-detected object to obtain a first feature of the first spectrum and a second feature of the second spectrum; determining a first matching degree of the first feature and a second matching degree of the second feature according to the object features of the to-be-detected article; based on the first matching degree and the second matching degree, performing feature fusion on the first feature and the second feature to obtain a spectral feature of the to-be-detected article; and inputting the spectral features into a pre-trained hazardous chemical substance identification model to obtain a hazardous chemical substance identification result of the to-be-detected article. The accuracy of the spectral characteristics and the detection precision of the hazardous chemical substances can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hazardous chemical identification, and in particular, to a method and device for identifying hazardous chemicals based on multispectral fusion. Background Art

[0002] Hazardous chemicals, such as strong acids, organic solvents, cyanides, etc., refer to chemical substances or mixtures with physical hazards, health hazards, or environmental hazards. Hazardous chemicals pose a serious threat to human health, public safety, or the ecological environment. Such chemicals are usually strictly controlled, and their production, storage, transportation, and use need to comply with specific standards and specifications.

[0003] While international trade in hazardous chemicals meets industrial needs and technological exchanges, it may also bring significant risks. Their cross-border circulation may exacerbate potential hazards such as environmental leaks and transportation accidents. For example, toxic substance leaks can contaminate water sources, or the transportation of flammable goods can cause explosions. Currently, infrared spectroscopy detection technology is often used as a rapid screening method in on-site detection. Through infrared spectroscopy detection technology, the component identification of the characteristic spectral peaks of substances can be carried out to preliminarily determine whether the goods contain hazardous chemicals such as flammable, explosive, or toxic substances.

[0004] However, infrared spectroscopy detection technology is easily affected by the interference of mixtures, resulting in misjudgment or missed detection. Moreover, it has insufficient sensitivity to low-concentration substances or complex matrices. In addition, due to the limited coverage of the infrared spectroscopy database, it is difficult to deal with new or modified chemicals. Summary of the Invention

[0005] Embodiments of the present invention provide a method and device for identifying hazardous chemicals based on multispectral fusion to solve the problems of low detection accuracy and difficulty in dealing with new or modified chemicals when using infrared spectroscopy detection technology for hazardous chemical detection.

[0006] In a first aspect, embodiments of the present invention provide a method for identifying hazardous chemicals based on multispectral fusion, including: Performing feature extraction on the first spectrum and the second spectrum of the item to be detected respectively to obtain the first feature of the first spectrum and the second feature of the second spectrum; Determining the first matching degree of the first feature and the second matching degree of the second feature according to the object feature of the item to be detected; Based on the first matching degree and the second matching degree, performing feature fusion on the first feature and the second feature to obtain the spectral feature of the item to be detected; Inputting the spectral feature into a pre-trained hazardous chemical identification model to obtain the hazardous chemical identification result of the item to be detected.

[0007] In a second aspect, embodiments of the present invention provide a device for identifying hazardous chemicals based on multispectral fusion, including: An extraction module for respectively extracting features from the first spectrum and the second spectrum of the item to be detected, obtaining the first feature of the first spectrum and the second feature of the second spectrum; A determination module for determining the first matching degree of the first feature and the second matching degree of the second feature according to the object feature of the item to be detected; A fusion module for performing feature fusion on the first feature and the second feature based on the first matching degree and the second matching degree to obtain the spectral feature of the item to be detected; An identification module for inputting the spectral feature into a pre-trained hazardous chemical identification model to obtain the hazardous chemical identification result of the item to be detected.

[0008] In the embodiment of the present invention, by synchronously extracting the first spectrum and the second spectral feature of the item to be detected, combining the object feature to evaluate the feature matching degree, performing feature fusion according to the feature matching degree to obtain the spectral feature, and inputting the spectral feature into a pre-trained hazardous chemical identification model to obtain the hazardous chemical identification result of the item to be detected. Using the object feature to obtain the matching degree and obtaining the spectral feature according to the feature of the matching degree can improve the accuracy of feature representation. Combining with the deep learning-based hazardous chemical identification model can accurately identify known hazardous chemicals through spectral features, and determine whether there is danger when the item is not a known hazardous chemical according to the spectral features. Therefore, this application can effectively improve the accuracy of spectral features and the detection accuracy of hazardous chemicals, and has the ability to identify new or modified chemicals. Description of the Drawings

[0009] Figure 1 is the implementation flowchart of the hazardous chemical identification method based on multi-spectral fusion provided by the embodiment of the present invention; Figure 2 is the implementation flowchart of step S120 of the hazardous chemical identification method based on multi-spectral fusion provided by the embodiment of the present invention; Figure 3 is the implementation flowchart of step S130 of the hazardous chemical identification method based on multi-spectral fusion provided by the embodiment of the present invention; Figure 4 is the implementation flowchart of step S140 of the hazardous chemical identification method based on multi-spectral fusion provided by the embodiment of the present invention; Figure 5 is the implementation flowchart of step S1402 of the hazardous chemical identification method based on multi-spectral fusion provided by the embodiment of the present invention; Figure 6 is the structural schematic diagram of the hazardous chemical identification device based on multi-spectral fusion provided by the embodiment of the present invention. Detailed Embodiments

[0010] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0011] Refer to Figure 1 , which shows the implementation flowchart of the hazardous chemical identification method based on multispectral fusion provided by the embodiments of the present invention, and is described in detail as follows: Step S110: Extract features from the first spectrum and the second spectrum of the item to be detected respectively, to obtain the first feature of the first spectrum and the second feature of the second spectrum.

[0012] In some embodiments, the first spectrum and the second spectrum of the item to be detected can be obtained through spectral detection. Specifically, the item to be detected can be first subjected to spectral detection to obtain the initial first spectrum and the initial second spectrum of the item to be detected. Then, preprocessing is performed on the initial first spectrum and the initial second spectrum to obtain the first spectrum and the second spectrum. The preprocessing can include data denoising processing and baseline correction processing.

[0013] Specifically, spectral detection is a technology for obtaining substance composition information by analyzing the interaction between substances and electromagnetic waves. When light of a specific wavelength irradiates the surface of a substance, the substance will absorb or scatter light of a specific wavelength due to molecular vibration, electron transition, etc., forming a characteristic spectrum. For example, in the detection of hazardous chemicals, infrared spectral detection can identify the C-H bond absorption peak of organic solvents, while Raman spectroscopy can detect the molecular vibration mode, and the combination of the two can analyze chemical components more comprehensively.

[0014] It should be noted that in this application, two different detection methods are required to perform spectral detection on the item to be detected. The infrared spectral detection technology and the Raman spectral detection technology are adopted in this application. The initial first spectrum and the initial second spectrum are the raw data directly obtained by different spectral detection methods. The initial first spectrum is the infrared spectral response measured by using the infrared spectral detection technology for hazardous chemicals, while the initial second spectrum is the Raman spectral response obtained by measuring hazardous chemicals through the Raman spectral detection technology. Both contain unprocessed noise and baseline drift. For example, there will be environmental light interference in the infrared spectrum.

[0015] In some embodiments, data denoising is used to eliminate the interference of non-target signals in the spectrum through algorithms. For example, the Savitzky-Golay filter can be used to smooth the infrared spectrum, removing the small fluctuations during instrument acquisition while retaining the effective peak shape. Baseline correction is used to eliminate the non-linear background offset in the spectrum. For example, when detecting strong acid solutions, the fluorescence effect will cause the overall baseline of the Raman spectrum to rise. At this time, the iterative least squares method can be used to fit the baseline curve and then subtract it from the original spectrum. In addition, the preprocessing can also include processing methods such as spectral smoothing, normalization, wavelength calibration, and scattering processing.

[0016] By synchronously collecting the initial spectral data of the item to be measured through multispectral fusion, combined with data denoising and baseline correction preprocessing, it is possible to effectively eliminate noise interference and baseline drift, improve the authenticity and signal-to-noise ratio of spectral features, thereby enhancing the accuracy of subsequent feature fusion and model recognition.

[0017] In this way, through preprocessing, noise interference can be eliminated, and the signal-to-noise ratio of the first spectrum and the second spectrum can be improved.

[0018] In some embodiments, feature extraction is a process of screening out discriminative key information from the first spectrum and the second spectrum. For example, the derivative method is used for infrared spectra to enhance the resolution of overlapping peaks, or principal component analysis is performed on Raman spectra for dimensionality reduction, and the first 3 principal components are extracted as feature vectors representing the overall changes of the spectrum.

[0019] In some embodiments, the first feature is a quantitative index extracted from the first spectrum, including the intensity, peak area, or waveform parameters at specific wavelengths. For example, the first feature of an infrared spectrum may include the absorbance value of the ester group C=O bond at 1740 cm -1 and the full width at half maximum of this peak. The second feature is a quantitative index extracted from the preprocessed second spectrum, including parameters such as the intensity, peak area, full width at half maximum, and peak position shift at specific Raman shifts. For example, when detecting sulfuric acid, its S-O bond vibration generates a characteristic peak at 980 cm -1 and the second feature can quantify the intensity, peak area ratio, and full width at half maximum of this peak.

[0020] Step S120: Determine the first matching degree of the first feature and the second matching degree of the second feature according to the object features of the item to be detected.

[0021] In some embodiments, the object features refer to the physical properties of the object to be detected, including packaging materials, packaging forms, packaging surface characteristics, and possible interfering factors. For example, a viscous liquid contained in a brown plastic bottle, its object features include plastic material, opaque packaging, and liquid form.

[0022] In some embodiments, the first matching degree is used to measure the compatibility between the infrared spectral features and the object features. For example, if the item to be measured is a powder in a metal can, the high reflectivity of the metal surface to infrared light will cause a decrease in the signal-to-noise ratio, and at this time, the infrared feature matching degree is relatively low. The second matching degree is used to evaluate the correlation between the Raman spectral features and the object features. For example, a transparent glass container has little interference on Raman signals, and the Raman feature matching degree is relatively high.

[0023] See Figure 2 For the above, the specific processing of step S120 may include steps S1201 - S1203, and the specific content is as follows: Step S1201: Determine the packaging information and interference information of the item to be detected based on the object features of the item to be detected.

[0024] In some embodiments, the packaging information includes packaging material, packaging form, and packaging surface characteristics. For example, taking powdered substances sealed in an aluminum metal can as an example, its packaging information is an aluminum metal can, a cylindrical form, and a matte coating on the surface. Interference information refers to the interference generated by the packaging or the item itself on the detection signal. For example, a glass bottle may cause ambient light noise to appear in the Raman spectrum due to surface reflection, and liquid sloshing may cause the baseline of the infrared spectrum to drift.

[0025] Step S1202: Compare the packaging information and interference information with the acquisition requirements of the first spectrum to determine the first matching degree of the first feature.

[0026] In some embodiments, the first matching degree is used to quantify the compatibility between the infrared spectrum features and the object features of the item to be detected. The first matching degree can be calculated based on the influence degree of the packaging information and interference information on the infrared spectrum acquisition requirements. Specifically, by analyzing the matching relationship between the object features and the infrared light penetration and anti-interference capabilities, converting key parameters into weight indicators, and combining with a preset threshold for weighted scoring, the first matching degree can be obtained. For example, if the item to be detected is an opaque plastic bottle, the penetration of infrared light is blocked, resulting in attenuation of the effective signal and an increase in the proportion of reflected noise. At this time, the system will calculate the first matching degree according to the influence degree of the opaque plastic bottle, and finally output a value in the range of 0 - 1. The higher the value, the more reliable the infrared spectrum features.

[0027] In some embodiments, the acquisition requirements refer to the core requirements put forward by different spectral technologies for the physical properties of the item to be measured to obtain effective signals. For example, infrared spectroscopy relies on light transmittance and low reflection interference, while Raman spectroscopy requires high scattering efficiency and low fluorescence background.

[0028] Step S1203: Compare the packaging information and interference information with the acquisition requirements of the second spectrum to determine the second matching degree of the second feature.

[0029] In some embodiments, the second matching degree is used to evaluate the suitability of the Raman spectrum features and the object features of the item to be detected. The second matching degree can be calculated based on the satisfaction degree of the packaging information and interference information with the Raman spectrum acquisition requirements. Specifically, by analyzing the correlation between the object features and the Raman light scattering efficiency and anti-fluorescence interference capabilities, converting key parameters into quantification indicators, and combining with a preset weight for comprehensive scoring, the second matching degree can be obtained. For example, if the item to be detected is a transparent glass bottle, its Raman signal intensity is significant and the characteristic peaks are clear. At this time, the system will calculate the matching degree score according to the scattering efficiency and fluorescence background intensity, and output a value in the range of 0 - 1. The higher the value, the more suitable the Raman spectrum features are for the current detection scenario.

[0030] In some embodiments, the matching degree calculation formula is: matching degree = Σ (parameter score × weight coefficient). Table 1 shows examples of partial object features and acquisition requirements. Referring to Table 1, through the mapping table of object features and acquisition requirements and the above matching degree calculation formula, the first matching degree and the second matching degree can be calculated. The parameter score can be comprehensively calculated in combination with the scoring level and the scoring range. For example, if object A is a transparent glass bottle, the scoring level of the light transmittance of the material in its infrared spectrum is high, and the scoring range is 0.8 - 1, and the corresponding specific light transmittance is 85%, then the corresponding score is 0.85 points; if object B is a colorless transparent quartz container, the scoring level of the fluorescence interference intensity in its Raman spectrum is low, and the scoring range is 0.8 - 1.0, and the corresponding fluorescence interference intensity is 6%, then when calculating the corresponding score, it is necessary to consider the correspondence between the range of fluorescence interference intensity corresponding to the scoring level and the scoring range. Since the scoring level of fluorescence interference intensity is low, its range is [0 - 10%], and the corresponding scoring range is [0.8 - 1.0]. After calculation, when the fluorescence interference intensity is 6%, the corresponding score is 0.88 points.

[0031] Table 1 Mapping Table of Object Features and Acquisition Requirements

[0032] By dynamically combining object features to analyze packaging information and interference information, and evaluating the matching degree according to the acquisition requirements of different spectral technologies, multi-dimensional detection optimization is achieved.

[0033] Step S130, based on the first matching degree and the second matching degree, perform feature fusion on the first feature and the second feature to obtain the spectral feature of the item to be detected.

[0034] In some embodiments, feature fusion refers to the process of integrating two types of features according to the matching degree. For example, when detecting a viscous liquid in a brown plastic bottle, the infrared matching degree is low due to the opaque packaging, while the Raman matching degree is high due to effective scattering. The system will assign a higher weight to the Raman feature, linearly fuse the two types of features, and finally generate a spectral feature including weighted peak intensity and peak position correction information.

[0035] See Figure 3 , the specific processing of the above step S130 may include steps S1301 - S1302, and the specific content is as follows: Step S1301, based on the first matching degree, perform feature screening on the first feature, and based on the second matching degree, perform feature screening on the second feature.

[0036] In some embodiments, feature screening involves selecting or weighting features based on the matching degree. For example, when the object to be measured is a metal can, due to the poor light transmittance of the metal can, the infrared matching degree is low, and the system will eliminate the band features affected by reflection interference.

[0037] Step S1302: According to the first matching degree and the second matching degree, perform feature fusion on the filtered first feature and the filtered second feature to obtain the spectral feature of the item to be detected.

[0038] In some embodiments, the filtered first feature refers to the effective infrared features selected according to the first matching degree. For example, the ester group peak at 1200 cm in the infrared spectrum that is not interfered by packaging reflection is retained. The filtered second feature refers to the effective Raman features retained according to the second matching degree. -1 The filtered second feature refers to the effective Raman features retained according to the second matching degree.

[0039] In some embodiments, feature fusion means integrating two types of features according to the matching degree. For example, the infrared matching degree is 0.2 and the Raman matching degree is 0.9. The intensities of the two feature peaks are fused into a composite spectral feature in a ratio of 2:9. The spectral feature refers to the fused multi-dimensional feature vector, such as including the weighted ester group peak intensity, S-O peak intensity, and joint peak position shift parameter, which is used to accurately characterize the components of hazardous chemicals.

[0040] In one implementation, the specific processing method of step S1302 is: based on the first matching degree and the second matching degree, determine the first credibility of the first feature and the second credibility of the second feature; according to the first credibility and the second credibility, perform feature fusion on the filtered first feature and the filtered second feature to obtain the spectral feature of the item to be detected.

[0041] In some embodiments, the first credibility and the second credibility are the feature reliability weights calculated according to the matching degree. The first credibility is the quotient of the first matching degree divided by the sum of the first matching degree and the second matching degree, and the second credibility is the quotient of the second matching degree divided by the sum of the first matching degree and the second matching degree. For example, when the infrared matching degree is 0.3 and the Raman matching degree is 0.8, the first credibility corresponding to the infrared is 0.27, and the second credibility corresponding to the Raman is 0.73, indicating that the Raman feature is more reliable.

[0042] In some embodiments, in the feature fusion stage, first, effective features need to be selected according to the matching degree, such as eliminating the invalid bands in the infrared spectrum due to metal packaging reflection interference, and then dynamically weighting and fusing the first feature and the second feature of the infrared spectrum and the Raman spectrum according to the credibility. The credibility, as a weight coefficient, directly determines the contribution ratio of each feature.

[0043] By dynamically fusing multi-spectral features, the robustness and accuracy of hazardous chemical detection are improved. By screening and eliminating the disturbed invalid features through matching degree, and at the same time assigning higher weights to the highly adaptable features according to the credibility, accurate weighted fusion is achieved. This method not only adapts to complex scenarios, but also makes up for the limitations of single technology through multi-spectral complementarity, and finally generates spectral features, which can effectively improve the recognition accuracy of hazardous chemicals.

[0044] Step S140: Input the spectral features into a pre-trained hazardous chemical recognition model to obtain the hazardous chemical recognition result of the item to be detected.

[0045] In some embodiments, the pre-trained hazardous chemical recognition model refers to a machine learning model trained with a large amount of hazardous chemical spectral data, which can learn the mapping relationship between spectral features and hazardous chemical categories. For example, a convolutional neural network is used to train the fused spectral data of thousands of hazardous chemicals, so that the model can distinguish ammonium nitrate from sulfuric acid. The hazardous chemical recognition result refers to the classification conclusion output by the model, including the substance category, hazard level, etc. For example, after inputting the fused spectral features of a certain liquid, the model output result is that the probability of concentrated sulfuric acid is 98%, and the hazard level: first-class corrosive.

[0046] It should be noted that the training process of the pre-trained hazardous chemical recognition model is as follows: First, a spectral enhancement technology is adopted to simulate diverse interference scenarios, and a synthetic data set covering different material packages and environmental interferences is constructed by adding noise and baseline drift; Second, a deep hybrid network architecture is designed, which combines the residual structure and the attention mechanism to hierarchically analyze the spectral features, uses the shallow network to extract the basic waveform features, the deep network to mine the co-variation law of molecular bonds, and introduces a domain adaptation module to eliminate device differences; Finally, a two-stage dynamic training strategy is adopted. First, the general visual feature extraction ability is migrated, and then the parameters are fine-tuned for the spectral characteristics. At the same time, an uncertainty quantification mechanism is integrated, and the risk of unknown substances is dynamically evaluated through probability modeling. This training mode enables the model to not only identify the subtle spectral differences of known hazardous chemicals, but also infer the new compound category through feature association, and achieve stable and reliable generalization performance in complex detection scenarios.

[0047] See Figure 4 , the specific processing of the above step S140 may include steps S1401 - S1402, and the specific content is as follows: Step S1401: Input the spectral features into a pre-trained hazardous chemical recognition model to obtain the composition information of the item to be detected.

[0048] In some embodiments, the composition information refers to the chemical composition and properties of the item to be detected, including the substance name, concentration, chemical bond characteristics, etc.

[0049] Step S1402: Obtain the identification result of hazardous chemicals of the item to be detected based on the composition information.

[0050] See Figure 5 , the specific processing of the above step S1402 may include steps S14021 - S14022, and the specific content is as follows: Step S14021: Compare the composition information with the composition information of each known hazardous chemical in the hazardous chemical information database to obtain a comparison result.

[0051] In some embodiments, the hazardous chemical information database is a database storing standardized data of known hazardous chemicals, covering information such as substance name, CAS number, composition ratio, and hazard class. For example, the entry of potassium nitrate in the database is marked as "oxidizing solid, hazard class 5.1, concentration ≥ 10% is regarded as a hazardous substance". The comparison result refers to the degree of association calculated between the component to be measured and the hazardous chemicals in the database. For example, the Raman peak at 783 cm -1 and the infrared peak at 1720 cm -1 of a certain liquid component completely match, and the similarity is as high as 98%.

[0052] Step S14022: Obtain the identification result of hazardous chemicals of the item to be detected based on the composition information and the comparison result.

[0053] In one implementation, the specific processing method of step S14022 is: according to the comparison result, determine the similarity between the composition information of the item to be detected and the composition information of each known hazardous chemical in the hazardous chemical information database; if there is a target known hazardous chemical whose similarity to the item to be detected exceeds the preset threshold, then determine the item to be detected as the target known hazardous chemical; if there is no target known hazardous chemical whose similarity to the item to be detected exceeds the preset threshold, then calculate the risk level of the item to be detected based on the composition information of the item to be detected, and obtain the identification result of hazardous chemicals of the item to be detected based on the risk level.

[0054] In some embodiments, the similarity refers to a matching value quantified by algorithms such as cosine similarity or spectral peak matching, such as the Raman peak matching degree of 85% between a certain gas and methane. The preset threshold is a critical value for determining whether there is a match. The risk level refers to a risk value calculated based on attributes such as the toxicity and flammability of the component, such as a certain mixture containing 50% ethanol, and the risk level is 0.8. The identification result of hazardous chemicals refers to the final classification conclusion, such as being determined as "ammonium nitrate, explosive" or marked as "unknown liquid with high risk level".

[0055] By inputting multi-spectral fusion features into a deep learning model, the intelligent level and detection accuracy of hazardous chemical identification are significantly improved. First, spectral features are analyzed through a convolutional neural network to accurately extract substance composition information. Second, a two-way comparison is carried out in combination with a hazardous chemical information database, and algorithms such as cosine similarity are used to achieve high-precision matching, effectively solving the problem of difficult identification of new chemicals and modified chemicals by traditional methods.

[0056] In some embodiments, after obtaining the hazardous chemical identification result, corresponding alarm information can also be sent. Specifically, after obtaining the hazardous chemical identification result, the hazard category and hazard level of the item to be detected can be determined according to the hazardous chemical identification result; and corresponding alarm information for the item to be detected can be sent according to the hazard category and hazard level.

[0057] Specifically, the hazard category refers to the type of hazard divided according to chemical properties, including explosiveness, flammability, and corrosiveness. The hazard level is a hierarchical index used to quantify the degree of hazard. For example, level one indicates a relatively high degree of hazard, and level two indicates a moderate degree of hazard. The alarm information is a graded response signal triggered by hazard assessment, including audible and visual alarms, emergency instructions, etc. For example, detecting a level one explosive triggers an evacuation alarm for the entire area, and a level two corrosive triggers a warning light in the isolation area.

[0058] An infrared and Raman spectroscopy complementary acquisition strategy is adopted, and feature adaptive screening is realized through matching degree evaluation, which can significantly reduce the interference of mixtures and the influence of packaging materials; a spectral feature fusion model based on deep learning is constructed, combined with a two-way comparison mechanism of a hazardous chemical information database, breaking through the coverage limitation of traditional databases, not only improving the detection sensitivity to low-concentration substances, but also being able to identify new chemicals and modified chemicals through feature correlation reasoning; an object feature-driven credibility evaluation system is introduced to realize dynamic weight allocation of multi-source features, so that the detection system maintains high robustness in complex scenarios. Therefore, this application can effectively improve the detection accuracy of hazardous chemicals and has the ability to identify new chemicals and modified chemicals.

[0059] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0060] The following is an apparatus embodiment of the present invention. For details not described in detail, reference can be made to the corresponding method embodiments above.

[0061] Figure 6 The structural schematic diagram of a hazardous chemical identification apparatus based on multi-spectral fusion provided by an embodiment of the present invention is shown. For the sake of convenience of description, only parts related to the embodiment of the present invention are shown, and are described in detail as follows: As Figure 6As shown in the figure, the hazardous chemical identification device 6 based on multispectral fusion includes: An extraction module 61, configured to respectively perform feature extraction on the first spectrum and the second spectrum of the item to be detected, and obtain the first feature of the first spectrum and the second feature of the second spectrum; A determination module 62, configured to determine the first matching degree of the first feature and the second matching degree of the second feature according to the object feature of the item to be detected; A fusion module 63, configured to perform feature fusion on the first feature and the second feature based on the first matching degree and the second matching degree, and obtain the spectral feature of the item to be detected; An identification module 64, configured to input the spectral feature into a pre-trained hazardous chemical identification model, and obtain the hazardous chemical identification result of the item to be detected.

[0062] In a possible implementation manner, the extraction module 61 is specifically configured to: perform spectral detection on the item to be detected to obtain the initial first spectrum and the initial second spectrum of the item to be detected; perform preprocessing on the initial first spectrum and the initial second spectrum to obtain the first spectrum and the second spectrum; wherein, the preprocessing includes data denoising processing and baseline correction processing.

[0063] In a possible implementation manner, the determination module 62 is specifically configured to: determine the packaging information and the interference information of the item to be detected based on the object feature of the item to be detected; compare the packaging information and the interference information with the acquisition requirements of the first spectrum to determine the first matching degree of the first feature; compare the packaging information and the interference information with the acquisition requirements of the second spectrum to determine the second matching degree of the second feature.

[0064] In a possible implementation manner, the fusion module 63 is specifically configured to: perform feature screening on the first feature based on the first matching degree, and perform feature screening on the second feature based on the second matching degree; perform feature fusion on the screened first feature and the screened second feature according to the first matching degree and the second matching degree, and obtain the spectral feature of the item to be detected.

[0065] In a possible implementation manner, the fusion module 63 is further configured to: determine the first credibility of the first feature and the second credibility of the second feature based on the first matching degree and the second matching degree; perform feature fusion on the screened first feature and the screened second feature according to the first credibility and the second credibility, and obtain the spectral feature of the item to be detected.

[0066] In a possible implementation manner, the identification module 64 is specifically configured to: input the spectral feature into a pre-trained hazardous chemical identification model to obtain the component information of the item to be detected; obtain the hazardous chemical identification result of the item to be detected based on the component information.

[0067] In a possible implementation, the recognition module 64 is further configured to: compare the ingredient information with the ingredient information of each known hazardous chemical in the hazardous chemical information database to obtain a comparison result; and obtain the hazardous chemical recognition result of the item to be detected based on the ingredient information and the comparison result.

[0068] In a possible implementation, the recognition module 64 is further configured to: determine the similarity between the ingredient information of the item to be detected and the ingredient information of each known hazardous chemical in the hazardous chemical information database according to the comparison result; if there is a target known hazardous chemical whose similarity to the item to be detected exceeds a preset threshold, determine the item to be detected as the target known hazardous chemical; if there is no target known hazardous chemical whose similarity to the item to be detected exceeds the preset threshold, calculate the risk level of the item to be detected based on the ingredient information of the item to be detected, and obtain the hazardous chemical recognition result of the item to be detected based on the risk level.

[0069] In a possible implementation, the recognition module 64 is further configured to: determine the hazard category and hazard level of the item to be detected based on the hazardous chemical recognition result; and send out the corresponding alarm information of the item to be detected according to the hazard category and hazard level.

[0070] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. If there is no special explanation and logical conflict, the terms and / or descriptions between different embodiments are consistent and can be mutually referred to, and the technical features in different embodiments can be combined to form a new embodiment according to their inherent logical relationships.

[0071] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the same; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and shall all be included in the protection scope of the present invention.

Claims

1. A method for identifying hazardous chemicals based on multispectral fusion, characterized in that, Including: Performing feature extraction on the first spectrum and the second spectrum of the item to be detected respectively to obtain the first feature of the first spectrum and the second feature of the second spectrum; Determining the first matching degree of the first feature and the second matching degree of the second feature according to the object feature of the item to be detected; Based on the first matching degree and the second matching degree, performing feature fusion on the first feature and the second feature to obtain the spectral feature of the item to be detected; Inputting the spectral feature into a pre-trained hazardous chemical identification model to obtain the hazardous chemical identification result of the item to be detected.

2. The method for identifying hazardous chemicals based on multispectral fusion according to claim 1, wherein The step of performing feature fusion on the first feature and the second feature based on the first matching degree and the second matching degree to obtain the spectral feature of the item to be detected includes: Performing feature screening on the first feature based on the first matching degree and performing feature screening on the second feature based on the second matching degree; According to the first matching degree and the second matching degree, performing feature fusion on the screened first feature and the screened second feature to obtain the spectral feature of the item to be detected.

3. The method for identifying hazardous chemicals based on multispectral fusion according to claim 2, wherein The step of performing feature fusion on the screened first feature and the screened second feature according to the first matching degree and the second matching degree to obtain the spectral feature of the item to be detected includes: Based on the first matching degree and the second matching degree, determining the first credibility of the first feature and the second credibility of the second feature; According to the first credibility and the second credibility, performing feature fusion on the screened first feature and the screened second feature to obtain the spectral feature of the item to be detected.

4. The method for identifying hazardous chemicals based on multispectral fusion according to claim 1, wherein The step of determining the first matching degree of the first feature and the second matching degree of the second feature according to the object feature of the item to be detected includes: Based on the object feature of the item to be detected, determining the packaging information and interference information of the item to be detected; Comparing the packaging information and the interference information with the acquisition requirements of the first spectrum to determine the first matching degree of the first feature; Comparing the packaging information and the interference information with the acquisition requirements of the second spectrum to determine the second matching degree of the second feature.

5. The method for identifying hazardous chemicals based on multispectral fusion according to claim 1, wherein The step of inputting the spectral feature into a pre-trained hazardous chemical identification model to obtain the hazardous chemical identification result of the item to be detected includes: Inputting the spectral feature into a pre-trained hazardous chemical identification model to obtain the component information of the item to be detected; Based on the component information, obtaining the hazardous chemical identification result of the item to be detected.

6. The method for identifying hazardous chemicals based on multispectral fusion according to claim 5, wherein The step of obtaining the hazardous chemical identification result of the item to be detected based on the component information includes: Comparing the component information with the component information of each known hazardous chemical in the hazardous chemical information database to obtain a comparison result; According to the component information and the comparison result, obtaining the hazardous chemical identification result of the item to be detected.

7. The method for identifying hazardous chemicals based on multispectral fusion according to claim 6, wherein The step of obtaining the hazardous chemical identification result of the item to be detected according to the component information and the comparison result includes: According to the comparison result, determine the similarity between the component information of the item to be detected and the component information of each known hazardous chemical in the hazardous chemical information database; If there is a target known hazardous chemical whose similarity to the item to be detected exceeds a preset threshold, then determine the item to be detected as the target known hazardous chemical; If there is no target known hazardous chemical whose similarity to the item to be detected exceeds a preset threshold, then calculate the risk level of the item to be detected based on the component information of the item to be detected, and obtain the hazardous chemical identification result of the item to be detected based on the risk level.

8. The method for identifying hazardous chemicals based on multispectral fusion according to claim 1, wherein The method further includes: Based on the hazardous chemical identification result, determine the hazard class and hazard level of the item to be detected; According to the hazard class and the hazard level, send out the corresponding alarm information for the item to be detected.

9. The method for identifying hazardous chemicals based on multispectral fusion according to claim 1, characterized in that The method further includes: Perform spectral detection on the item to be detected to obtain an initial first spectrum and an initial second spectrum of the item to be detected; Perform preprocessing on the initial first spectrum and the initial second spectrum to obtain a first spectrum and a second spectrum; wherein, the preprocessing includes data denoising processing and baseline correction processing.

10. A hazardous chemical identification device based on multispectral fusion, characterized in that, It includes: An extraction module, configured to perform feature extraction on the first spectrum and the second spectrum of the item to be detected respectively, to obtain a first feature of the first spectrum and a second feature of the second spectrum; A determination module, configured to determine a first matching degree of the first feature and a second matching degree of the second feature according to the object feature of the item to be detected; A fusion module, configured to perform feature fusion on the first feature and the second feature based on the first matching degree and the second matching degree, to obtain a spectral feature of the item to be detected; An identification module, configured to input the spectral feature into a pre-trained hazardous chemical identification model to obtain the hazardous chemical identification result of the item to be detected.

Images