Antibiotic identification method based on three-dimensional fluorescence spectrum Reeb graph

Abstract

The invention discloses an antibiotic identification method based on a three-dimensional fluorescence spectrum Reeb graph. The method comprises the following steps: step 1, constructing an EEM sample data set containing various known antibiotics; step 2, constructing a Reeb graph based on the preprocessed EEM data; 3, quantitatively extracting Reeb graph features from the constructed Reeb graph, including the number of maximum value points, the number of branches, the total length of the branches, the integration of the fluorescence intensity of the branches and the ratio of the intensity to the length of the branches, and forming feature vectors of the antibiotics; step 4, constructing an antibiotic feature library; and step 5, inputting the feature vector into a pre-trained machine learning classifier to perform antibiotic category identification. According to the method, the morphological information of the EEM is effectively captured by utilizing the three-dimensional Reeb graph characteristics of the EEM; and the recognition accuracy of antibiotics in a complex water body environment and the robustness of concentration and peak drift are remarkably improved. The method is suitable for rapid and accurate monitoring of antibiotic pollution in water, and provides technical support for environmental protection.

Description

Technical Field

[0001] This invention relates to the field of water quality testing technology, specifically to a method for antibiotic identification based on three-dimensional fluorescence Reeb maps. Background Technology

[0002] Water pollution is one of the most serious challenges facing the world today, and the rapid and accurate identification of pollutants in water bodies is crucial for environmental protection and public health. In particular, antibiotics, as a widely used class of drugs, inevitably enter the aquatic environment during their production and use, posing a potential threat to ecosystems and human health. Rapid and accurate identification of antibiotics in the aquatic environment is an urgent problem to be solved. Excitation-emission matrix (EEM), as a fingerprint spectroscopy technique, can comprehensively reflect the excitation and emission characteristics of fluorescent substances in water samples, and is widely used in water quality monitoring due to its high sensitivity and rich information content.

[0003] However, most existing EEM-based water pollutant identification systems and methods focus on data-level analysis of two-dimensional EEM matrices, such as extracting peak intensities, areas, or using two-dimensional image processing techniques. This two-dimensional analysis method struggles to effectively distinguish antibiotics with similar fluorescence peak positions or shapes (i.e., overlapping spectra). The main reason is that two-dimensional matrix analysis lacks sensitivity to the three-dimensional morphological features of EEMs (such as peak tilt, concavity, and connectivity), easily losing crucial three-dimensional spatial structural information, thus limiting the accuracy and robustness of identification. Furthermore, factors such as changes in pollutant concentration in the water, fluorescence peak drift, and interference from coexisting substances can all affect the identification performance of traditional methods.

[0004] Therefore, there is an urgent need for a method that can more fully extract the three-dimensional morphological information of EEMs and is insensitive to factors such as concentration and peak drift, so as to achieve faster, more accurate and more robust antibiotic identification, especially to effectively distinguish antibiotics with overlapping fluorescence spectra. Summary of the Invention

[0005] To overcome the limitations of existing two-dimensional EEM analysis methods in distinguishing antibiotics with overlapping fluorescence spectra and resisting concentration changes and peak shifts, this invention provides an antibiotic identification method based on three-dimensional fluorescence spectrum Reeb diagrams. By introducing the novel three-dimensional feature of Reeb diagrams, this invention achieves highly sensitive and robust identification of antibiotics in water.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] An antibiotic identification method based on three-dimensional fluorescence Reeb maps includes the following steps:

[0008] Step 1: Construct an EEM sample dataset containing a variety of known antibiotics;

[0009] Step 2: Construct a Reeb plot based on the preprocessed EEM data;

[0010] Step 3: Quantize and extract Reeb features from the constructed Reeb graph to form the feature vector of the antibiotic;

[0011] Step 4, Construction of antibiotic feature library;

[0012] Step 5: Input the feature vector into a pre-trained machine learning classifier to identify the antibiotic category.

[0013] Furthermore, in step 1, a variety of typical and representative antibiotic pure substances or standard solutions are collected, with particular attention paid to those species that are common in actual aquatic environments or have overlapping fluorescence spectra (e.g., tetracyclines, sulfonamides, β-lactams, etc.). In order to improve the generalization ability and anti-interference ability of the model, simulated river water can be added to the dataset in an appropriate amount.

[0014] Collect a sufficient number of samples to ensure that each antibiotic category contains abundant samples and covers possible concentration variations and environmental factors. Divide the obtained data into a training set (approximately 80%), a validation set (approximately 10%), and a test set (approximately 10%). Perform EEM scanning on each prepared antibiotic sample to ensure the standardization and repeatability of the acquisition process. Record the fluorescence intensity data of each sample in the excitation-emission wavelength range. Associate the collected raw EEM data with the corresponding antibiotic type label and concentration information to form a structured dataset.

[0015] Furthermore, in step 2, the three-dimensional EEM data of the water sample is acquired and preprocessed to obtain a standardized EEM intensity surface f(E). x E m ), where E x E m These are the lengths of the excitation wavelength and the emission wavelength, respectively. Subsequently, the normalized surface is considered as a function defined on the excitation-emission wavelength plane, and the gradient information ▽f(E) of this function is analyzed. x E m The Reeb plot is constructed by considering the changes in isosurface connectivity, thereby abstracting the three-dimensional morphological features of the EEM into a graph structure.

[0016] Furthermore, in step 3, the number N, including the maximum value point, is quantitatively calculated from the constructed Reeb graph. max Number of branches N B The coordinates of the maximum point (E) xiE mi ), representing the excitation wavelength and emission wavelength corresponding to the i-th fluorescence peak, and the maximum point intensity I. F,i , representing the intensity value corresponding to the i-th fluorescence peak, and the total branch length S. B Branch fluorescence intensity integral S F and the ratio of branch strength to length R F The discriminative structural features, including those included, form the Reeb graph feature vector F. topo =[N max E x1 E m1 ,I F,1 ,…,N B ,S B ,S F ,R F ,…];

[0017]

[0018] Among them, L j It is the length of the i-th branch in the Reeb plot. The branch length can be defined as the Euclidean distance of the branch in wavelength space or the length of the intensity change path.

[0019]

[0020] Where, path j It is the path of the j-th branch, I norm (E x E m ) is the normalized fluorescence intensity at each point on the path, and dl is the differential length of the path;

[0021]

[0022] Among them, R F It represents the ratio of the integral of the branch fluorescence intensity to the total branch length, reflecting the "steepness" or "compactness" of the EEM peak.

[0023] In step 4, steps 1 to 3 are repeated for the EEM data of each known antibiotic to construct its Reeb map and extract the corresponding Reeb map features, forming a feature library with corresponding antibiotic type labels. The extracted Reeb map feature vector F topo,k and antibiotic label Y k The antibiotic features are stored in a structured database, forming an antibiotic feature library, which will serve as the basis for subsequent training and validation of machine learning classifiers.

[0024] In step 5, the topological feature vector F of the water sample to be tested is... topoThe antibiotic is compared and matched with features in the antibiotic feature library, and the antibiotic type is identified using a trained machine learning classifier, and the identification result is output.

[0025] This invention innovatively transforms the three-dimensional morphological features of the fluorescence excitation-emission matrix (EEM) into Reeb plots for analysis, thereby extracting unique spectral features that are insensitive to changes in antibiotic concentration and fluorescence peak shifts and can effectively distinguish overlapping spectrum antibiotics.

[0026] The beneficial effects of this invention are mainly reflected in:

[0027] 1. High-precision identification: Reeb images can effectively capture the three-dimensional morphological features of EEMs, and can effectively distinguish even overlapping spectral antibiotics with similar EEM shapes or positions that are difficult to distinguish using traditional methods, thus significantly improving identification accuracy.

[0028] 2. Robustness and anti-interference ability: The Reeb plot has position invariance, making it insensitive to factors such as changes in antibiotic concentration and fluorescence peak drift, thereby enhancing the robustness and practical application value of the identification method.

[0029] 3. Concise and efficient features: Reeb diagrams can compress complex 3D morphological information into a concise 3D structure, effectively removing redundant information and making subsequent feature extraction and classification more efficient. Attached Figure Description

[0030] Figure 1 This is the overall flowchart of the present invention.

[0031] Figure 2 A schematic diagram of the process of building a Reeb diagram.

[0032] Figure 3 This is a schematic diagram of Reeb graph feature extraction. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the specific embodiments of this invention will be described in detail below with reference to the accompanying drawings. These specific embodiments are not intended to limit the invention, but rather to explain it.

[0034] Reference Figures 1-3 An antibiotic identification method based on three-dimensional fluorescence Reeb maps includes the following steps:

[0035] Step 1, Dataset Construction, as follows:

[0036] Collect a variety of typical and representative antibiotic pure substances or standard solutions, paying particular attention to those species that are common in real aquatic environments or have overlapping fluorescence spectra (e.g., tetracyclines, sulfonamides, β-lactams, etc.). To improve the model's generalization ability and anti-interference ability, simulated river water can be added to the dataset as appropriate.

[0037] Collect a sufficient number of samples, ensuring a rich sample count for each antibiotic category and covering potential concentration variations and environmental influences. Divide the resulting data into a training set (approximately 80%), a validation set (approximately 10%), and a test set (approximately 10%). Perform EEM scanning on each prepared antibiotic sample. Ensure the acquisition process is standardized and reproducible, recording fluorescence intensity data within the excitation-emission wavelength range for each sample. Correlate the acquired raw EEM data with the corresponding antibiotic category labels and concentration information to form a structured dataset.

[0038] Step 2: Construct a Reeb plot based on the preprocessed EEM data, that is, the number of preprocessed EEMs f(E) is calculated. x E m The data was converted into a Reeb plot, as follows:

[0039] First, the preprocessed EEM is considered as a two-dimensional plane (E x E m A scalar function defined on ), f(E) x E m ) = I norm (E x E m ), where f(E) x E m ) indicates that at the excitation wavelength E x and emission wavelength E m The normalized fluorescence intensity was then calculated. Next, the gradient information ▽f(E) of this function was analyzed. x E m The key points were identified, including local maxima (corresponding to the peaks of fluorescence), local minima (corresponding to the valleys of fluorescence), and saddle points (the turning points connecting different peaks or valleys). All these key points satisfy the condition that the gradient ▽f(E) is zero. x E m ) = 0.

[0040] By performing streamline tracing (i.e., along the gradient descent or ascent direction) on all non-critical points on the EEM surface, the convergence and separation patterns are observed, and the changes in topological connectivity of contour lines are analyzed. Based on this, a Reeb plot R is constructed. fThe graph is the quotient space M / ~ of the domain M (the excitation-emission plane of the EEM) under a specific equivalence relation ~, where the equivalence relation ~ considers all points p1 and p2 with the same function value f(p1) = f(p2) and located in the same connected component as equivalent (i.e., all points with the same function value and within the connected region are mapped to the same point in the Reeb graph). Ultimately, the Reeb graph is represented by a graph structure, where nodes correspond to key points or intensity values ​​where the topology of contour lines on the EEM intensity surface changes, and edges represent connectivity paths between these key points in terms of intensity values. This construction process effectively abstracts the complex three-dimensional morphological features of the EEM into a concise and topologically invariant graph structure.

[0041] Step 3: Extract discriminative Reeb feature vectors from the constructed Reeb graph. Quantify and extract Reeb graph features from the constructed Reeb graph, including the number of maximum points, the number of branches, the total length of branches, the integral of branch fluorescence intensity, and the ratio of branch intensity to length, to form the feature vector of the antibiotic.

[0042] These features can capture the three-dimensional morphological properties of EEM and are invariant to intensity scaling and small perturbations. The extracted features include, but are not limited to:

[0043] Number of Maximum Points (N) max ): indicates the number of fluorescence peaks in the EEM.

[0044] Coordinates of Maximum Points (Ex) i Em i ): represents the excitation wavelength and emission wavelength corresponding to the i-th fluorescence peak.

[0045] Intensity of Maximum Points (I) F,i ): represents the intensity value corresponding to the i-th fluorescence peak.

[0046] Number of Branches (N) B ): This represents the total number of branches (i.e. edges) in the Reeb graph.

[0047] Total Branch Lengths (S) B ): represents the total length of all edges in the Reeb graph, reflecting the scalability of the EEM shape. Its calculation formula is:

[0048]

[0049] Among them, Lj It is the length of the i-th branch in the Reeb plot. The branch length can be defined as the Euclidean distance or the length of the intensity change path of the branch in the wavelength space.

[0050] Sum of Branch Fluorescence Values ​​(S) F ): This represents the sum of the integral values ​​of the fluorescence intensity corresponding to all branches in the Reeb plot, reflecting the overall fluorescence contribution of the region represented by each branch. Its calculation formula is:

[0051]

[0052] Where, path j It is the path of the j-th branch, I norm (E x E m ) is the normalized fluorescence intensity at each point on the path, and dl is the differential length of the path.

[0053] Ratio of Fluorescence to Branch Length (R) F The value represents the ratio of the integral of the branched fluorescence intensity to the total branch length, reflecting the "steepness" or "compactness" of the EEM peak. Its calculation formula is:

[0054]

[0055] Reeb graph vectors can also include abstract features such as the adjacency matrix of the Reeb graph, the degree sequence of nodes, and the presence or absence of specific cycles. All extracted Reeb graph features are combined into a high-dimensional Reeb graph feature vector F. topo =[N max E x1 E m1 ,I F,1 ,…,N B ,S B ,S F ,R F ,…]

[0056] Step 4, construction of the antibiotic feature library, as follows:

[0057] This step aims to establish a reference database for antibiotic identification and classification. Steps 1 through 3 are repeated for EEM data of each known antibiotic to construct its Reeb map and extract the corresponding Reeb map features, forming a feature library with labels for the corresponding antibiotic type. The extracted Reeb map feature vector F... topo,k and antibiotic label Y kThe features are stored in a structured database, forming an antibiotic feature library. This library will serve as the basis for subsequent training and validation of machine learning classifiers.

[0058] Step 5, antibiotic identification, as follows:

[0059] This module is responsible for comparing the Reebok map features of the water sample to be tested with the features of known antibiotics, and outputting the identification results. It also analyzes the feature vector F obtained in the previous steps. topo,test The data is input into a trained classifier, which will output a prediction result Y of the type of antibiotic in the water sample based on the learned model. pred This module enables accurate identification of antibiotics in water samples.

[0060] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for antibiotic identification based on three-dimensional fluorescence Reebok images, characterized in that, The method includes the following steps: Step 1: Construct an EEM sample dataset containing a variety of known antibiotics; Step 2: Construct a Reeb plot based on the preprocessed EEM data; Step 3: Quantize and extract Reeb features from the constructed Reeb graph to form the feature vector of the antibiotic; Step 4, Construction of antibiotic feature library; Step 5: Input the feature vector into a pre-trained machine learning classifier to identify the antibiotic category.

2. The antibiotic identification method based on three-dimensional fluorescence Reebok image as described in claim 1, characterized in that, In step 1, a sufficient number of samples are collected to ensure that each antibiotic category contains abundant samples and covers possible concentration variations and environmental factors. The obtained data is divided into training set, validation set, and test set. Each prepared antibiotic sample is subjected to EEM scanning to ensure the standardization and repeatability of the collection process. Fluorescence intensity data in the excitation-emission wavelength range of each sample is recorded. The collected raw EEM data is associated with the corresponding antibiotic type label and concentration information to form a structured dataset.

3. The antibiotic identification method based on three-dimensional fluorescence Reebok image as described in claim 1 or 2, characterized in that, In step 2, the three-dimensional EEM data of the water sample is acquired and preprocessed to obtain the standardized EEM intensity surface f(E). x E m ), where E x E m These are the lengths of the excitation wavelength and the emission wavelength, respectively. Subsequently, the normalized surface is considered as a function defined on the excitation-emission wavelength plane, and the gradient information ▽f(E) of this function is analyzed. x E m By analyzing the changes in isosurface connectivity, a Reeb graph is constructed, thereby abstracting the three-dimensional morphological features of the EEM into a graph structure.

4. The antibiotic identification method based on three-dimensional fluorescence Reebok image as described in claim 3, characterized in that, In step 3, the number N, including the maximum value point, is quantitatively calculated from the constructed Reeb graph. max Number of branches N B The coordinates of the maximum point (E) xi E mi ), representing the excitation wavelength and emission wavelength corresponding to the i-th fluorescence peak, and the maximum point intensity I. F,i , representing the intensity value corresponding to the i-th fluorescence peak, and the total branch length S. B Branch fluorescence intensity integral S F and the ratio of branch strength to length R F The discriminative structural features, including those included, form the Reeb graph feature vector F. topo =[N max E x1 E m1 ,I F,1 ,…,N B ,S B ,S F ,R F ,…]; Among them, L j It is the length of the i-th branch in the Reeb plot. The branch length can be defined as the Euclidean distance of the branch in wavelength space or the length of the intensity change path. Where, path j It is the path of the j-th branch, I norm (E x E m ) is the normalized fluorescence intensity at each point on the path, and dl is the differential length of the path; Among them, R F It represents the ratio of the integral of the branch fluorescence intensity to the total branch length, reflecting the "steepness" or "compactness" of the EEM peak.

5. The antibiotic identification method based on three-dimensional fluorescence Reebok image as described in claim 4, characterized in that, In step 4, steps 1 to 3 are repeated for the EEM data of each known antibiotic to construct its Reeb map and extract the corresponding Reeb map features, forming a feature library with labels for the corresponding antibiotic types. The extracted Reeb map feature vector F topo,k and antibiotic label Y k The antibiotic features are stored in a structured database, forming an antibiotic feature library, which will serve as the basis for subsequent training and validation of machine learning classifiers.

6. The antibiotic identification method based on three-dimensional fluorescence Reeb plot as described in claim 5, characterized in that, In step 5, the topological feature vector F of the water sample to be tested is... topo The antibiotic is compared and matched with features in the antibiotic feature library, and the antibiotic type is identified using a trained machine learning classifier, and the identification result is output.

Images