Agricultural product PFAS detection method based on multi-source spectrum image fusion

Abstract

This invention discloses a PFAS detection method for agricultural products based on multi-source spectral fusion, belonging to the field of agricultural product contaminant detection technology. The method includes: acquiring multi-source spectral response maps formed by pretreated agricultural product extracts in a PFAS spectrochemical sensor array; associating and labeling these maps according to spectral sensing channels, detection locations, and acquisition sequences to generate fingerprint-calibrated multi-source spectra; establishing a PFAS fingerprint anchor point set based on the correspondence between the fingerprint-calibrated multi-source spectra and preset PFAS structural spectral segments; and using the PFAS fingerprint anchor point set to register local spectral segments of each spectral sensing channel to obtain a PFAS fingerprint aligned spectral map set. This invention achieves accurate correspondence of the same PFAS structural response in different spectral sources by establishing the correspondence between PFAS structural spectral segments and the PFAS fingerprint anchor point set, reducing mismatches between heterogeneous spectral segments and matrix background interference, thus making PFAS feature recognition more stable.

Description

Technical Field

[0001] This invention relates to the field of agricultural product contaminant detection technology, and in particular to a PFAS detection method for agricultural products based on multi-source spectral fusion. Background Technology

[0002] The detection of perfluorinated and polyfluorinated alkyl substances (PFAS) in agricultural products has become an important direction in food safety and the regulation of new pollutants. PFAS are characterized by structural stability, strong environmental persistence, complex migration pathways, and potential cumulative risks. They may exist in fruits, vegetables, grains, aquatic products, and processed products in low concentrations and in complex matrices. With the development of spectral analysis, chemical sensors, and multi-source data processing technologies, the detection of PFAS in agricultural products is gradually shifting from single-response identification to multi-source spectral joint analysis. By obtaining information on the structural response, intensity changes, and matrix effects of target pollutants through different spectral sensing channels, a technical foundation is provided for rapid screening and risk assessment of PFAS.

[0003] However, existing methods have some shortcomings. The response mechanisms, spectral ranges, and signal morphologies of different spectral sensing channels vary. If only full-spectrum splicing or uniform dimensionality reduction is performed, local fingerprint features related to the PFAS structure are easily mismatched, causing the effective response to be masked by the background of the agricultural product matrix or irrelevant spectral bands. In addition, the response stability, background interference level, and PFAS structure matching degree of different spectral sources in complex extracts are not consistent. Fixed-weight fusion or simple average fusion is difficult to distinguish between high-confidence and low-confidence responses, and abnormal channels or noise fluctuations can easily affect PFAS category identification, concentration range estimation, and suspected risk judgment. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a PFAS detection method for agricultural products based on multi-source spectral fusion to solve the problems of difficulty in aligning heterogeneous spectral bands and insufficient differentiation of spectral source reliability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a PFAS detection method for agricultural products based on multi-source spectral fusion, comprising: acquiring multi-source spectral response maps formed by pretreated agricultural product extracts in a PFAS spectrochemical sensor array, and associating and labeling them according to spectral sensing channels, detection locations, and acquisition sequences to generate fingerprint-calibrated multi-source spectra; establishing a PFAS fingerprint anchor point set based on the correspondence between the fingerprint-calibrated multi-source spectra and preset PFAS structural spectral segments, and using the PFAS fingerprint anchor point set to register local spectral segments of each spectral sensing channel to obtain a PFAS fingerprint-aligned spectral map set; extracting PFAS target response features based on the PFAS fingerprint-aligned spectral map set, and determining the response stability... The source confidence weights are generated by considering the degree of background interference suppression and the degree of fingerprint anchor matching. The PFAS target response features are then fused according to these source confidence weights to form source confidence fusion features. Based on these source confidence fusion features, candidate results for PFAS categories, candidate results for concentration ranges, and suspected PFAS risk markers in agricultural product extracts are determined. The source consistency status is then determined based on the support relationship between each spectral sensing channel and the candidate results for PFAS categories, candidate results for concentration ranges, and suspected PFAS risk markers. When the source consistency status is consistent, valid agricultural product PFAS detection information is generated. When the source consistency status is inconsistent, agricultural product PFAS detection information to be verified is generated.

[0007] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the specific steps for generating the fingerprint-calibrated multi-source spectral image are as follows: The pretreated agricultural product extract is fed into the PFAS spectrochemical sensor array, so that the agricultural product extract makes detection contact with each spectral sensing channel, forming a spectral response area to be collected; Based on the spectral response region to be acquired, the spectral response corresponding to each spectral sensing channel is acquired respectively, and background subtraction, baseline correction and intensity unification processing are performed on each spectral response to obtain the channel-corrected spectral response map. Based on the spectral sensing channel, detection position, and acquisition sequence corresponding to the channel calibration spectral response map, an association marker is generated and written into the channel calibration spectral response map. This establishes a correspondence between the responses of each spectral sensing channel at the same detection position and acquisition sequence, generating a fingerprint calibration multi-source spectrum.

[0008] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the specific steps for obtaining the PFAS fingerprint aligned spectral set are as follows: By fingerprinting the correspondence between the spectral sensing channel markers in the multi-source spectrum and the preset PFAS structure spectral segments, the range of target spectral segments corresponding to the PFAS structure in each spectral sensing channel is determined, and the PFAS target spectral segment index is obtained. Based on the PFAS target spectral band index, local spectral bands corresponding to each spectral sensing channel are extracted from the fingerprint calibration multi-source spectrum, and the peak position, peak shape and response change characteristics in the local spectral bands are matched with the correspondence of PFAS structural spectral bands to establish a PFAS fingerprint anchor point set. Based on the PFAS fingerprint anchor point set, peak position correction and spectral registration are performed on local spectral segments to establish alignment relationships between local spectral segments corresponding to the same PFAS structure, resulting in a PFAS fingerprint aligned spectral set.

[0009] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the specific steps for forming the spectral source confidence fusion feature are as follows: Based on the PFAS fingerprint aligned spectral atlas, the peak position change, peak intensity change, peak shape change and spectral band response area corresponding to the PFAS fingerprint anchor point in each spectral sensing channel are extracted to form the PFAS target response features. Based on the target response characteristics of PFAS, the response stability, background interference suppression degree and fingerprint anchor point matching degree of each spectral sensing channel are determined respectively, and the source confidence weight is generated according to the response stability, background interference suppression degree and fingerprint anchor point matching degree. The PFAS target response features of each spectral sensing channel at the same detection location and the same acquisition time sequence are fused according to the source confidence weight to form the source confidence fusion feature.

[0010] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the steps of generating valid PFAS detection information for agricultural products when the spectral source consistency status is consistent, and generating PFAS detection information for agricultural products to be verified when the spectral source consistency status is inconsistent are as follows: Based on the correspondence between the source confidence fusion characteristics and the preset PFAS category spectral fingerprint records and the preset PFAS concentration range response results, candidate results for PFAS categories and candidate results for concentration ranges in agricultural product extracts are determined. Based on the candidate results of PFAS categories and concentration ranges, and combined with the response content in the source confidence fusion features that correspond to the PFAS fingerprint anchors but are not classified into the candidate results of PFAS categories, suspected PFAS risk markers are identified. The consistency status of the spectral source is determined based on the support relationship between each spectral sensing channel and the candidate results of PFAS category, candidate results of concentration range, and suspected PFAS risk markers. When the spectral source consistency status is consistent, valid PFAS detection information for agricultural products is generated; when the spectral source consistency status is inconsistent, PFAS detection information for agricultural products to be verified is generated.

[0011] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the steps for establishing the preset PFAS structure spectral segment correspondence are as follows: Known PFAS standard samples were selected as calibration objects, and standard spectral response maps of the known PFAS standard samples were acquired in the PFAS spectrochemical sensor array. The standard spectral response map is subjected to background subtraction, baseline correction and intensity unification to obtain the standard channel corrected spectral response map; When the same PFAS representative substance shows the corresponding spectral segment response in multiple repeated acquisitions, and the corresponding spectral segment response does not form a stable response in the same direction in the blank spectral response map, the current spectral segment position is determined as the stable spectral segment position. The stable spectral band positions are mapped to the corresponding PFAS structure names to determine the range of structural characteristic spectral bands for each PFAS structure name in each spectral sensing channel. Establish corresponding records for PFAS structure names, applicable spectral sensing channels, and structural characteristic spectral ranges to form a PFAS structure spectral range correspondence.

[0012] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the specific steps for establishing the PFAS fingerprint anchor point set are as follows: According to the target spectral range in the PFAS target spectral index, a local spectral segment is extracted from the channel correction spectral response map corresponding to the fingerprint calibration multi-source spectrum. The position of the spectral segment with the largest continuous response change amplitude in the local spectral segment is determined as the peak position. The width, symmetry and continuity of the response change on both sides of the peak position are determined as the peak shape. The direction of enhancement or weakening of the peak position relative to the background response at both ends of the local spectral segment is determined as the response change characteristic. The peak position, peak shape and response change characteristics of the local spectral segment are matched with the correspondence of the PFAS structural spectral segment, and the local spectral segment that meets the matching relationship is determined as the PFAS candidate fingerprint anchor point. PFAS candidate fingerprint anchors from different spectral sensing channels and pointing to the same PFAS structure name at the same detection location and under the same acquisition time sequence are grouped together to form a PFAS fingerprint anchor set.

[0013] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the specific steps for peak position correction and spectral registration of local spectral bands are as follows: Based on the PFAS structure name, detection location marker, and acquisition time sequence marker, the PFAS candidate fingerprint anchor points in the PFAS fingerprint anchor point set are grouped to form anchor point groups. Within each anchor group, the peak position of the PFAS candidate fingerprint anchor point is compared with the position of the stable spectral segment with the same PFAS structure name in the PFAS structure spectral segment correspondence to determine the peak position shift direction and peak position shift amount. Based on the peak position shift direction and peak position shift amount, the peak position of the local spectral segment in the corresponding spectral sensing channel is shifted so that the peak position in the local spectral segment is aligned with the position of the stable spectral segment corresponding to the PFAS structure name. Using the PFAS structure name as the registration reference, spectral registration is performed on the local spectral segments that have completed peak position correction to form an alignment record; A PFAS fingerprint alignment unit is formed by multiple alignment records at the same detection location and under the same acquisition time sequence, and a PFAS fingerprint alignment spectrum set is formed by multiple PFAS fingerprint alignment units.

[0014] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the specific steps for generating spectral source confidence weights are as follows: Each spectral sensing channel under the same detection location and the same acquisition sequence was used as a comparison object; Based on the differences in peak position variation, peak intensity variation direction and the continuity of spectral band response area variation in the PFAS target response characteristics, the response stability of each spectral sensing channel is ranked. Based on the peak intensity variation, spectral band response area, average level of response intensity at both ends of local spectral bands, and response fluctuations on both sides of the peak position in the PFAS target response characteristics, the background interference suppression degree of each spectral sensing channel is ranked. Based on the enhancement or weakening directions reflected by the peak position change, peak shape change and peak intensity change in the PFAS target response characteristics, the fingerprint anchor point matching degree of each spectral sensing channel is ranked. The ranking of response stability, background interference suppression, and fingerprint anchor matching is converted into a ranking score, and the ranking scores of the same spectral sensing channel in the three rankings are summed to obtain the comprehensive confidence score of the corresponding spectral sensing channel. The source confidence weight of the corresponding spectral sensing channel is obtained by calculating the ratio of the overall confidence score of each spectral sensing channel to the sum of the overall confidence scores of all spectral sensing channels.

[0015] As a preferred embodiment of the PFAS detection method for agricultural products based on multi-source spectral fusion described in this invention, the spectral source confidence fusion feature includes multiple single-group spectral source confidence fusion features, each single-group spectral source confidence fusion feature corresponding to a detection location marker, an acquisition time sequence marker, and a PFAS structure name; The single-source confidence fusion features include fusion peak position variation, fusion peak intensity variation, fusion peak shape variation, fusion spectral band response area, and fusion response direction; The fusion peak position change, fusion peak intensity change, fusion peak shape change, and fusion spectral band response area are obtained by weighting the peak position change, peak intensity change, peak shape change, and spectral band response area of ​​each spectral sensing channel within the same fusion group according to the corresponding spectral source confidence weight. The fusion response direction is obtained by weighting the response change direction of each spectral sensing channel within the same fusion group according to the corresponding spectral source confidence weight.

[0016] The beneficial effects of this invention are as follows: By establishing the correspondence between PFAS structural spectral segments and the PFAS fingerprint anchor set, and using the PFAS fingerprint anchor set to perform peak position correction and spectral segment registration for local spectral segments of each spectral sensing channel, accurate correspondence of the same PFAS structural response in different spectral sources is achieved, reducing mismatch of heterogeneous spectral segments and matrix background interference, making PFAS feature recognition more stable; by generating spectral source confidence weights based on response stability, background interference suppression degree and fingerprint anchor matching degree, and fusing the PFAS target response features of each spectral sensing channel according to the spectral source confidence weights, the distinction and weighting of different spectral source confidence levels are achieved, reducing the impact of abnormal channels and noise fluctuations on detection results, and improving the reliability of PFAS category identification, concentration range judgment and suspected risk labeling. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a PFAS detection method for agricultural products based on multi-source spectral fusion.

[0019] Figure 2 Flowchart for generating multi-source spectra for fingerprint calibration.

[0020] Figure 3 Flowchart for PFAS fingerprint anchor registration.

[0021] Figure 4 This is a flowchart for source confidence fusion and consistency determination. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a PFAS detection method for agricultural products based on multi-source spectral fusion, comprising the following steps: S1. Collect the multi-source spectral response map formed by the pretreated agricultural product extract in the PFAS spectrochemical sensor array, and associate and mark it according to the spectral sensing channel, detection position and acquisition time sequence to generate fingerprint calibration multi-source spectrum.

[0026] S1.1 The pretreated agricultural product extract is fed into the PFAS spectrochemical sensor array, so that the agricultural product extract makes detection contact with each spectral sensing channel, forming the spectral response area to be collected.

[0027] It should be noted that the PFAS spectrochemical sensor array includes an inlet, a dispensing channel, a spectral sensing channel, and an outlet. The spectral sensing channel includes a Raman-enhanced spectroscopy sensing channel, an infrared absorption spectroscopy sensing channel, a fluorescence response spectroscopy sensing channel, and a hyperspectral reflectance spectroscopy sensing channel. Each spectral sensing channel has a detection position, and the surface of the detection position is provided with a PFAS chemical recognition layer for forming detection contact with the agricultural product extract. The PFAS chemical recognition layer can be one of a fluorine-containing affinity material, an ion exchange material, or a porous adsorption material. The PFAS chemical recognition layer is used to enrich or retain PFAS-related components in the agricultural product extract and to generate collectable spectral response differences in the corresponding spectral sensing channel for the PFAS-related components. The spectral response differences include at least one of Raman enhancement response, infrared absorption change, fluorescence intensity change, or reflectance spectral shape change.

[0028] The agricultural product samples to be tested are sampled, crushed, and homogenized. The extraction solvent is selected according to the water content, lipid content, pigment content, and PFAS category of the agricultural product to be tested. The extraction solvent can be acetonitrile, methanol, water-acetonitrile mixture, water-methanol mixture, or water-organic solvent mixture containing buffer salts. For fruit and vegetable samples with high water content, acetonitrile or methanol is preferred as the extraction solvent. For grain, aquatic products, and processed product samples with high lipid or pigment content, water-acetonitrile mixture or water-methanol mixture is used, and filtration is used to reduce interference from non-target matrices. The sample after adding the extraction solvent is shaken for extraction, and the liquid phase after extraction is subjected to solid-liquid separation and filtration to remove particulate residues and suspended impurities that affect spectral acquisition, resulting in a pretreated agricultural product extract.

[0029] The pretreated agricultural product extract is uniformly mixed and fed into the injection end of the PFAS spectrochemical sensor array. The agricultural product extract is distributed to each spectral sensing channel along a fixed flow direction through the distribution channel. In each spectral sensing channel, the agricultural product extract continuously covers the corresponding detection position and forms detection contact with the PFAS chemical recognition layer on the surface of the detection position. The detection position that has formed detection contact is determined as the spectral response area to be collected.

[0030] S1.2. Based on the spectral response region to be acquired, the spectral response corresponding to each spectral sensing channel is acquired respectively, and background subtraction, baseline correction and intensity unification processing are performed on each spectral response to obtain the channel-corrected spectral response map.

[0031] It should be noted that each spectral sensing channel acquires the corresponding spectral response region to be acquired, thus obtaining the initial spectral response map of each spectral sensing channel.

[0032] Before acquiring the initial spectral response map of each spectral sensing channel, the background spectral response of each spectral sensing channel that has not been in contact with the agricultural product extract is acquired. The initial spectral response map of each spectral sensing channel and the corresponding channel background spectral response are then subjected to background subtraction to remove non-sample responses caused by the dark response of the light source, the background of the optical path, and the background of the detection position itself, thus obtaining the background-subtracted spectral response map.

[0033] The spectral segment with continuous response change and no sharp peak abrupt change is selected from the background spectral response of the channel as the baseline reference spectral segment, and the response value of the baseline reference spectral segment in the background subtraction spectral response map is used as the baseline control point. Adjacent baseline control points are connected in the order of the spectral segments to form a piecewise linear baseline curve, and the piecewise linear baseline curve is used as the baseline drift trend. The baseline drift trend is subtracted from the background subtraction spectral response map to obtain the baseline correction spectral response map.

[0034] The average response intensity of the baseline reference spectral band in the background spectral response of the same spectral sensing channel is used as the reference response intensity. The ratio of the response intensity at each spectral band position in the baseline-corrected spectral response map to the reference response intensity is calculated to obtain the response intensity at a uniform intensity scale. This ensures that the spectral responses at different detection positions and acquisition times are on the same intensity scale, resulting in the channel-corrected spectral response map, expressed as follows: in This represents the channel-corrected spectral response diagram; This represents the response intensity of the initial spectral response plot; Indicates the background spectral response of the channel; Indicates the baseline drift trend; Indicates the reference response strength; Indicates the spectral sensing channel number; Indicates the detection location number; Indicates the data acquisition time sequence number; This indicates the spectral band position. The above formula is used to explain the formation process of the channel correction spectral response map, that is, subtracting the channel background spectral response and baseline drift trend from the initial spectral response map, and using the reference response intensity to unify the intensity, so that the spectral response at different detection positions and different acquisition sequences is on the same intensity scale.

[0035] S1.3. Generate association markers based on the spectral sensing channels, detection positions, and acquisition sequences corresponding to the channel calibration spectral response map, and write the association markers into the channel calibration spectral response map to establish a correspondence between the responses of each spectral sensing channel under the same detection position and the same acquisition sequence, and generate a fingerprint calibration multi-source spectrum.

[0036] It should be noted that the spectral sensing channel name is used to indicate that the channel-corrected spectral response map originates from the Raman enhanced spectral sensing channel, infrared absorption spectral sensing channel, fluorescence response spectral sensing channel, or hyperspectral reflectance spectral sensing channel. The detection position record is used to indicate the contact position of the agricultural product extract in the PFAS spectrochemical sensor array. The acquisition time sequence record is used to indicate the acquisition sequence of the channel-corrected spectral response map.

[0037] Furthermore, the spectral sensing channel name, detection position record, and acquisition timing record corresponding to each channel calibration spectral response map are used as spectral sensing channel markers, detection position markers, and acquisition timing markers, respectively, and are combined in the order of "spectral sensing channel marker - detection position marker - acquisition timing marker" to form an associated marker.

[0038] The associated markers are written into the corresponding channel calibration spectral response map, and the channel calibration spectral response maps are aggregated according to the detection location markers and acquisition time sequence markers, so that the channel calibration spectral response maps from different spectral sensing channels at the same detection location and the same acquisition time sequence are aggregated into the same group of multi-source spectral responses.

[0039] The same set of multi-source spectral responses is used as a fingerprint calibration unit, and multiple fingerprint calibration units constitute a fingerprint calibration multi-source spectrum.

[0040] S2. Based on the fingerprint calibration multi-source spectrum and the pre-defined PFAS structure spectral segment correspondence, establish a PFAS fingerprint anchor point set, and use the PFAS fingerprint anchor point set to register the local spectral segments of each spectral sensing channel to obtain the PFAS fingerprint aligned spectrum set.

[0041] S2.1 By fingerprinting the correspondence between the spectral sensing channel markers in the multi-source spectrum and the preset PFAS structure spectral segments, the target spectral segment range corresponding to the PFAS structure in each spectral sensing channel is determined, and the PFAS target spectral segment index is obtained.

[0042] It should be noted that the pre-defined PFAS structure-spectral correspondence is established before the detection of agricultural product extracts. During the establishment process, known PFAS standard samples are selected as calibration objects. The known PFAS standard samples include at least one representative PFAS substance with at least one of the following structures: perfluoroalkyl chain, carbon-fluorine bond, sulfonic acid group, and carboxylic acid group. The known PFAS standard samples are sent into the PFAS spectrochemical sensor array, and the standard spectral response diagrams corresponding to each spectral sensing channel are acquired.

[0043] Background subtraction, baseline correction, and intensity unification are performed on the standard spectral response map to obtain the standard channel calibrated spectral response map. In the standard channel calibrated spectral response map, the same PFAS representative substance is repeatedly sampled at least three times. When the same spectral segment position is within the same candidate spectral segment range in the repeated sampling, and the peak position deviation does not exceed the allowable peak position fluctuation range obtained from the repeated sampling of the standard sample, the response direction remains consistent, and the peak intensity change does not exceed the allowable intensity fluctuation range obtained from the repeated sampling of the standard sample, this spectral segment position is determined as the candidate stable spectral segment position. The allowable peak position fluctuation range and allowable intensity fluctuation range are determined by the repeated sampling results of known PFAS standard samples, and can be formed using the mean and standard deviation of the repeated sampling data. Further, a blank extract is fed into the PFAS spectrochemical sensor array, and a blank spectral response map is acquired. If the candidate stable spectral segment position does not show a continuous and stable response change in the same direction as the PFAS representative substance in the blank spectral response map, this candidate stable spectral segment position is determined as the stable spectral segment position, and a correspondence is established between the stable spectral segment position and the corresponding PFAS structure name to form the structural characteristic spectral segment range of the corresponding PFAS structure name.

[0044] By fingerprinting the spectral sensing channel markers in the multi-source spectrum, the spectral sensing channel markers are matched with the applicable spectral sensing channels in the PFAS structure spectral segment correspondence, thus determining the target spectral segment range corresponding to the PFAS structure in each spectral sensing channel.

[0045] The spectral sensing channel markers, detection location markers, acquisition timing markers, PFAS structure names, and target spectral ranges in the same fingerprint calibration unit are registered accordingly, so that each target spectral range can be associated with a specific spectral sensing channel, detection location, acquisition timing, and PFAS structure name. The PFAS target spectral range index is formed by each target spectral range and its corresponding registration content.

[0046] S2.2 Based on the PFAS target spectral band index, extract the local spectral bands corresponding to each spectral sensing channel from the fingerprint calibration multi-source spectrum, and match the peak position, peak shape and response change characteristics in the local spectral bands with the preset PFAS structure spectral band correspondence to establish a PFAS fingerprint anchor point set.

[0047] It should be noted that, according to the target spectral range in the PFAS target spectral index, a local spectral segment is extracted from the channel correction spectral response map corresponding to the fingerprint calibration multi-source spectrum, so that the local spectral segment retains the corresponding spectral sensing channel mark, detection position mark, acquisition timing mark and PFAS structure name.

[0048] When extracting features from a local spectral band, the local background response is determined based on the response intensities at both ends of the local spectral band. Within the target spectral band, the position of the spectral band with the largest absolute change relative to the local background response is located, and this position is designated as the peak. The direction of response change is determined based on the relationship between the peak response intensity and the local background response. The direction of response change includes both enhancement and attenuation directions, expressed as follows:

[0049] in, Peak position in the spectrum segment; This represents the average level of the response intensity at both ends of a local spectral band; This indicates the position where the response difference within the parentheses reaches its maximum. Indicates the PFAS structure name number; Indicates the PFAS structure name In the spectral sensing channel The target spectral range; and Indicates the target spectral range The left and right spectral segments; The value indicates the direction of response change; a positive value indicates an enhancement direction, and a negative value indicates a weakening direction. This formula is used to determine the peak position in a local spectral band. The local background response is calculated using the response intensities at both ends of the target spectral band. The position of the spectral band with the largest change relative to the local background response within the target spectral band is then identified as the peak position.

[0050] The width, symmetry, and continuity of the response changes on both sides of the peak position are defined as the peak shape; the direction of enhancement or weakening of the background response relative to the two ends of the local spectral band is defined as the response change characteristic. The peak position, peak shape, and response change characteristics corresponding to the local spectral band are matched with the correspondence between the PFAS structure spectral bands; when the peak position falls within the structural characteristic spectral band range of the corresponding PFAS structure name, the peak shape corresponds to the stable spectral band shape of the corresponding PFAS structure name in the standard channel calibrated spectral response diagram, and the response change characteristic corresponds to the response direction of the corresponding PFAS structure name, the local spectral band is determined as a PFAS candidate fingerprint anchor point.

[0051] PFAS candidate fingerprint anchors from different spectral sensing channels and pointing to the same PFAS structure name at the same detection location and at the same acquisition time are aggregated. The spectral sensing channel marker, detection location marker, acquisition time sequence marker, PFAS structure name, target spectral range, peak position, peak shape and response change characteristics corresponding to the PFAS candidate fingerprint anchors are included in the aggregation results to form a PFAS fingerprint anchor set.

[0052] S2.3. Based on the PFAS fingerprint anchor point set, perform peak position correction and spectral registration on local spectral segments to establish alignment relationships between local spectral segments corresponding to the same PFAS structure, and obtain the PFAS fingerprint aligned spectral set.

[0053] It should be noted that during peak position correction, the PFAS candidate fingerprint anchors in the PFAS fingerprint anchor set are grouped according to the PFAS structure name, detection location marker, and acquisition time sequence marker. This ensures that PFAS candidate fingerprint anchors pointing to the same PFAS structure name at the same detection location and acquisition time sequence are grouped into the same anchor group. Within each anchor group, the peak position of the PFAS candidate fingerprint anchor is compared with the stable spectral segment position of the same PFAS structure name in the PFAS structure spectral segment correspondence. The peak position shift direction is determined based on the relative position of the peak position to the stable spectral segment position, and the peak position shift amount is determined based on the spectral segment position difference between the peak position and the stable spectral segment position. The expression is as follows: in, Indicates peak position offset; Indicates the PFAS structure name In the spectral sensing channel The location of stable spectral segments; Indicates the local spectral response intensity after peak position correction; superscript This represents a local spectral band after peak position correction. The above formula is used to explain the peak position correction process for a local spectral band. With spectral position Within the same spectral coordinate scale, this method is used to perform translation correction on the spectral coordinates of local spectral segments. It calculates the peak position offset between the actual peak position and the stable spectral segment position, and translates the local spectral segments based on this offset to align them with the corresponding PFAS structure. Based on the peak position offset direction and amount, it performs peak position translation on the local spectral segments in the corresponding spectral sensing channel, aligning the peak positions in the local spectral segments with the stable spectral segment positions of the corresponding PFAS structure names. After peak position translation, it preserves the peak shape and response change characteristics of the local spectral segments, ensuring that the peak-corrected local spectral segments still reflect the spectral response changes of agricultural product extracts in the corresponding spectral sensing channel.

[0054] During spectral registration, the PFAS structure name is used as the registration reference. Local spectral segments that have completed peak position correction within the same anchor point group are grouped according to the spectral sensing channel labels. Based on the range of local spectral segments after peak position correction and the range of structural feature spectral segments in the correspondence relationship of PFAS structure spectral segments, the registration range of spectral segments matching the corresponding PFAS structure name is determined, so that local spectral segments pointing to the same PFAS structure name in different spectral sensing channels establish a correspondence relationship. Spectral registration between different spectral sensing channels uses the PFAS structure name and the range of structural feature spectral segments as the correspondence reference, and it is not required that different spectral sensing channels have the same spectral physical coordinates.

[0055] The local spectral segments that have completed peak position correction and spectral segment registration, along with the corresponding spectral sensing channel markers, detection location markers, acquisition time sequence markers, PFAS structure names, peak positions, peak shapes, and response change characteristics, are retained as alignment records. Multiple alignment records under the same detection location and the same acquisition time sequence constitute a PFAS fingerprint alignment unit, and multiple PFAS fingerprint alignment units constitute a PFAS fingerprint alignment spectral atlas. It should also be noted that existing technologies typically complete PFAS detection of agricultural products by splicing multi-source spectra as a whole, uniformly reducing dimensions, or directly classifying them. However, they lack the basis for the correspondence between PFAS structure and each spectral sensing channel, which can easily lead to spectral mismatch and masking of effective features under complex matrix interference. This scheme determines the target spectral range by pre-setting the PFAS structure spectral segment correspondence, establishes a PFAS fingerprint anchor point set, and uses the PFAS fingerprint anchor point set to perform peak position correction and spectral segment registration on local spectral segments to obtain a PFAS fingerprint aligned spectrum set. This enables local spectral segments pointing to the same PFAS structure in different spectral sensing channels to form an accurate correspondence, reduces interference from irrelevant background spectral segments, improves the accuracy of multi-source spectrum alignment and the stability of PFAS feature recognition, and provides a structurally clear spectral basis for subsequent source confidence fusion.

[0056] S3. Extract PFAS target response features based on PFAS fingerprint alignment spectral atlas, and generate spectral source confidence weights based on response stability, background interference suppression degree and fingerprint anchor matching degree. Then, fuse the PFAS target response features according to the spectral source confidence weights to form spectral source confidence fusion features.

[0057] S3.1 Based on the PFAS fingerprint aligned spectral atlas, extract the peak position changes, peak intensity changes, peak shape changes, and spectral band response areas corresponding to the PFAS fingerprint anchor points in each spectral sensing channel to form PFAS target response features.

[0058] It should be noted that the PFAS fingerprint alignment unit in the PFAS fingerprint alignment spectral set is used as the processing object. According to the spectral sensing channel mark, detection position mark, acquisition time sequence mark and PFAS structure name, the local spectral segment for peak position correction and spectral segment registration is determined, and the peak position and peak shape corresponding to the PFAS structure name in the PFAS fingerprint anchor point set are used as the feature extraction reference.

[0059] The location of the spectral segment with the largest continuous response change amplitude in the local spectral band is determined, and the spectral segment location is compared with the corresponding peak position in the PFAS fingerprint anchor point set to obtain the peak position change; the response intensity corresponding to the peak position is determined, and the response intensity corresponding to the peak position is compared with the average level of the response intensity at both ends of the local spectral band to obtain the peak intensity change.

[0060] The width, symmetry, and continuity of the response changes on both sides of the peak are compared with the corresponding peak shape in the PFAS fingerprint anchor set to obtain the peak shape change; the response intensity within the target spectral range on both sides of the peak is continuously accumulated to obtain the spectral response area.

[0061] Peak position changes, peak intensity changes, peak shape changes, and spectral band response areas obtained under the same spectral sensing channel, the same detection location, the same acquisition sequence, and the same PFAS structure name are recorded accordingly to form the PFAS target response characteristics, expressed as follows: in, Indicates the target response characteristics of PFAS; Indicates the normalized peak position variation; Indicates changes in peak intensity; Indicates peak shape variation; Indicates the normalized spectral response area; The spectral width represents the range of the target spectral band. The peak shape record value representing a local spectral segment is formed by normalizing the width, symmetry, and continuity of the response changes on both sides of the peak position; This represents the standard peak-shaped record value corresponding to the PFAS structure name in the PFAS fingerprint anchor set; The first norm (L1) is used to calculate the difference between the current peak shape record value and the standard peak shape record value. When the peak shape record value consists of three normalized components—width, symmetry, and continuity—the L1 represents the sum of the absolute values ​​of the differences between the three components. This formula is used to form the PFAS target response features. Specifically, normalized peak position variation represents peak position shift, peak intensity variation represents the intensity difference at the peak position relative to the background response, peak shape variation represents the difference between the local peak shape and the peak shape of the PFAS fingerprint anchor point, and normalized spectral band response area represents the overall response intensity within the target spectral band. These four features together constitute the PFAS target response features. S3.2. Based on the target response characteristics of PFAS, determine the response stability, background interference suppression degree and fingerprint anchor point matching degree of each spectral sensing channel, and generate spectral source confidence weights based on the response stability, background interference suppression degree and fingerprint anchor point matching degree.

[0062] It should be noted that each spectral sensing channel under the same detection location and the same acquisition sequence is used as the comparison object so that the source confidence weight can reflect the reliability of the response of different spectral sensing channels to the same PFAS structure name.

[0063] When determining response stability, the differences in peak position variation, the consistency of peak intensity variation direction, and the continuity of spectral response area variation for the same PFAS structure name under adjacent acquisition time sequences for the same spectral sensing channel are used as evaluation criteria for response stability. The differences in peak position variation, the consistency of peak intensity variation direction, and the continuity of spectral response area variation are ranked separately, and each ranking is converted into a ranking score. The smaller the difference in peak position variation, the more consistent the peak intensity variation direction, and the more continuous the spectral response area variation, the higher the ranking score. The ranking scores of the same spectral sensing channel in the three evaluation criteria are summed to obtain a comprehensive response stability score, and the response stability is ranked from highest to lowest according to the comprehensive response stability score. When the comprehensive response stability scores are the same, the spectral sensing channel with the smaller difference in peak position variation is prioritized; if they are still the same, the same ranking is retained.

[0064] When determining the degree of background interference suppression, the peak intensity change and spectral band response area in the PFAS target response characteristics are used as the PFAS target response content, and the average level of the response intensity at both ends of the local spectral band and the response fluctuations on both sides of the peak are used as the background response content. The background interference suppression degree of each spectral sensing channel is ranked according to the difference between the PFAS target response content and the background response content, and the response fluctuations on both sides of the peak are ranked, so as to obtain the background interference suppression degree of each spectral sensing channel.

[0065] When determining the fingerprint anchor matching degree, the enhancement or weakening direction reflected by the peak position change, peak shape change and peak intensity change in the PFAS target response features is compared with the peak position, peak shape and response change features corresponding to the PFAS structure name in the PFAS fingerprint anchor set. According to the ranking of peak position difference, peak shape difference and response direction consistency, the fingerprint anchor matching degree of each spectral sensing channel is sorted to obtain the fingerprint anchor matching degree of each spectral sensing channel.

[0066] The ranking of response stability, background interference suppression, and fingerprint anchor matching is converted into ranking scores. In each ranking, a ranking score is assigned to each spectral sensing channel based on its ranking. Spectral sensing channels with higher rankings receive higher ranking scores than those with lower rankings. Channels with the same ranking receive the same ranking score. Ranking scores are assigned in descending order of the number of spectral sensing channels involved in the ranking, with the highest ranking channel receiving the highest score and the lowest ranking channel receiving the lowest score. The ranking scores of the same spectral sensing channel in all three rankings are summed to obtain the comprehensive confidence score for that channel. The ratio of the comprehensive confidence score of each spectral sensing channel to the sum of the comprehensive confidence scores of all spectral sensing channels is calculated to obtain the source confidence weight of that channel, expressed as follows: in, Indicates the source confidence weight; This indicates the overall credibility score; This refers to a set of spectral sensing channels that participate in fusion under the same detection location, the same acquisition sequence, and the same PFAS structure name; This indicates the number of spectral sensing channels participating in the sorting under the same detection location, the same acquisition time sequence, and the same PFAS structure name; Indicates the ranking of response stability; superscript Indicates response stability; Indicates the ranking of background interference suppression levels; superscript Indicates the degree of background interference suppression; Indicates the ranking of fingerprint anchor point matching degree; superscript This indicates the degree of fingerprint anchor matching. The above formula is used to generate the spectral source confidence weight. The comprehensive confidence score is calculated based on the ranking of response stability, background interference suppression, and fingerprint anchor matching, and then converted into the spectral source confidence weight.

[0067] When a spectral sensing channel ranks last in any of the three categories—response stability, background interference suppression, and fingerprint anchor matching—and the corresponding PFAS target response feature does not form a stable correspondence with the PFAS fingerprint anchor set, this spectral sensing channel is identified as a low-confidence spectral sensing channel. Stable correspondence means that peak position changes fall within the structural feature spectral range of the corresponding PFAS structure name and the allowable peak position deviation range; peak shape changes do not exceed the allowable peak shape difference range; and the direction of response change is consistent with the corresponding response change feature in the PFAS fingerprint anchor set. The allowable peak position deviation range and allowable peak shape difference range are determined by the peak position fluctuation and peak shape fluctuation obtained from repeated acquisitions of known PFAS standard samples. The low-confidence spectral sensing channel is downgraded by setting its ranking score in all three categories—response stability, background interference suppression, and fingerprint anchor matching—to the lowest possible ranking score, and then summed again to obtain the downgraded comprehensive confidence score. The ratio between the downgraded comprehensive confidence score and the comprehensive confidence scores of the other spectral sensing channels is recalculated to obtain the updated source confidence weight.

[0068] S3.3. The PFAS target response features of each spectral sensing channel at the same detection location and the same acquisition time sequence are fused according to the spectral source confidence weight to form the spectral source confidence fusion feature.

[0069] It should be noted that PFAS target response features from different spectral sensing channels but with the same detection location marker, the same acquisition timing marker, and the same PFAS structure name are grouped into the same fusion group; the PFAS target response features in each fusion group include peak position variation, peak intensity variation, peak shape variation, and spectral band response area, and the PFAS target response features of each spectral sensing channel correspond to a spectral source confidence weight.

[0070] The peak position changes of each spectral sensing channel within the same fusion group are weighted according to the corresponding source confidence weight to obtain the fused peak position changes; the peak intensity changes of each spectral sensing channel are weighted according to the corresponding source confidence weight to obtain the fused peak intensity changes; the peak shape changes of each spectral sensing channel are weighted according to the corresponding source confidence weight to obtain the fused peak shape changes; and the spectral response area of ​​each spectral sensing channel is weighted according to the corresponding source confidence weight to obtain the fused spectral response area.

[0071] The response change directions of each spectral sensing channel are weighted and voted according to the corresponding spectral source confidence weights to obtain the fused response direction. Among them, the enhancement direction is denoted as positive support, and the reduction direction is denoted as negative support. When the sum of the spectral source confidence weights corresponding to positive support is greater than the sum of the spectral source confidence weights corresponding to negative support, the fused response direction is determined as the enhancement direction; when the sum of the spectral source confidence weights corresponding to negative support is greater than the sum of the spectral source confidence weights corresponding to positive support, the fused response direction is determined as the reduction direction; when the two are the same, the response change direction corresponding to the spectral sensing channel with the highest spectral source confidence weight is determined as the fused response direction.

[0072] The variations in fusion peak position, fusion peak intensity, fusion peak shape, fusion spectral band response area, and fusion response direction within the same fusion group are recorded in correspondence with the corresponding detection location markers, acquisition time sequence markers, and PFAS structure names to form a single-group source confidence fusion feature. The source confidence fusion feature is formed by combining the single-group source confidence fusion features under different detection locations, different acquisition times, and different PFAS structure names, expressed as follows: in, This represents the source confidence fusion feature. It should also be noted that existing technologies typically process multi-source spectral data through fixed-weight fusion, full-spectrum stitching, or direct classification algorithms. However, the response stability and interference levels of different spectral sensing channels under the complex matrix of agricultural products are inconsistent, easily leading to low-confidence sources affecting the final detection results. This scheme extracts PFAS target response features and generates source confidence weights based on response stability, background interference suppression, and fingerprint anchor point matching, performing confidence fusion on the PFAS target response features. This enables spectral sensing channels with stable responses, low background interference, and consistent anchor point matching to play a major role in the fusion, reducing the impact of abnormal channels, noise fluctuations, and matrix background on the detection results, and improving the anti-interference capability, feature expression reliability, and PFAS recognition stability of multi-source spectrum fusion.

[0073] S4. Based on the source confidence fusion characteristics, determine the candidate results of PFAS category, candidate results of concentration range, and suspected PFAS risk markers in agricultural product extracts. Based on the support relationship between each spectral sensing channel and the candidate results of PFAS category, candidate results of concentration range, and suspected PFAS risk markers, determine the source consistency status. When the source consistency status is consistent, generate valid agricultural product PFAS detection information. When the source consistency status is inconsistent, generate agricultural product PFAS detection information to be verified.

[0074] S4.1 Based on the correspondence between the source confidence fusion characteristics and the preset PFAS category spectral fingerprint records and the preset PFAS concentration range response relationship, determine the candidate results of PFAS category and concentration range in the agricultural product extract.

[0075] It should be noted that the preset PFAS category spectral fingerprint records and preset PFAS concentration range response relationships are established before the detection of agricultural product extracts, and are used to provide corresponding basis for the category determination and concentration range determination of spectral source confidence fusion features.

[0076] Category calibration detection was performed using known PFAS standard samples, which included standard substances with known PFAS category names and corresponding PFAS structure names. Repeated detections were performed on the known PFAS standard samples, and the multi-source spectral response maps generated from each detection were sequentially associated, registered with fingerprint anchors, extracted as target response features, and subjected to source confidence fusion processing to form standard source confidence fusion features. Based on the standard source confidence fusion features obtained from repeated detections, the stable standard fusion peak position changes, standard fusion peak shape changes, and standard response directions under the corresponding PFAS category name were determined. When the standard fusion peak position changes and standard fusion peak shape changes corresponding to the same PFAS category name remained consistent in repeated detections, and the standard response directions were consistent, the PFAS category name, PFAS structure name, standard fusion peak position changes, standard fusion peak shape changes, standard response directions, and repeated detection records were saved accordingly to form a PFAS category spectral fingerprint record.

[0077] PFAS standard samples of known concentrations were used for stepwise concentration calibration. Multiple standard concentration levels were set according to the target detection range, and PFAS standard samples at each standard concentration level were repeatedly tested. The multi-source spectral response maps generated from each test were sequentially associated and labeled, fingerprint anchored, target response features extracted, and spectral source confidence fusion processed to form standard spectral source confidence fusion features corresponding to each standard concentration level. The concentration interval boundaries were determined based on the standard fusion peak intensity variation range and the standard fusion spectral band response area range at each standard concentration level. When the standard fusion peak intensity variation range and the standard fusion spectral band response area range corresponding to adjacent standard concentration levels can be distinguished, the concentration range between adjacent standard concentration levels is divided into one concentration interval. When the response ranges corresponding to adjacent standard concentration levels overlap, the corresponding concentration ranges are merged into the same concentration interval. The PFAS category name, upper limit of concentration interval, lower limit of concentration interval, standard fusion peak intensity variation range, standard fusion spectral band response area range, number of calibration repetitions, and allowable response fluctuation range are recorded to form the PFAS concentration interval response relationship. The allowable response fluctuation range is determined by the repeated test results at the same standard concentration level.

[0078] The fusion peak position change, fusion peak shape change, and fusion response direction in the source confidence fusion features are compared with the standard fusion peak position change, standard fusion peak shape change, and standard response direction in the PFAS category spectral fingerprint record. It is determined whether the fusion response direction is consistent with the standard response direction. PFAS category names with inconsistent response directions are not considered as priority category candidates. For PFAS category names with consistent response directions, the differences in fusion peak position change and fusion peak shape change are calculated separately and converted into difference rankings. The smaller the differences in fusion peak position change and fusion peak shape change, the higher the degree of category correspondence. The ranking scores corresponding to the fusion peak position change difference ranking and the fusion peak shape change difference ranking are summed to obtain the category correspondence score, and the category correspondence score is ranked from high to low. When the category correspondence scores are the same, the PFAS category name with the smaller difference in fusion peak position change is prioritized, and the PFAS category name with the highest category correspondence ranking is determined as the PFAS category candidate result in the agricultural product extract.

[0079] Using the source confidence fusion features corresponding to the PFAS category candidate results as the judgment object, the fusion peak intensity change and fusion band response area in the source confidence fusion features are compared with the standard fusion peak intensity change range and standard fusion band response area range under the corresponding PFAS category name in the PFAS concentration interval response relationship. If both the fusion peak intensity change and the fusion band response area fall within the standard response range corresponding to the same concentration interval, this concentration interval is determined as the priority concentration interval. If there are multiple priority concentration intervals, the difference between the fusion peak intensity change and the standard fusion peak intensity change center value of each priority concentration interval, and the difference between the fusion band response area and the standard fusion band response area center value of each priority concentration interval are calculated respectively. The concentration intervals are ranked according to the sum of the two differences from smallest to largest to determine the degree of correspondence. If the fusion peak intensity change and the fusion band response area do not fall within the same concentration interval at the same time, the minimum difference with the boundary of the standard response range of each concentration interval is calculated respectively. The concentration intervals are ranked according to the sum of the minimum differences from smallest to largest to determine the degree of correspondence. The concentration interval with the highest degree of correspondence is determined as the candidate concentration interval for agricultural product extract.

[0080] S4.2. Based on the candidate results of PFAS categories and concentration ranges, and combined with the response content in the source confidence fusion features that corresponds to the PFAS fingerprint anchor but does not completely correspond to the candidate results of PFAS categories, suspected PFAS risk markers are identified.

[0081] It should be noted that the suspected PFAS risk marker is used to characterize the presence of PFAS structure-directed responses in the source confidence fusion features, but the PFAS structure-directed responses cannot be jointly explained by the current PFAS category candidate results and concentration range candidate results. The PFAS structure-directed response refers to the response content that can correspond to the peak position, peak shape and response change features in the PFAS fingerprint anchor set, but has not been classified into the PFAS category candidate results.

[0082] Based on the PFAS category candidate results, the PFAS structure name, standard fusion peak position change, standard fusion peak shape change, and standard response direction corresponding to the PFAS category candidate results are determined from the PFAS category spectral fingerprint records. The fusion peak position change, fusion peak shape change, and fusion response direction in the source confidence fusion features are compared with the PFAS category spectral fingerprint records corresponding to the PFAS category candidate results, and the response content that can correspond to the PFAS category candidate results is determined as the assigned response content.

[0083] For the remaining response content in the source confidence fusion features that has not been identified as the assigned response content, it is further determined whether the remaining response content still corresponds to the peak position, peak shape and response change features in the PFAS fingerprint anchor set; when the remaining response content still corresponds to the PFAS fingerprint anchor set, the remaining response content is identified as the response content to be labeled.

[0084] The corresponding results of the concentration range candidate results are checked against the content of the response to be labeled. When the fusion peak intensity change and fusion spectrum response area in the content of the response to be labeled do not fall within the standard fusion peak intensity change range, standard fusion spectrum response area range and allowable response fluctuation range corresponding to the concentration range candidate results, the PFAS structure name, detection location mark, acquisition time sequence mark and risk source content corresponding to the content of the response to be labeled are recorded as suspected PFAS risk marks. When there is no content of the response to be labeled, or the content of the response to be labeled can fall within the standard fusion peak intensity change range, standard fusion spectrum response area range and allowable response fluctuation range corresponding to the concentration range candidate results, it is determined that no suspected PFAS risk marks are found.

[0085] S4.3. Based on the support relationship between each spectral sensing channel and the candidate results of PFAS category, candidate results of concentration range, and suspected PFAS risk markers, determine the consistency status of the spectral source. When the consistency status of the spectral source is consistent, generate valid PFAS detection information for agricultural products. When the consistency status of the spectral source is inconsistent, generate PFAS detection information for agricultural products to be reviewed.

[0086] It should be noted that the PFAS category candidate results, concentration range candidate results, and suspected PFAS risk markers are considered as content to be confirmed. Before forming the source confidence fusion features, each spectral sensing channel retains the corresponding PFAS target response features. The PFAS target response features are used to determine the support relationship between each spectral sensing channel and the content to be confirmed.

[0087] The enhancement or weakening directions reflected by the peak position change, peak shape change, and peak intensity change in the PFAS target response characteristics of each spectral sensing channel are compared with the PFAS category spectral fingerprint record corresponding to the PFAS category candidate result. When the category name obtained from the comparison is consistent with the PFAS category candidate result, it is determined that the corresponding spectral sensing channel supports the PFAS category candidate result; otherwise, it is determined that the corresponding spectral sensing channel does not support the PFAS category candidate result.

[0088] The peak intensity variation and spectral band response area in the PFAS target response features of each spectral sensing channel are compared with the PFAS concentration interval response relationship corresponding to the PFAS category candidate results. When the concentration interval obtained by the comparison is consistent with the concentration interval candidate results, it is determined that the corresponding spectral sensing channel supports the concentration interval candidate results; otherwise, it is determined that the corresponding spectral sensing channel does not support the concentration interval candidate results.

[0089] When a suspected PFAS risk marker exists, the PFAS target response characteristics under the PFAS structure name, detection location marker, and acquisition time sequence marker corresponding to the suspected PFAS risk marker in each spectral sensing channel are compared with the peak position, peak shape, and response change characteristics in the PFAS fingerprint anchor set. If the comparison result can point to the risk source content in the suspected PFAS risk marker, it is determined that the corresponding spectral sensing channel supports the suspected PFAS risk marker; otherwise, it is determined that the corresponding spectral sensing channel does not support the suspected PFAS risk marker. When no suspected PFAS risk marker exists, no support relationship determination is made for the suspected PFAS risk marker.

[0090] The support relationships of each spectral sensing channel for PFAS category candidate results, concentration range candidate results, and suspected PFAS risk markers are verified accordingly. The spectral source consistency status is determined by combining the spectral source confidence weights corresponding to each spectral sensing channel. Specifically, the sum of spectral source confidence weights corresponding to spectral sensing channels supporting PFAS category candidate results, the sum of spectral source confidence weights corresponding to spectral sensing channels supporting concentration range candidate results, and the sum of spectral source confidence weights corresponding to spectral sensing channels supporting suspected PFAS risk markers when they exist are calculated. The spectral source consistency status is determined to be consistent when the sum of spectral source confidence weights supporting PFAS category candidate results is not lower than the lower limit of category consistency support weight, the sum of spectral source confidence weights supporting concentration range candidate results is not lower than the lower limit of concentration consistency support weight, and the sum of spectral source confidence weights supporting suspected PFAS risk markers when they exist is not lower than the lower limit of risk marker consistency support weight. Otherwise, the spectral source consistency status is determined to be inconsistent.

[0091] The lower limits of support weights for category consistency, concentration consistency, and risk label consistency are determined by the repeated calibration results of known PFAS standard samples. Specifically, during the repeated calibration of known PFAS standard samples, the sum of the source confidence weights corresponding to the spectral sensing channels that support the correct PFAS category name, the sum of the source confidence weights corresponding to the spectral sensing channels that support the correct concentration range, and the sum of the source confidence weights corresponding to the spectral sensing channels that support the correct risk label when a known risk response exists are calculated in each calibration. The weights and values ​​that can obtain correct detection results in multiple repeated calibrations are summarized, and the minimum weight sum, or the lowest stable weight sum after removing abnormal calibration results, is determined as the lower limits of support weights for category consistency, concentration consistency, and risk label consistency, respectively. Low-confidence spectral sensing channels are included in the support weight calculation after being downgraded in order of priority, but are not used as a direct basis for determining inconsistencies in source consistency status.

[0092] When the spectral source consistency status is consistent, the PFAS category candidate results, concentration range candidate results, suspected PFAS risk markers, detection location markers, and acquisition time sequence markers are written into the detection record to generate valid agricultural product PFAS detection information; when the spectral source consistency status is inconsistent, the PFAS category candidate results, concentration range candidate results, suspected PFAS risk markers, detection location markers, acquisition time sequence markers, and inconsistent spectral sensor channel names are written into the detection record to generate agricultural product PFAS detection information to be reviewed.

[0093] In summary, this invention achieves accurate correspondence of the same PFAS structural response in different spectral sources by: establishing a PFAS structural spectral segment correspondence and a PFAS fingerprint anchor set; and using the PFAS fingerprint anchor set to perform peak position correction and spectral segment registration on local spectral segments of each spectral sensing channel, thereby reducing mismatch of heterogeneous spectral segments and matrix background interference, and making PFAS feature recognition more stable; by generating spectral source confidence weights based on response stability, background interference suppression degree, and fingerprint anchor matching degree, and fusing the PFAS target response features of each spectral sensing channel according to the spectral source confidence weights, it achieves differentiation and weighted application of different spectral source confidence levels, reduces the impact of abnormal channels and noise fluctuations on detection results, and improves the reliability of PFAS category identification, concentration range judgment, and suspected risk labeling.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A PFAS detection method for agricultural products based on multi-source spectral fusion, characterized in that, include: Multi-source spectral response maps of pretreated agricultural product extracts were collected in a PFAS spectrochemical sensor array and associated and labeled according to spectral sensing channels, detection locations and acquisition time sequences to generate fingerprint-calibrated multi-source spectra. Based on the correspondence between the fingerprint calibration multi-source spectrum and the preset PFAS structure spectral segment, a PFAS fingerprint anchor point set is established, and the local spectral segments of each spectral sensing channel are registered using the PFAS fingerprint anchor point set to obtain the PFAS fingerprint aligned spectrum set. Based on the PFAS fingerprint alignment spectral atlas, PFAS target response features are extracted, and spectral source confidence weights are generated according to response stability, background interference suppression degree and fingerprint anchor matching degree. The PFAS target response features are fused according to the spectral source confidence weights to form spectral source confidence fusion features. Based on the source confidence fusion characteristics, candidate results for PFAS categories, candidate results for concentration ranges, and suspected PFAS risk markers in agricultural product extracts are determined. The source consistency status is determined based on the support relationship between each spectral sensing channel and the candidate results for PFAS categories, candidate results for concentration ranges, and suspected PFAS risk markers. When the source consistency status is consistent, valid agricultural product PFAS detection information is generated. When the source consistency status is inconsistent, agricultural product PFAS detection information to be verified is generated.

2. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 1, characterized in that, The specific steps for generating the multi-source spectral map for fingerprint calibration are as follows: The pretreated agricultural product extract is fed into the PFAS spectrochemical sensor array, so that the agricultural product extract makes detection contact with each spectral sensing channel, forming a spectral response area to be collected; Based on the spectral response region to be acquired, the spectral response corresponding to each spectral sensing channel is acquired respectively, and background subtraction, baseline correction and intensity unification processing are performed on each spectral response to obtain the channel-corrected spectral response map. Based on the spectral sensing channel, detection position, and acquisition sequence corresponding to the channel calibration spectral response map, an association marker is generated and written into the channel calibration spectral response map. This establishes a correspondence between the responses of each spectral sensing channel at the same detection position and acquisition sequence, generating a fingerprint calibration multi-source spectrum.

3. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 1, characterized in that, The specific steps for obtaining the PFAS fingerprint alignment spectral set are as follows: By fingerprinting the correspondence between the spectral sensing channel markers in the multi-source spectrum and the preset PFAS structure spectral segments, the range of target spectral segments corresponding to the PFAS structure in each spectral sensing channel is determined, and the PFAS target spectral segment index is obtained. Based on the PFAS target spectral band index, local spectral bands corresponding to each spectral sensing channel are extracted from the fingerprint calibration multi-source spectrum, and the peak position, peak shape and response change characteristics in the local spectral bands are matched with the correspondence of PFAS structural spectral bands to establish a PFAS fingerprint anchor point set. Based on the PFAS fingerprint anchor point set, peak position correction and spectral registration are performed on local spectral segments to establish alignment relationships between local spectral segments corresponding to the same PFAS structure, resulting in a PFAS fingerprint aligned spectral set.

4. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 1, characterized in that, The specific steps for forming the spectral source confidence fusion feature are as follows: Based on the PFAS fingerprint aligned spectral atlas, the peak position change, peak intensity change, peak shape change and spectral band response area corresponding to the PFAS fingerprint anchor point in each spectral sensing channel are extracted to form the PFAS target response features. Based on the target response characteristics of PFAS, the response stability, background interference suppression degree and fingerprint anchor point matching degree of each spectral sensing channel are determined respectively, and the source confidence weight is generated according to the response stability, background interference suppression degree and fingerprint anchor point matching degree. The PFAS target response features of each spectral sensing channel at the same detection location and the same acquisition time sequence are fused according to the source confidence weight to form the source confidence fusion feature.

5. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 1, characterized in that, When the spectral source consistency status is consistent, valid PFAS detection information for agricultural products is generated; when the spectral source consistency status is inconsistent, PFAS detection information for agricultural products to be verified is generated. The specific steps are as follows: Based on the correspondence between the source confidence fusion characteristics and the preset PFAS category spectral fingerprint records and the preset PFAS concentration range response relationship, candidate results for PFAS categories and candidate results for concentration ranges in agricultural product extracts are determined. Based on the candidate results of PFAS categories and concentration ranges, and combined with the response content in the source confidence fusion features that can correspond to the PFAS fingerprint anchor but are not classified into the candidate results of PFAS categories, suspected PFAS risk markers are identified. The consistency status of the spectral source is determined based on the support relationship between each spectral sensing channel and the candidate results of PFAS category, candidate results of concentration range, and suspected PFAS risk markers. When the spectral source consistency status is consistent, valid PFAS detection information for agricultural products is generated; when the spectral source consistency status is inconsistent, PFAS detection information for agricultural products to be verified is generated.

6. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 3, characterized in that, The pre-defined PFAS structure spectral segment correspondence is established through the following steps: Known PFAS standard samples were selected as calibration objects, and standard spectral response maps of the known PFAS standard samples were acquired in the PFAS spectrochemical sensor array. The standard spectral response map is subjected to background subtraction, baseline correction and intensity unification to obtain the standard channel corrected spectral response map; When the same PFAS representative substance shows the corresponding spectral segment response in multiple repeated acquisitions, and the corresponding spectral segment response does not form a stable response in the same direction in the blank spectral response map, the current spectral segment position is determined as the stable spectral segment position. The stable spectral band positions are mapped to the corresponding PFAS structure names to determine the range of structural characteristic spectral bands for each PFAS structure name in each spectral sensing channel. Establish corresponding records for PFAS structure names, applicable spectral sensing channels, and structural characteristic spectral ranges to form a PFAS structure spectral range correspondence.

7. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 3, characterized in that, The specific steps for establishing the PFAS fingerprint anchor point set are as follows: According to the target spectral range in the PFAS target spectral index, a local spectral segment is extracted from the channel correction spectral response map corresponding to the fingerprint calibration multi-source spectrum. The position of the spectral segment with the largest continuous response change amplitude in the local spectral segment is determined as the peak position. The width, symmetry and continuity of the response change on both sides of the peak position are determined as the peak shape. The direction of enhancement or weakening of the peak position relative to the background response at both ends of the local spectral segment is determined as the response change characteristic. The peak position, peak shape and response change characteristics of the local spectral segment are matched with the correspondence of the PFAS structural spectral segment, and the local spectral segment that meets the matching relationship is determined as the PFAS candidate fingerprint anchor point. PFAS candidate fingerprint anchors from different spectral sensing channels and pointing to the same PFAS structure name at the same detection location and under the same acquisition time sequence are grouped together to form a PFAS fingerprint anchor set.

8. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 7, characterized in that, The specific steps for peak position correction and spectral registration of local spectral bands are as follows: Based on the PFAS structure name, detection location marker, and acquisition time sequence marker, the PFAS candidate fingerprint anchor points in the PFAS fingerprint anchor point set are grouped to form anchor point groups. Within each anchor group, the peak position of the PFAS candidate fingerprint anchor point is compared with the position of the stable spectral segment with the same PFAS structure name in the PFAS structure spectral segment correspondence to determine the peak position shift direction and peak position shift amount. Based on the peak position shift direction and peak position shift amount, the peak position of the local spectral segment in the corresponding spectral sensing channel is shifted so that the peak position in the local spectral segment is aligned with the position of the stable spectral segment corresponding to the PFAS structure name. Using the PFAS structure name as the registration reference, spectral registration is performed on the local spectral segments that have completed peak position correction to form an alignment record; A PFAS fingerprint alignment unit is formed by multiple alignment records at the same detection location and under the same acquisition time sequence, and a PFAS fingerprint alignment spectrum set is formed by multiple PFAS fingerprint alignment units.

9. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 4, characterized in that, The specific steps for generating the spectral source confidence weights are as follows: Each spectral sensing channel under the same detection location and the same acquisition sequence was used as a comparison object; Based on the differences in peak position variation, peak intensity variation direction and the continuity of spectral band response area variation in the PFAS target response characteristics, the response stability of each spectral sensing channel is ranked. Based on the peak intensity variation, spectral band response area, average level of response intensity at both ends of local spectral bands, and response fluctuations on both sides of the peak position in the PFAS target response characteristics, the background interference suppression degree of each spectral sensing channel is ranked. Based on the enhancement or weakening directions reflected by the peak position change, peak shape change and peak intensity change in the PFAS target response characteristics, the fingerprint anchor point matching degree of each spectral sensing channel is ranked. The ranking of response stability, background interference suppression, and fingerprint anchor matching is converted into a ranking score, and the ranking scores of the same spectral sensing channel in the three rankings are summed to obtain the comprehensive confidence score of the corresponding spectral sensing channel. The source confidence weight of the corresponding spectral sensing channel is obtained by calculating the ratio of the overall confidence score of each spectral sensing channel to the sum of the overall confidence scores of all spectral sensing channels.

10. The PFAS detection method for agricultural products based on multi-source spectral fusion as described in claim 4 or 9, characterized in that, The spectral source confidence fusion feature includes multiple single-group spectral source confidence fusion features, each of which corresponds to a detection location marker, an acquisition time sequence marker, and a PFAS structure name; The single-source confidence fusion features include fusion peak position variation, fusion peak intensity variation, fusion peak shape variation, fusion spectral band response area, and fusion response direction; The fusion peak position change, fusion peak intensity change, fusion peak shape change, and fusion spectral band response area are obtained by weighting the peak position change, peak intensity change, peak shape change, and spectral band response area of ​​each spectral sensing channel within the same fusion group according to the corresponding spectral source confidence weight. The fusion response direction is obtained by weighting the response change direction of each spectral sensing channel within the same fusion group according to the corresponding spectral source confidence weight.

Images