Multistage wavelet transform-based aliasing spectrum functional group data structured storage method
By using a multi-level wavelet transform method to perform baseline correction and separation on asphalt infrared spectral data, and combining dynamic cascade decomposition and non-uniform sampling of the db8 discrete wavelet basis, the problems of aliasing and storage redundancy in asphalt infrared spectral data were solved, achieving high-precision feature extraction and structured storage, and promoting the digital development of road engineering materials.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-07
AI Technical Summary
Existing methods for analyzing asphalt infrared spectral data suffer from severe baseline drift, broad and highly overlapping functional group absorption peaks, resulting in insufficient feature extraction success rate and accuracy. Furthermore, data storage is redundant and lacks structure, hindering in-depth data mining and efficient utilization.
A multi-level wavelet transform method is used to perform baseline correction and filtering on asphalt infrared spectral data, identify and separate characteristic functional group regions, compress data through dynamic cascade decomposition of db8 discrete wavelet basis and non-uniform sampling strategy, and construct metadata knowledge graph for structured storage.
It achieves high-precision decoupling and structured management of asphalt infrared spectral data, improves the success rate and accuracy of feature extraction, reduces storage redundancy, and supports the digital transformation of road engineering materials.
Smart Images
Figure CN121812017A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to a method for structuring storage of mixed spectrum functional group data based on multi-level wavelet transform, and belongs to the technical field of road engineering materials. BACKGROUND
[0002] As the core material of road construction, the accurate analysis of the chemical composition of asphalt binder is crucial for evaluating and improving the performance of the pavement. Infrared spectroscopy is one of the main technical means for analyzing the chemical functional groups of asphalt. However, the existing methods for analyzing asphalt infrared spectrum data have several technical problems that need to be solved urgently:
[0003] Firstly, the chemical composition of asphalt is complex, and its infrared spectrum generally has serious baseline drift. The absorption peaks of different functional groups are wide and highly overlapped, forming serious mixed signals. Traditional baseline correction methods (such as linear fitting and polynomial fitting) and spectral peak decomposition algorithms (such as second derivative method and Gaussian-Lorentz peak fitting) cannot adaptively handle the complex situation of coexistence of wide peaks and sharp peaks, resulting in insufficient success rate and accuracy of feature extraction, which cannot meet the requirements of fine analysis.
[0004] Secondly, the original infrared spectrum data is usually stored in a high-dimensional and unstructured form (such as all data points of the entire spectrum), which lacks a unified and lightweight data standard. This not only causes huge storage redundancy, but also makes the functional group feature information with clear chemical meaning buried in a large amount of redundant data. The functional group metadata and the original spectrum data are isolated from each other, forming a "data island", which seriously hinders the deep mining and efficient utilization of data.
[0005] Therefore, an integrated solution is needed to decouple the functional group data of asphalt infrared spectrum data, compress it with high precision, and manage and store the structured data. SUMMARY
[0006] In view of the problems of mixed spectrum functional group data and data storage redundancy in the existing asphalt infrared spectrum data, the present application provides a method for structuring storage of mixed spectrum functional group data based on multi-level wavelet transform.
[0007] The method for structuring storage of mixed spectrum functional group data based on multi-level wavelet transform of the present application comprises:
[0008] Baseline drift correction and filtering are performed on the original infrared spectrum data of asphalt binder to obtain corrected spectrum data. A plurality of characteristic functional group regions corresponding to the key chemical structure of asphalt are identified and separated in the corrected spectrum data. The intensity values of the plurality of characteristic functional group regions are enhanced to obtain the overall enhanced characteristic functional group region.
[0009] A dynamic cascaded decomposition system based on the db8 discrete wavelet basis is constructed, and adaptive layer decomposition is performed on each enhanced feature functional group region to obtain multi-level wavelet frequency domain data.
[0010] A non-uniform sampling strategy is implemented on the multi-level wavelet frequency domain data, and lossless compression technology based on entropy coding is used to compress the non-uniform sampled data to obtain key feature vectors of multiple feature functional groups.
[0011] A metadata knowledge graph containing multiple feature functional groups is constructed based on key feature vectors of multiple feature functional groups. A relational-document hybrid storage architecture is adopted, and the key feature vectors of multiple feature functional groups are structured and stored through streaming parallel computing.
[0012] According to the present invention, the method for structured storage of aliased spectral functional group data based on multi-level wavelet transform, the method for obtaining the corrected spectral data includes:
[0013] The baseline drift of the raw infrared spectrum data of asphalt binder was corrected by using the asymmetric least squares method, and then the Savitzky-Golay smoothing algorithm was used for filtering to obtain the corrected spectral data.
[0014] According to the structured storage method for aliased spectral functional group data based on multi-level wavelet transform of the present invention, based on prior knowledge of the chemical composition of asphalt binder, 15 characteristic functional group regions corresponding to the key chemical structures of asphalt are selected; the 15 characteristic functional groups include aliphatic CH bonds, aromatic CH bonds, monosubstituted benzene ring CH bonds, trisubstituted benzene ring CH bonds, aromatic ring C=C bonds, trans-butadiene, carbonyl, carbonyl / amide, vinyl, and long-chain aliphatic... Methylene, methyl, aliphatic ethers, sulfoxides and SBS modifiers; using the absorption peak regions of 15 characteristic functional groups as identification criteria, identification and separation are performed according to the spectral positions corresponding to each characteristic functional group.
[0015] According to the structured storage method for aliased spectral functional group data based on multi-level wavelet transform of the present invention, the method for enhancing the intensity values of multiple characteristic functional group regions to obtain the enhanced characteristic functional group regions is as follows:
[0016] The spectral intensity values in the 15 characteristic functional group regions were magnified by 1.2 times, and then a window function of length 5 was used to perform convolution smoothing on the magnified spectral intensity values to obtain a smooth spectral curve.
[0017] Calculate the first gradient of the smoothed spectral curve and mark the points on the smoothed spectral curve whose absolute value of the first gradient exceeds the preset noise threshold as candidate peak points; amplify the intensity value of the candidate peak points on the smoothed spectral curve by 1.5 times to obtain the enhanced spectral curve as the enhanced feature functional group region.
[0018] According to the structured storage method for aliased spectral functional group data based on multi-level wavelet transform of the present invention, the entirety of 15 enhanced feature functional group regions is divided according to wavenumber range to obtain 15 enhanced feature functional group regions.
[0019] The structured storage method for aliased spectral functional group data based on multi-level wavelet transform according to the present invention includes an adaptive layer decomposition method for each enhanced feature functional group region, comprising:
[0020] Based on the peak morphology of the functional group spectrum in each enhanced functional group region, the number of wavelet decomposition layers is adaptively configured, and discrete wavelet transform is performed on the enhanced functional group regions. Multi-level wavelet frequency domain data, including low-frequency approximation coefficients and high-frequency detail coefficients, are obtained through a recursive filter bank algorithm.
[0021] The present invention provides a structured storage method for aliased spectral functional group data based on multi-level wavelet transform, wherein the method for adaptively configuring the number of wavelet decomposition levels includes:
[0022] For wavenumber span greater than 150cm -1 The enhanced feature functional group region was decomposed into 7 wavelet decomposition layers.
[0023] For wavenumber span less than 50cm -1 The enhanced feature functional group region is decomposed into 5 wavelet decomposition layers.
[0024] For the enhanced feature functional group region in the aliasing peak region, the wavelet decomposition layer is selected as 6 layers.
[0025] According to the present invention, the structured storage method for aliased spectral functional group data based on multi-level wavelet transform employs a 7-level wavelet decomposition with a wavenumber span greater than 150 cm. -1 The method for enhancing the feature functional group region is as follows:
[0026] The first layer decomposes the enhanced feature functional group region into low-frequency approximation coefficients. and high frequency detail coefficient The second layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient The third layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient Similarly, the seventh layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient ;
[0027] Finally, the low-frequency approximation coefficients are obtained. and high frequency detail coefficient Multi-level wavelet frequency domain data.
[0028] The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to the present invention, wherein the key eigenvectors of multiple characteristic functional groups are obtained as follows:
[0029] The energy of multi-level wavelet frequency domain data is calculated, and the multi-level wavelet frequency domain data are sorted from largest to smallest energy. Multi-level wavelet frequency domain data with energy greater than the energy contribution rate threshold are selected and compressed into key feature vectors containing 15 feature functional groups.
[0030] According to the present invention, the structured storage method for aliased spectral functional group data based on multi-level wavelet transform is used to deploy functional group location information table and functional group multi-level decomposition data table on the metadata knowledge graph, and to create a data access port.
[0031] Through the data access port, the key feature vector containing 15 feature functional groups is processed concurrently by multiple threads and controlled by transactions, and then stored in a structured manner in the database of metadata knowledge graph.
[0032] The beneficial effects of this invention are as follows: The method of this invention performs baseline correction on the original spectral data using asymmetric least squares and Savitzky-Golay smoothing; it identifies and separates the standard processed data by defining the location information of 15 characteristic functional groups; it deploys a seven-level cascaded decomposition system based on db8 discrete wavelet basis to perform dimensionality-upgrading and decoupling decomposition of the functional group spectral type; it uses a non-uniform sampling strategy to capture subtle features of spectral peaks and global information of the acquired multi-level wavelet frequency domain data signals, realizing multi-scale decomposition of infrared data; it uses composite compression technology to dynamically select and retain key peak detail coefficient features according to energy contribution rate, compressing them into key feature vectors containing 15 characteristic functional groups; it adopts a streaming parallel computing method, integrating a multi-threaded file processor and transaction control mechanism to perform parallel processing and batch import of the original data, realizing structured data management and lightweight storage of post-processed infrared data.
[0033] The method of this invention provides an integrated solution for functional group data decoupling, high-precision compression, and structured data management and storage of asphalt infrared spectral data, which can promote the digital transformation of the road engineering materials field and achieve high-quality development of road engineering materials. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the spectral correction data of asphalt binder in an embodiment of the present invention;
[0035] Figure 2This is a schematic diagram showing the location information of characteristic functional groups in asphalt binders;
[0036] Figure 3 It is a wavelet coefficient heatmap based on multi-level wavelet decoupling;
[0037] Figure 4 This is a schematic diagram of the decoupling-reconstruction accuracy of characteristic functional groups in asphalt binders. Detailed Implementation
[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0039] Specific Implementation Method 1: Combination Figures 1 to 4 As shown, this invention provides a method for structured storage of aliased spectral functional group data based on multi-level wavelet transform, comprising:
[0040] Baseline drift correction and filtering were performed on the raw infrared spectral data of asphalt binder to obtain corrected spectral data. Multiple characteristic functional group regions corresponding to the key chemical structure of asphalt were identified and separated from the corrected spectral data. The intensity values of multiple characteristic functional group regions were enhanced to obtain the overall enhanced characteristic functional group regions.
[0041] A dynamic cascaded decomposition system based on the db8 discrete wavelet basis is constructed, and adaptive layer decomposition is performed on each enhanced feature functional group region to obtain multi-level wavelet frequency domain data.
[0042] A non-uniform sampling strategy is implemented on the multi-level wavelet frequency domain data, and lossless compression technology based on entropy coding is used to compress the non-uniform sampled data to obtain key feature vectors of multiple feature functional groups.
[0043] A metadata knowledge graph containing multiple feature functional groups is constructed based on key feature vectors of multiple feature functional groups. A relational-document hybrid storage architecture is adopted, and the key feature vectors of multiple feature functional groups are structured and stored through streaming parallel computing.
[0044] This implementation provides a method for decoupling, high-precision compression, and structured storage of aliased spectral functional group data based on multi-level wavelet transform. Specifically, it includes: using algorithms such as asymmetric least squares and Savitzky-Golay smoothing to correct the data baseline of the spectral image; constructing a seven-level dynamic cascaded decomposition system based on the db8 discrete wavelet basis; decomposing the standard spectral data into multiple frequency sub-bands using strategies such as discrete wavelet transform and functional group region signal enhancement; implementing a non-uniform sampling strategy based on wavelet energy contribution rate; and using an adaptive parameter configuration mechanism to capture low-frequency details in wide-peak regions using 5-7 level decomposition. For sharp peaks... The region and overlapping peak region are decomposed into 3-5 layers to capture sharp peak details, thus preserving the approximate coefficients in full and achieving dimensionality-up decomposition of overlapping data. The composite compression technology is used to dynamically select and retain key peak detail coefficient features according to energy contribution rate, compressing them into key feature vectors containing 15 feature functional groups, achieving lossless compression of data dimensionality reduction. A standardized metadata knowledge graph containing key functional groups is created, and a streaming parallel computing method is used to perform structured database management and batch storage through variable-length wavelet coefficient vector data. The three-level data storage system of "sample-feature peak-wavelet coefficient" is used to realize the structured storage and batch processing of spectral files.
[0045] Furthermore, the method for obtaining the corrected spectral data includes:
[0046] Asymmetric least squares method was used to correct baseline drift in the raw infrared spectral data of asphalt binder to eliminate background interference. Then, the Savitzky-Golay smoothing algorithm was applied for filtering to suppress high-frequency noise, resulting in corrected spectral data. Finally, the corrected spectral data underwent specific separation based on peak identification to form independent processing units.
[0047] As an example, based on prior knowledge of the chemical composition of asphalt binders, 15 characteristic functional group regions corresponding to the key chemical structures of asphalt were selected; these 15 characteristic functional groups include aliphatic CH bonds, aromatic CH bonds, monosubstituted benzene ring CH bonds, trisubstituted benzene ring CH bonds, aromatic ring C=C bonds, trans-butadiene, carbonyl, carbonyl / amide, vinyl, and long-chain aliphatic groups. Methylene, methyl, aliphatic ethers, sulfoxides and SBS modifiers; using the absorption peak regions of 15 characteristic functional groups as identification criteria, identification and separation are performed according to the spectral positions corresponding to each characteristic functional group.
[0048] The 15 characteristic functional groups are shown in Table 1:
[0049] Table 1
[0050]
[0051] In this embodiment, the method for enhancing the intensity of multiple feature functional group regions to obtain enhanced feature functional group regions is as follows:
[0052] Based on the characteristic peak location information, Hanning window gradient detection is used to identify the peak positions. Spatial signal enhancement technology is employed to enhance the data signal points, facilitating discrete wavelet separation of spectral peak data. For the obtained baseline-corrected spectral data, based on the wavenumber range location information of 15 characteristic functional groups, the following processing is performed on each functional group region to generate enhanced spectral signals:
[0053] The spectral intensity values in 15 characteristic functional group regions were magnified by 1.2 times to initially improve the signal-to-noise ratio; then, a window function of length 5 was used to perform convolution smoothing on the magnified spectral intensity values to obtain a smooth spectral curve.
[0054] The first-order gradient of the smoothed spectral curve is calculated, and points on the smoothed spectral curve whose absolute value of the first-order gradient exceeds a preset noise threshold are marked as candidate peak points. The intensity values of the candidate peak points on the smoothed spectral curve are amplified by a factor of 1.5 to obtain the enhanced spectral curve as the enhanced feature functional group region. The signal intensity of the feature peaks in the enhanced spectral curve is significantly improved relative to the baseline noise, providing high-quality input data for the subsequent multi-scale decomposition of discrete wavelet transform.
[0055] The 15 enhanced characteristic functional group regions were divided according to wavenumber range to obtain the 15 enhanced characteristic functional group regions.
[0056] Furthermore, methods for adaptive layer decomposition of each enhanced feature functional group region include:
[0057] Based on the peak morphology of the functional group spectrum in each enhanced functional group region, the number of wavelet decomposition layers is adaptively configured, and discrete wavelet transform is performed on the enhanced functional group regions. Multi-level wavelet frequency domain data, including low-frequency approximation coefficients and high-frequency detail coefficients, are obtained through a recursive filter bank algorithm.
[0058] As an example, methods for adaptively configuring the number of wavelet decomposition layers include:
[0059] For wavenumber span greater than 150cm -1 The enhanced feature functional group region was decomposed into 7 wavelet decomposition layers.
[0060] For wavenumber span less than 50cm -1 The enhanced feature functional group region is decomposed into 5 wavelet decomposition layers.
[0061] For the enhanced feature functional group region in the aliasing peak region, the wavelet decomposition layer is selected as 6 layers.
[0062] Discrete wavelet transform is used to decouple the spectral peak features of independent functional groups. A multi-level data decoupling mechanism is used to convert the spectral lines of the feature functional groups into multi-level wavelet frequency domain data signals. The high-scale frequency sub-band is used to capture the subtle features of the spectral peaks, while the low-scale frequency sub-band is used to capture the global features.
[0063] As an example, a 7-level wavelet decomposition is used with a wavenumber span greater than 150cm. -1 The method for enhancing the feature functional group region is as follows:
[0064] The first layer decomposes the enhanced feature functional group region into low-frequency approximation coefficients. and high frequency detail coefficient The second layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient The third layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient Similarly, the seventh layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient ;
[0065] Finally, the low-frequency approximation coefficients are obtained. and high frequency detail coefficient Multi-level wavelet frequency domain data.
[0066] As an example, the generated enhanced spectral signal is divided into regions according to the wavenumber range of 15 characteristic functional groups. For each region's spectral data segment, a multi-level discrete wavelet transform based on the Daubechies 8th order wavelet basis (db8) is performed. The specific process is as follows:
[0067] First, based on the peak morphology characteristics of functional groups, the wavelet decomposition level is adaptively configured: for wide peak regions (such as CH aliphatic, ROR ether bonds, etc. with a wavenumber span greater than 150 cm⁻¹), the wavelet decomposition level is adjusted accordingly. -1 (Functional groups), employing a 7-layer decomposition to fully extract low-frequency global contour information; for sharp peak regions (such as C=C trans double bonds, CH monosubstituted benzene rings, etc., with wavenumber spans less than 50 cm⁻¹), a 7-layer decomposition is used. -1 (functional groups), employing a 5-layer decomposition to capture key high-frequency local details; for regions with overlapping peaks (such as... Shear vibration and (Shear vibration superposition area), a 6-layer decomposition is used to extract information while taking into account both global and detailed information.
[0068] Secondly, a discrete wavelet transform is performed on the selected functional group region signal segment. A recursive filter bank algorithm is then used to decompose the signal layer by layer into low-frequency approximate coefficients and high-frequency detail coefficients. Taking a 7-level decomposition as an example, the first level decomposes the original signal into approximate coefficients. (Includes 0-π / 2 frequency band information) and detail coefficients (Includes π / 2-π frequency band information); the second layer continues to... Decompose to obtain and This process continues up to the seventh level, ultimately yielding a set of approximate coefficients. and seven sets of detail coefficients The multi-level wavelet frequency domain data signal is denoted as .
[0069] In this data structure, the low-scale frequency subband specifically refers to the highest-level approximation coefficients. This corresponds to the 0-π / 128 frequency band of the original signal (taking a 7-layer decomposition as an example). This sub-band coefficient characterizes the overall profile features of the spectral curve, including global information such as baseline trend, center position of the broad peak, and amplitude range; the high-scale frequency sub-band specifically refers to the detail coefficients of each layer. ,in The highest frequency component corresponding to the π / 2-π frequency band is used to capture subtle features such as sharp edges of spectral peaks and noise jitter. The relatively low-frequency components corresponding to the π / 64-π / 128 frequency band are used to capture mid-frequency information in the transition region between broad and sharp peaks.
[0070] Through the above multi-level decomposition, different frequency components that were originally overlapping and difficult to distinguish in the wavenumber domain are mapped to non-overlapping frequency sub-bands, achieving data decoupling: for example, the low-frequency main component of the aromatic broad peak is concentrated in... In the middle, the high-frequency components of the characteristic peaks of the SBS modifier superimposed on it were separated to In the middle; for example, and After decomposition, the spectral differences between the overlapping peaks of the shear vibrations are amplified at different detail coefficient levels. This causes the overlapping peaks, which originally required complex mathematical fitting for separation, to appear as differences in the energy distribution of different sub-band coefficients in the wavelet domain, thus achieving the dimensionality-upgrading deconstruction of the aliased functional group signal. Finally, the spectral data of each functional group region is transformed from the original one-dimensional wavenumber-intensity sequence into a multi-dimensional data structure containing 8 (7-level decomposition) or 6 (5-level decomposition) frequency sub-band coefficient arrays. This structure serves as the multi-level wavelet frequency domain data signal used for subsequent energy filtering and feature compression. .
[0071] Furthermore, the method for obtaining the key eigenvectors of multiple feature functional groups is as follows:
[0072] The energy of multi-level wavelet frequency domain data is calculated, and the multi-level wavelet frequency domain data are sorted from largest to smallest energy. Multi-level wavelet frequency domain data with energy greater than the energy contribution rate threshold are selected and compressed into key feature vectors containing 15 feature functional groups.
[0073] In this embodiment, dynamic threshold screening technology is used. Based on the spectral peak morphology, a 7-layer decomposition is configured to retain the main body of the spectral line in the wide peak area, while a 5-layer decomposition is used to capture the details of the sharp peak in the sharp peak area and the overlapping peak area, so as to achieve high-precision decoupling processing of complex overlapping spectral line images.
[0074] Based on the energy distribution of wavelet coefficients, a non-uniform sampling strategy is implemented to retain all approximate coefficients to construct spectral profiles. Key peak detail coefficient features are dynamically selected and retained according to the energy contribution rate, compressing the original high-dimensional spectral data into a key feature vector containing 15 feature functional groups.
[0075] The spectral reconstruction algorithm is used to construct the spectral contour of the image by fully preserving the approximation coefficients, and then the spectral curves of specific functional groups are merged in reverse to form a reconstructed infrared spectral image. The reconstruction accuracy of the algorithm is verified by the data reconstruction error distribution.
[0076] A functional group location information table and a functional group multi-level decomposition data table are set up for the metadata knowledge graph, and a data access port is created.
[0077] Through the data access port, the key feature vector containing 15 feature functional groups is processed concurrently by multiple threads and controlled by transactions, and then stored in a structured manner in the database of metadata knowledge graph.
[0078] This implementation creates a metadata knowledge graph containing 15 functional groups, and sets up multiple target data tables such as the functional group positioning information table and the functional group multi-level decomposition data table based on the infrared parameters obtained by multi-level wavelet decomposition, and creates an analysis data access port.
[0079] Lossless compression technology based on entropy coding is used to compress spectral data, and a data file conversion module is established to perform formatted conversion of spectral analysis data, so as to realize lightweight processing and automated transmission of spectral post-processing files.
[0080] Employing a streaming parallel computing approach, integrating a multi-threaded file processor and transaction control mechanism, the system performs parallel processing and batch import of raw data, achieving automated processing of data recognition, functional group recognition, multi-level wavelet decoupling, and batch import.
[0081] Example: The method of the present invention will be described in detail below through specific examples to verify the beneficial effects of the present invention. The selected infrared spectral data are matrix asphalt from different oil source areas, as shown in Table 2.
[0082] Table 2. Asphalt binders from different oil sources
[0083]
[0084] Step 1: Identification and separation of functional groups in the infrared spectrum of asphalt binder:
[0085] 1) Baseline correction of the original spectral data was performed using the asymmetric least squares (ALS) method and Savitzky-Golay smoothing. Spatial signal enhancement techniques were then used to enhance the data signal points, such as... Figure 1 As shown;
[0086] 2) Select 15 functional group localization information as spectral line feature recognition regions, and use Hanning window gradient detection to identify peak positions, such as... Figure 2 As shown;
[0087] 3) Multi-level discrete wavelet transform is used to decouple the spectral peak features of functional groups in each independent unit. The high-frequency details and low-frequency features of functional groups are identified in a directional manner through the multi-level data decoupling mechanism. The extraction results are shown in Table 3.
[0088] Table 3. Extraction effect of characteristic functional groups of asphalt binder in infrared spectroscopy
[0089]
[0090] Step 2: Decoupling and high-precision reconstruction of infrared spectral characteristic functional groups of asphalt binder:
[0091] 1) Adaptive threshold filtering technology is used to dynamically adjust the wavelet series based on the spectral characteristics of broad peak regions, sharp peak regions, and overlapping peak regions, thereby achieving decoupling processing of complex overlapping spectral lines, such as... Figure 3 As shown;
[0092] 2) Implement a non-uniform sampling strategy based on wavelet coefficient energy distribution, retain the key peak detail coefficient features according to the energy contribution rate, and compress the original high-dimensional spectral data into key feature vectors;
[0093] 3) Develop a spectral decoupling-reconstruction algorithm to inversely merge spectral curves of specific functional groups. Verify the segmentation accuracy of the decoupling module through data reconstruction precision. Figure 4 As shown.
[0094] Step 3: Lightweight processing and structured storage method for spectral data based on entropy coding compression and structured storage:
[0095] 1) Create a spectral data input interface, create a metadata knowledge graph containing 15 functional groups in the database storage, and deploy the required functional group location information table and functional group multi-level decomposition data table.
[0096] 2) Establish a data file conversion and lossless compression module, define a structured format for storing target processing data point sets, and realize automated storage of spectral post-processing files through the data end storage interface;
[0097] 3) Define an interactive front-end interface for spectral data, and use streaming parallel computing to realize batch interactive data upload-processing-entry-download.
[0098] This implementation proposes a multi-scale discrete wavelet spectral data decoupling mechanism. It employs multi-level discrete wavelet transform and dynamic wavelet decomposition threshold screening technology to decouple asphalt infrared spectral correction data at multiple levels, thereby achieving the decoupling of aliased spectral functional group data.
[0099] To address the functional group signal stacking problem in infrared spectral data of asphalt binders, this method employs algorithms such as asymmetric least squares and Savitzky-Golay smoothing to perform baseline correction and data enhancement. Based on functional group feature localization information, 15 characteristic functional group regions in the spectral data are identified, located, and separated. A seven-level dynamic cascaded decomposition system based on Dobessi discrete wavelets is used to preserve the main spectral line in broad peak regions, while a five-level decomposition is used to capture sharp peak details in sharp and overlapping peak regions. Data decomposition of overlapping functional groups is achieved through data dimensionality enhancement. Compared to general de-overlap methods such as continuous wavelet transform, this implementation method uses fingerprint recognition principles to specifically separate and extract characteristic functional groups in asphalt. Multi-level discrete wavelet transform is used to extract global information and detailed features of peak morphology at multiple scales. This mechanism effectively improves the accuracy of capturing weak peak signals, achieving a functional group extraction success rate of 93.3%.
[0100] This implementation proposes a non-uniform sampling strategy based on wavelet coefficient energy distribution and lossless compression technology based on entropy coding to create a streaming parallel computing method with a relational-document hybrid storage structure, thereby achieving high-precision compression and formatted batch storage of spectral data.
[0101] Based on multi-level wavelet data de-overlap, a non-uniform sampling strategy based on wavelet coefficient energy distribution is implemented. Lossless compression technology based on entropy coding is used to compress spectral data into key feature vectors containing 15 functional groups. Furthermore, a metadata knowledge graph containing 15 functional groups is created at the data storage end. A relational-document hybrid storage architecture is adopted, utilizing PostgreSQL's strong transactional management of structured metadata. Streaming parallel computing is used to convert the data point set into JSON encoded file format. Variable-length wavelet coefficient vector data is used for structured database management and batch storage. Compared to traditional lightweight data dimensionality reduction methods, the non-uniform sampling strategy adopted in this application can achieve high-fidelity feature preservation at extremely high compression ratios. Verification based on data reconstruction error distribution shows that this application successfully extracts signals from the processed functional groups while maintaining over 90% signal reconstruction accuracy while removing redundant information.
[0102] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.
Claims
1. A method for structured storage of aliased spectral functional group data based on multi-level wavelet transform, characterized in that... include: Baseline drift correction and filtering were performed on the raw infrared spectral data of asphalt binder to obtain the corrected spectral data. Multiple characteristic functional group regions corresponding to the key chemical structure of asphalt were identified and separated from the corrected spectral data; the intensity values of multiple characteristic functional group regions were enhanced to obtain the overall enhanced characteristic functional group regions. A dynamic cascaded decomposition system based on the db8 discrete wavelet basis is constructed, and adaptive layer decomposition is performed on each enhanced feature functional group region to obtain multi-level wavelet frequency domain data. A non-uniform sampling strategy is implemented on the multi-level wavelet frequency domain data, and lossless compression technology based on entropy coding is used to compress the non-uniform sampled data to obtain key feature vectors of multiple feature functional groups. A metadata knowledge graph containing multiple feature functional groups is constructed based on key feature vectors of multiple feature functional groups. A relational-document hybrid storage architecture is adopted, and the key feature vectors of multiple feature functional groups are structured and stored through streaming parallel computing.
2. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 1, characterized in that, The method for obtaining the corrected spectral data includes: The baseline drift of the raw infrared spectrum data of asphalt binder was corrected by using the asymmetric least squares method, and then the Savitzky-Golay smoothing algorithm was used for filtering to obtain the corrected spectral data.
3. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 2, characterized in that, Based on prior knowledge of the chemical composition of asphalt binders, 15 characteristic functional group regions corresponding to the key chemical structures of asphalt were selected. These 15 characteristic functional groups include aliphatic CH bonds, aromatic CH bonds, monosubstituted benzene ring CH bonds, trisubstituted benzene ring CH bonds, aromatic ring C=C bonds, trans-butadiene, carbonyl, carbonyl / amide, vinyl, and long-chain aliphatic groups. Methylene, methyl, aliphatic ethers, sulfoxides and SBS modifiers; using the absorption peak regions of 15 characteristic functional groups as identification criteria, identification and separation are performed according to the spectral positions corresponding to each characteristic functional group.
4. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 3, characterized in that, The method for enhancing the intensity of multiple feature functional group regions to obtain the enhanced feature functional group regions is as follows: The spectral intensity values in the 15 characteristic functional group regions were magnified by 1.2 times, and then a window function of length 5 was used to perform convolution smoothing on the magnified spectral intensity values to obtain a smooth spectral curve. Calculate the first gradient of the smoothed spectral curve and mark the points on the smoothed spectral curve whose absolute value of the first gradient exceeds the preset noise threshold as candidate peak points; amplify the intensity value of the candidate peak points on the smoothed spectral curve by 1.5 times to obtain the enhanced spectral curve as the enhanced feature functional group region.
5. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 4, characterized in that, The 15 enhanced characteristic functional group regions were divided according to wavenumber range to obtain the 15 enhanced characteristic functional group regions.
6. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 5, characterized in that, Methods for adaptive layer decomposition of each enhanced feature functional group region include: Based on the peak morphology of the functional group spectrum in each enhanced functional group region, the number of wavelet decomposition layers is adaptively configured, and discrete wavelet transform is performed on the enhanced functional group regions. Multi-level wavelet frequency domain data, including low-frequency approximation coefficients and high-frequency detail coefficients, are obtained through a recursive filter bank algorithm.
7. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 6, characterized in that, Methods for adaptively configuring the number of wavelet decomposition layers include: For wavenumber span greater than 150cm -1 The enhanced feature functional group region was decomposed into 7 wavelet decomposition layers. For wavenumber span less than 50cm -1 The enhanced feature functional group region is decomposed into 5 wavelet decomposition layers. For the enhanced feature functional group region in the aliasing peak region, the wavelet decomposition layer is selected as 6 layers.
8. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 7, characterized in that, Seven-level wavelet decomposition with a wavenumber span greater than 150cm was used. -1 The method for enhancing the feature functional group region is as follows: The first layer decomposes the enhanced feature functional group region into low-frequency approximation coefficients. and high frequency detail coefficient ; The second layer will use low-frequency approximation coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient ; The third layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient Similarly, the seventh layer will approximate the low-frequency coefficients. Decomposed into low-frequency approximation coefficients and high frequency detail coefficient ; Finally, the low-frequency approximation coefficients are obtained. and high frequency detail coefficient Multi-level wavelet frequency domain data.
9. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 8, characterized in that, The method for obtaining the key eigenvectors of multiple feature functional groups is as follows: The energy of multi-level wavelet frequency domain data is calculated, and the multi-level wavelet frequency domain data are sorted from largest to smallest energy. Multi-level wavelet frequency domain data with energy greater than the energy contribution rate threshold are selected and compressed into key feature vectors containing 15 feature functional groups.
10. The method for structured storage of aliased spectral functional group data based on multi-level wavelet transform according to claim 9, characterized in that, A functional group location information table and a functional group multi-level decomposition data table are set up for the metadata knowledge graph, and a data access port is created. Through the data access port, the key feature vector containing 15 feature functional groups is processed concurrently by multiple threads and controlled by transactions, and then stored in a structured manner in the database of metadata knowledge graph.