Extraction and qualitative recognition method of density comprehensive texture feature of three-dimensional fluorescence spectral image

Abstract

The present invention belongs to the field of analysis and detection, and discloses an organic matter identification method based on spectral image contour texture feature extraction and nearest neighbor optimization algorithm, specifically a method for extracting comprehensive features of concentration series changes in three-dimensional fluorescence spectral images and quickly distinguishing the types of organic pollutants. In view of the fact that the fluorescence spectral image features of different types of organic pollutants are similar and easily lead to misjudgment and the sample library traversal and comparison is time-consuming, a concentration series comprehensive feature extraction and organic pollutant category image recognition method is provided to quickly and efficiently distinguish aromatic hydrocarbon organic pollutants qualitatively, and can further be combined with the concentration series sample library to achieve semi-quantitative analysis.

Description

Technical Field

[0001] The present invention belongs to the field of analysis and detection, and specifically relates to an organic matter identification method based on spectral image contour texture feature extraction and a nearest neighbor optimization algorithm, and more particularly to a method for extracting comprehensive features of concentration series changes in three-dimensional fluorescence spectral images and quickly distinguishing the types of organic pollutants. Background Art

[0002] As industrial pollution becomes increasingly serious, it is of great significance to quickly and effectively identify and trace the source of pollutants. However, there are many types of pollutants. How to accurately and efficiently locate and identify the types of pollutants from the sample library has become an urgent problem that needs to be solved.

[0003] Three-dimensional fluorescence spectroscopy has one more dimension of information than two-dimensional fluorescence spectroscopy, providing a more effective identification method for the detection of complex systems. It is common for the characteristic areas of fluorescence spectral images of different organic pollutants to overlap and for the characteristic areas to change with concentration. Although organic substances with similar fluorescence spectral characteristics have clear distinguishing features at similar concentrations, the characteristics of different types of organic substances between high-concentration samples and low-concentration samples are similar, which can easily lead to misjudgment. In addition, the retrieval of unknown samples and samples of various concentrations is time-consuming. Therefore, in practical applications, when the amount of sample library data is large, it is particularly important to effectively improve the retrieval efficiency and comparison accuracy.

[0004] Based on the traversal comparison method of the sample library in our invention patent "Three-dimensional fluorescence spectroscopy organic pollutant identification method based on image edge detection feature extraction" (application number 202210574320.2, hereinafter referred to as Patent 1), the present invention further proposes a comprehensive texture feature extraction method taking into account the computational efficiency and accuracy, that is, the texture features of the concentration series samples after the contour edge feature extraction and preservation are superimposed and other related processing is performed, and the comprehensive texture images of various organic pollutants are drawn as comparison samples, and then the nearest neighbor optimization algorithm is used for similarity comparison to achieve accurate determination of the category of the pollutant to be tested. Summary of the invention

[0005] The purpose of the present invention is to provide a method for extracting comprehensive features of concentration series and identifying organic pollutant categories in order to address the situation where the fluorescence spectrum image features of different types of organic pollutants are similar and easily lead to misjudgment and the sample library traversal and comparison is time-consuming, so as to quickly and efficiently qualitatively distinguish aromatic hydrocarbon organic pollutants, and further combine the concentration series sample library to achieve semi-quantitative analysis. The present invention is like patent 1: Select a suitable MOF sensing material, mix it with the object to be detected for three-dimensional fluorescence detection, and extract the texture features of the three-dimensional fluorescence spectrum image after the fluorescence spectrum is imaged; then, the extracted features of the concentration series of various organic pollutants are linearly combined to form a comprehensive texture image of this type of organic matter, and a sample feature library is established to achieve rapid classification and identification of pollutants.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] A method for extracting comprehensive texture features and quickly identifying the concentration series of three-dimensional fluorescence spectrum images of organic pollutants includes the following steps:

[0008] (1) Obtaining a series of sample solutions of monomeric organic pollutants at different concentrations, uniformly mixing the fluorescent probe MOF-74 with each concentration sample, and collecting three-dimensional fluorescence spectrum data of each mixed solution; using the Canny operator to perform edge detection processing, extracting image texture features, and establishing a sample library of characteristic images of each organic matter concentration series;

[0009] (2) linearly superimposing the concentration series sample characteristic images of each monomer organic pollutant in the sample library obtained in step (1) to obtain a preliminary comprehensive texture characteristic image, and counting non-zero value pixel points;

[0010] The specific operation is: linearly superimpose the texture feature images of the same type of organic pollutants at different concentrations into one image, replace the zero-value pixels in the superimposed image with "NaN", and then count the number of non-zero-value pixels and draw the corresponding statistical graph to assist in the subsequent threshold selection.

[0011] (3) analyzing the preliminary comprehensive texture feature image and the statistical results of the number of non-zero value pixels obtained in step (2), using 1 as a threshold, filtering out pixels less than or equal to 1 in the image, and obtaining comprehensive texture feature images of each category of monomeric organic pollutants;

[0012] The specific operation is as follows: restore the pixel value of "NaN" to 0 to facilitate the subsequent calculation of relevant judgment indicators in the nearest neighbor algorithm, and avoid the error situation that "NaN" is still "NaN" after adding the specific value; replace the non-zero pixel values ​​less than or equal to the threshold 1 with 0, and replace the pixel values ​​greater than the threshold 1 with 1, and finally obtain the comprehensive texture feature image of each category of monomer organic pollutants.

[0013] (4) Using the optimized nearest neighbor algorithm to identify unknown pollutants, including the following sub-steps:

[0014] (4.1) reading the comprehensive texture feature images of each category of monomeric organic pollutants obtained in step (3) into the working area of ​​the data processing software;

[0015] (4.2) processing the three-dimensional fluorescence spectrum data of the unknown pollutant solution according to step (1) to obtain a three-dimensional fluorescence spectrum image and a corresponding texture feature image;

[0016] (4.3) First, the texture feature image of the unknown pollutant obtained in step (4.2) is compared with the comprehensive texture feature image samples of each category of monomer organic pollutants in the sample library, and the optimized nearest neighbor algorithm is used to calculate the overall texture similarity or multi-region weighted similarity to determine the pollutant category.

[0017] Specifically, the overall texture similarity is calculated first. There should be a category with significantly higher similarity than other categories. If the difference between the similarity and other categories is less than 20%, it is necessary to further calculate the multi-region weighted similarity to improve the discrimination accuracy. The calculation process of the overall texture similarity is as follows: the number of edge points of the texture feature image of the unknown pollutant is counted, that is, the number of non-zero value pixels in the global binary image, the texture feature image of the unknown pollutant is superimposed with the comprehensive texture image of various organic pollutants, and the number of pixels with a value of 2 is calculated, and the two are divided to obtain the overall texture similarity.

[0018] The calculation process of multi-region weighted similarity is as follows: according to the characteristics of the three-dimensional fluorescence spectrum image, the texture feature image and the comprehensive texture image of the unknown pollutant are divided into multiple sub-images, and the overall texture similarity of the corresponding sub-images is calculated respectively. Different weights are set according to the importance of the multiple sub-images, and the multi-region weighted similarity is obtained by adding the weights. One of the similarity calculations is selected according to the needs, and the category of the unknown pollutant solution is obtained after comparison and judgment.

[0019] Furthermore, the comprehensive texture feature image extraction in step (2) to step (3) is performed by linearly combining, pixel counting and screening the texture feature images of the same organic pollutant at different concentrations to obtain a comprehensive texture feature image that can represent the three-dimensional fluorescence spectral characteristics of this type of organic pollutant.

[0020] Furthermore, the threshold selected in step (3) can be adjusted according to different recognition algorithms. Here, selecting 1 is to remove sample points with no commonality.

[0021] Furthermore, the overall texture similarity (Ots:Overall texturesimilarity) in step (4.3) is calculated by the following formula:

[0022]

[0023] Among them, eq is a function that determines the equality of two parameters within a function, and returns a logical value of 1 when the two values ​​are equal;

[0024] S(m,n) represents the value of the texture feature image of the unknown organic pollutant solution obtained in step (4.2) at the pixel coordinate (m,n);

[0025] Y(m,n) represents the value of the comprehensive texture feature image of each category of monomer organic pollutant sample obtained in step (3) at the pixel coordinate (m,n);

[0026] Furthermore, in step (4.3), the multi-region weight similarity (Mws: Multi-region weightsimilarity) is calculated according to the following formula:

[0027]

[0028] Among them, eq is a function that determines the equality of two parameters within a function, and returns a logical value of 1 when the two values ​​are equal;

[0029] x i Represents the weight of the i-th sub-image after the texture feature image of the unknown pollutant solution is segmented in step (4.3).

[0030] S i (j, k) represents the value of the pixel coordinate (j, k) of the i-th sub-image after image segmentation of the texture feature image of the unknown pollutant solution in step (4.3).

[0031] Y i (j, k) represents the value of the pixel coordinate (j, k) of the i-th sub-image after image segmentation of the comprehensive texture feature image of the sample pollutant in step (4.3).

[0032] The monomer organic pollutant in the present invention is a monocyclic aromatic hydrocarbon.

[0033] The beneficial effects of the present invention are:

[0034] Based on Patent 1, the present invention further extracts the comprehensive texture features of the concentration series of the same type of samples, draws the comprehensive texture images of various organic pollutants as comparison samples, and uses the nearest neighbor optimization algorithm to compare the similarity to achieve accurate determination of the type of pollutants to be tested. The method of the present invention can play a unique role in the comparison of complex spectra and large samples in the field of water quality monitoring and pollutant identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flowchart for comprehensive texture feature extraction and rapid qualitative identification of concentration series of three-dimensional fluorescence spectral images of organic pollutants;

[0036] Figure 2 Preliminary comprehensive texture feature images of four monomer organic pollutants at ten concentrations; a-aniline, b-chlorobenzene, c-nitrobenzene, d-phenol;

[0037] Figure 3 The superimposed images and pixel numerical statistics of aniline organic pollutants; a-aniline comprehensive texture image, b-statistical results;

[0038] Figure 4 The image effect diagram of the comprehensive texture characteristics of four monomer organic pollutants: a-aniline, b-chlorobenzene, c-nitrobenzene, d-phenol;

[0039] Figure 5 The calculation results of the overall texture similarity between ten concentrations of aniline and four comprehensive texture feature images;

[0040] Figure 6 The calculation results of the overall texture similarity between ten concentrations of chlorobenzene and four comprehensive texture feature images;

[0041] Figure 7 The calculation results of the overall texture similarity between ten concentrations of nitrobenzene and four comprehensive texture feature images;

[0042] Figure 8 The calculation results of the overall texture similarity between ten concentrations of phenol and four comprehensive texture feature images;

[0043] Fig. 9 The effect diagram of image segmentation using chlorobenzene as an example; a-complete image, b-first part, c-second part, d-third part;

[0044] Fig.10 Taking chlorobenzene as an example, the statistical results of overall texture similarity and multi-region weighted similarity are calculated;

[0045] Fig.11 Taking phenol as an example, the statistical results of the overall texture similarity and multi-region weighted similarity are calculated. Specific implementation plan

[0046] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0047] The basic steps for extracting comprehensive texture features and rapid qualitative identification of organic pollutants from three-dimensional fluorescence spectral image concentration series are as follows: Figure 1 As shown, after obtaining three-dimensional fluorescence spectrum data and other steps as in Patent 1, contour edge features are extracted, and texture feature images of various organic pollutants of different concentrations in the sample library are extracted as comprehensive texture feature images of various organic pollutants using linear combination and threshold analysis screening methods, and a comprehensive texture feature image sample library is established; the overall texture similarity or multi-region weighted similarity between the texture feature image of the unknown pollutant solution and the comprehensive texture feature images of various organic pollutants is calculated, and the output classification result specifically includes the following steps:

[0048] S1: Obtain a series of sample solutions of monomer organic pollutant concentrations, uniformly mix a certain amount of MOF-74 with them, and collect three-dimensional fluorescence spectrum data of the mixed solutions at different concentrations of four organic pollutants (aniline, chlorobenzene, nitrobenzene, and phenol); use the Canny operator to perform edge detection processing and establish a three-dimensional fluorescence spectrum image texture feature sample library, in which each organic pollutant has multiple three-dimensional fluorescence spectrum image texture feature images at different concentrations, and the number of concentrations of each organic pollutant remains consistent.

[0049] S2: The texture features of the three-dimensional fluorescence spectrum images of various monomeric organic pollutants in the sample library obtained in step S1 are linearly superimposed and pixel-point screened to obtain preliminary comprehensive texture feature images of various organic pollutants. Specifically, the texture feature images corresponding to different concentrations of the same organic pollutant are linearly superimposed into one image, and then the zero-valued pixels are replaced with "NaN" to facilitate subsequent numerical statistics, that is, the number of non-zero-valued pixels is counted and the corresponding bar chart is drawn to facilitate comparison and analysis. Because the focus is on the number distribution of non-zero-valued pixels, the zero-valued pixels are marked as "NaN" which means countless values, and the zero-valued pixels will not be included in the scope of statistical analysis when performing numerical statistics.

[0050] S3: Analyze the preliminary comprehensive texture feature image and numerical statistical results obtained in step S2, determine the screening threshold, and screen out pixels in the image that are smaller than the threshold according to the threshold. Specifically, restore the pixel value of "NaN" to 0 to facilitate the subsequent calculation of relevant judgment indicators in the nearest neighbor algorithm, avoid the error situation that "NaN" is still "NaN" after adding the specific numerical value, replace the non-zero pixel values ​​less than the threshold "1" with 0, and replace the pixel values ​​greater than the threshold "1" with 1, and finally obtain a comprehensive texture feature image representing various organic pollutants.

[0051] S4: Identify unknown pollutants using the optimized nearest neighbor algorithm, including the following sub-steps:

[0052] S4.1: Reading the comprehensive texture image of the monomer organic pollutant and its pollutant category obtained in step S3 into the working area of ​​the data processing software;

[0053] S4.2: Processing the three-dimensional fluorescence spectrum data of the unknown pollutant solution according to step S1 to obtain a three-dimensional fluorescence spectrum image and a corresponding texture feature image;

[0054] S4.3: First, the texture feature image of the unknown pollutant obtained in step S4.2 is compared with the comprehensive texture feature images corresponding to various organic pollutants in the sample library, and the optimized nearest neighbor algorithm is used to calculate the overall texture similarity or multi-region weighted similarity to determine the pollutant category.

[0055] The calculation process of the overall texture similarity is as follows: count the number of edge points of the texture feature image of the unknown pollutant, that is, the number of non-zero value pixels in the global binary image, superimpose the texture feature image of the unknown pollutant with the comprehensive texture image of various organic pollutants and calculate the number of pixels with a value of 2, and divide the two to obtain the overall texture similarity.

[0056] The specific calculation process of multi-region weighted similarity is as follows: according to the spectral image characteristics reflected by the spectrum detection of MOF sensing materials, the texture feature image and the comprehensive texture image of the unknown pollutant are divided into three sub-images, and the overall texture similarity of the corresponding sub-images is calculated respectively. Different weights are set according to the importance of the three sub-images, and the multi-region weighted similarity is obtained by adding the weights. Select one of the similarity calculations as needed, and obtain the category of the unknown pollutant solution after comparison and judgment.

[0057] The calculation process of overall texture similarity (Ots:Overall texture similarity) and multi-region weight similarity (Mws:Multi-region weight similarity) is as follows:

[0058] Overall texture similarity (Ots):

[0059]

[0060] Multi-region weight similarity (Mws):

[0061]

[0062] Here, eq is a function that determines the equality of two parameters within a function and returns a logical value of 1 when the two values ​​are equal.

[0063] S(m,n) represents the value of the texture feature image of the unknown organic pollutant solution obtained in step S4.2 at the pixel coordinate (m,n).

[0064] Y(m,n) represents the value of the comprehensive texture feature image of the monomer organic pollutant sample obtained in step S3 at the pixel coordinate (m,n).

[0065] x i Represents the weight of the i-th sub-image after the texture feature image of the unknown pollutant solution is segmented in step S4.3.

[0066] S i (j, k) represents the value of the i-th sub-image at the coordinate (j, k) after the texture feature image of the unknown pollutant solution is segmented in step S4.3.

[0067] Y i (j, k) represents the value of the i-th sub-image at the coordinate (j, k) after image segmentation of the comprehensive texture feature image of the sample pollutant in step S4.3.

[0068] Example 1

[0069] This embodiment includes the following steps:

[0070] S1. Collect three-dimensional fluorescence spectrum data of samples as in Patent 1:

[0071] Obtain a monomer organic pollutant sample solution, mix a certain amount of MOF-74 with it, and collect three-dimensional fluorescence spectrum data of the mixed solution; the concentration of MOF-74 in the mixed solution is 0.03 mg / ml; low concentration detection range: 10 -6 ~10 -3 mol / L, the sample data used in this study included 1×10 -6 mol / L、2×10 -6 mol / L、5×10 -6 mol / L、8×10 -6 mol / L、1×10 -5 mol / L、2×10 -5 mol / L、3×10 -5 mol / L、4×10 -5 mol / L、5×10 -5 mol / L、1×10 -4 mol / L; the concentrations of chlorobenzene, nitrobenzene and phenol were the same, all 2×10 -6 mol / L、5×10 -6 mol / L、8×10 -6 mol / L、1×10 -5mol / L、2×10 -5 mol / L、4×10 -5 mol / L、5×10 -5 mol / L、1×10 -4 mol / L、1.5×10 -4 mol / L、2.5×10 -4 mol / L.

[0072] S2. Three-dimensional fluorescence spectral data collected after preprocessing step S1 of Patent 1.

[0073] S2.1: As in Patent 1, the encoding format of the three-dimensional fluorescence spectrum data output by the optical detection device is converted from ANSI to utf16 and redundant data is deleted;

[0074] S2.2: As in Patent 1, the cleaned three-dimensional fluorescence spectrum data is imported into the Matlab workspace and saved in the mat file format, and its type is converted from table to double for subsequent operations;

[0075] S3. Extract the texture features of the three-dimensional fluorescence spectrum data pre-processed in step S2 as in patent 1:

[0076] S3.1: As in Patent 1, the three-dimensional fluorescence spectrum data preprocessed in step S2 is plotted as a smooth three-dimensional fluorescence spectrum image, and the contour lines of the three-dimensional fluorescence spectrum image are plotted and saved as an image file.

[0077] S3.2: As in Patent 1, the image file obtained in step S3.1 is read and edge detection is performed using the Canny operator to obtain the texture features of the three-dimensional fluorescence spectrum image, specifically a global binary image, and a three-dimensional fluorescence spectrum image texture feature sample library is established, in which each organic pollutant has multiple three-dimensional fluorescence spectrum image texture feature images at different concentrations, and the number of concentrations of each organic pollutant remains consistent. Through the above operation, a sample library with four organic pollutants, namely aniline, chlorobenzene, nitrobenzene and phenol, and corresponding texture feature images of ten concentrations is established.

[0078] S3.3: The texture features of the three-dimensional fluorescence spectrum images of various monomeric organic pollutants in the sample library obtained in step S3.2 are linearly superimposed and pixel-point screened to obtain preliminary comprehensive texture feature images of various organic pollutants. Specifically, the texture feature images corresponding to different concentrations of the same organic pollutant are linearly superimposed into one image, and then the zero-valued pixels are replaced with "NaN" for the convenience of subsequent numerical statistics, that is, the number of non-zero-valued pixels is counted and the corresponding bar chart is drawn for comparison and analysis. Because the focus is on the number distribution of non-zero-valued pixels, the zero-valued pixels are marked as "NaN" which means countless values, and the zero-valued pixels will not be included in the scope of statistical analysis when performing numerical statistics. Taking aniline as an example, the numerical statistical results are as follows: Figure 3 shown.

[0079] like Figure 2 The following are preliminary comprehensive texture feature images of four organic pollutants: (a) is aniline, (b) is chlorobenzene, (c) is nitrobenzene, and (d) is phenol. Compared with a single texture feature image, the lines are thicker and the pixels in the feature area are mixed and difficult to distinguish.

[0080] Figure 3 Figure 5 is a preliminary comprehensive texture feature image of aniline and its numerical statistical results, where (a) is the preliminary comprehensive texture feature image, and (b) is the result of numerical statistics of the pixels of the image, where the pixel values ​​range from 1 to 10. From Figure (b), we can see that the number of pixels with a value of 1 accounts for the majority, which means that the three-dimensional fluorescence spectrum images of different concentrations of the same organic pollutant have their own changes, and as the value gradually increases, the number of corresponding pixels is also decreasing. The larger the value of the pixel, the more three-dimensional fluorescence spectrum images of different concentrations of the same organic pollutant are present at this point, and the greater the reliability of this point as a common feature of the organic pollutant.

[0081] S3.4: Analyze the preliminary comprehensive texture feature image and numerical statistical results obtained in step S3.3, determine the screening threshold, and screen out pixels in the image that are smaller than the threshold according to the threshold. Specifically, restore the pixel value of "NaN" to 0 to facilitate the subsequent calculation of relevant judgment indicators in the nearest neighbor algorithm, avoid the error situation that "NaN" is still "NaN" after adding the specific numerical value, replace the non-zero pixel values ​​less than the threshold with 0, and replace the pixel values ​​greater than the threshold with 1, and finally obtain a comprehensive texture feature image representing various organic pollutants. The comprehensive texture feature images of four organic pollutants in the embodiment of the present invention are shown in FIG. Figure 4As shown, by analyzing the numerical statistical results of step S3.3, the threshold is determined to be 1, so the pixels with a value of 1 are set to pixels with a value of 0, where Figure (a) is the comprehensive texture feature image of aniline; Figure (b) is the comprehensive texture feature image of chlorobenzene; Figure (c) is the comprehensive texture feature image of nitrobenzene; and Figure (d) is the comprehensive texture feature image of phenol. After pixel screening, it can be seen that the image at this time has thinner lines and clearer feature areas compared to the preliminary comprehensive texture feature image.

[0082] S4: Identify unknown pollutants using the optimized nearest neighbor algorithm, including the following sub-steps:

[0083] S4.1: Read the monomer organic pollutant comprehensive texture image and its pollutant category obtained in step S3.4 into the work area;

[0084] S4.2: Processing the three-dimensional fluorescence spectrum data of the unknown pollutant solution according to steps (2) to (3) to obtain a three-dimensional fluorescence spectrum image and a corresponding texture feature image;

[0085] S4.3: Compare the texture feature image of the unknown pollutant obtained in step S4.2 with the comprehensive texture feature images corresponding to various organic pollutants in the sample library, and use the optimized nearest neighbor algorithm to calculate the overall texture similarity or multi-region weighted similarity to distinguish the pollutant category. The calculation process of the overall texture similarity is as follows: count the number of edge points of the texture feature image of the unknown pollutant, that is, the number of non-zero value pixels in the global binary image, respectively superimpose the texture feature image of the unknown pollutant with the comprehensive texture image of various organic pollutants and calculate the number of pixels with a value of 2, and divide the two to obtain the overall texture similarity. The calculation process of the multi-region weighted similarity is as follows: according to the spectral image characteristics reflected by the spectral detection of the MOF sensing material, the texture feature image of the unknown pollutant and the comprehensive texture image are divided into three sub-images, and the overall texture similarity of the corresponding sub-images is calculated respectively. Different weights are set according to the importance of the three sub-images, and the multi-region weighted similarity is obtained by adding the weights. Select one of the similarity calculations as needed, and obtain the approximate category of the organic pollutants in the unknown pollutant solution after comparison and discrimination.

[0086] S4.4: Based on the approximate category obtained in step S4.3, the sample library of comprehensive texture feature images of each concentration obtained in step S3.4 is quickly located and retrieved to obtain the exact category and semi-quantitative identification.

[0087] In the embodiment of the present invention, the classification accuracy is 100%.

[0088] like Figure 5In order to test the overall texture similarity results of aniline with the comprehensive texture feature images of four organic pollutants at ten concentrations, it can be seen from the figure that the overall texture similarity of aniline at each concentration and the comprehensive texture feature image of aniline is maintained at about 70%, and the highest is close to 90%, while the overall texture similarity with the other three comprehensive texture feature images is basically maintained at about 20%, forming a large gap. This shows that the optimized nearest neighbor algorithm can use overall texture similarity as the judgment indicator to achieve accurate discrimination and classification of ten concentrations of aniline in the sample library.

[0089] Figure 6 The overall texture similarity results of chlorobenzene at ten concentrations and the comprehensive texture feature images of four organic pollutants were tested. The overall texture similarity of chlorobenzene at each concentration in the figure with the comprehensive texture feature image of this type is much different from that of aniline comprehensive texture feature image, and is relatively small with the comprehensive texture feature images of nitrobenzene and phenol, but it can still ensure the accurate classification of chlorobenzenes.

[0090] Figure 7 The overall texture similarity results of nitrobenzene with the comprehensive texture feature images of four organic pollutants were tested for ten concentrations of nitrobenzene, which can also ensure the accurate classification of nitrobenzene.

[0091] Figure 8 The overall texture similarity results of phenol at ten concentrations and the comprehensive texture feature images of four organic pollutants were tested, which can also ensure the accurate classification of phenols.

[0092] like Fig. 9 The concentration is shown as 2×10 -5 The effect diagram of image segmentation is shown in Figure 1, taking chlorobenzene (0.1 mol / L) as an example. A texture feature image is divided into three parts and the weight of the first part is set to 1, the weight of the second part is set to 0.1, and the weight of the third part is set to 1.

[0093] like Fig.10 The results are shown for a concentration of 2×10 -5 mol / L chlorobenzene is taken as an example to compare the calculation results of overall texture similarity and multi-region weighted similarity. Due to the setting of weights of three sub-images, the multi-region weighted similarity is generally lower than the overall texture similarity, but accurate classification can still be guaranteed. The multi-region weighted similarity of the comprehensive texture feature image of chlorobenzene is significantly higher than the multi-region weighted similarity with other comprehensive texture feature images.

[0094] like Fig.11 The concentration is shown as 2×10 -5mol / L phenol was taken as an example to compare the calculation results of overall texture similarity and multi-region weighted similarity. The multi-region weighted similarity of the comprehensive texture feature image of phenol was significantly higher than that of the multi-region weighted similarities with other comprehensive texture feature images.

[0095] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, etc. made within the spirit and principles of the invention should be included in the protection scope of the invention.

Claims

1. A method for extracting and qualitatively identifying the comprehensive texture features of three-dimensional fluorescence spectral image concentration, characterized in that: The following steps are involved: (1) Obtaining a series of sample solutions of monomeric organic pollutants at different concentrations, uniformly mixing the fluorescent probe MOF-74 with each concentration sample, and collecting three-dimensional fluorescence spectrum data of each mixed solution; using the Canny operator to perform edge detection processing, extracting image texture features, and establishing a sample library of characteristic images of each organic matter concentration series; (2) linearly superimposing the concentration series sample characteristic images of each monomer organic pollutant in the sample library obtained in step (1) to obtain a preliminary comprehensive texture characteristic image, and counting non-zero value pixel points; (3) analyzing the preliminary comprehensive texture feature image and the statistical results of the number of non-zero value pixels obtained in step (2), using 1 as a threshold, filtering out pixels less than or equal to 1 in the image, and obtaining comprehensive texture feature images of each category of monomeric organic pollutants; (4) Using the optimized nearest neighbor algorithm to identify unknown pollutants, including the following sub-steps: (4.1) reading the comprehensive texture feature images of each category of monomeric organic pollutants obtained in step (3) into the working area of ​​the data processing software; (4.2) processing the three-dimensional fluorescence spectrum data of the unknown pollutant solution according to step (1) to obtain a three-dimensional fluorescence spectrum image and a corresponding texture feature image; (4.3) First, the texture feature image of the unknown pollutant obtained in step (4.2) is compared with the comprehensive texture feature image samples of each category of monomer organic pollutants in the sample library, and the optimized nearest neighbor algorithm is used to calculate the overall texture similarity or multi-region weighted similarity to distinguish the pollutant category; Specifically, the overall texture similarity is calculated first. If the difference between the similarity and other categories is less than 20%, it is necessary to further calculate the multi-region weighted similarity to improve the discrimination accuracy. The overall texture similarity calculation process is specifically as follows: the number of edge points of the texture feature image of the unknown pollutant is counted, that is, the number of non-zero value pixels in the texture feature image, the texture feature image of the unknown pollutant is superimposed with the comprehensive texture images of various organic pollutants, and the number of pixels with a value of 2 is calculated, and the two are divided to obtain the overall texture similarity; The calculation process of multi-region weighted similarity is as follows: according to the characteristics of the three-dimensional fluorescence spectrum image, the texture feature image and the comprehensive texture image of the unknown pollutant are divided into multiple sub-images, and the overall texture similarity of the corresponding sub-images is calculated respectively. Different weights are set according to the importance of the multiple sub-images, and the multi-region weighted similarity is obtained by adding the weights. One of the similarity calculations is selected according to the needs, and the category of the unknown pollutant solution is obtained after comparison and judgment.

2. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 1, characterized in that: The specific operation of step (2) is as follows: linearly superimpose the texture feature images of the same type of organic pollutants at different concentrations into one image, replace the zero-valued pixels in the superimposed image with "NaN", and then count the number of non-zero-valued pixels and draw a corresponding statistical graph to assist in the subsequent selection of thresholds.

3. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 2, characterized in that: The specific operation of step (3) is as follows: the pixel value of "NaN" is restored to 0 to facilitate the subsequent calculation of relevant judgment indicators in the nearest neighbor algorithm, and to avoid the error situation that "NaN" is still "NaN" after adding the specific value; the non-zero pixel values ​​less than or equal to the threshold 1 are uniformly replaced with 0, and the pixel values ​​greater than the threshold 1 are uniformly replaced with 1, and finally the comprehensive texture feature image of each category of monomer organic pollutants is obtained.

4. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 1, characterized in that: The comprehensive texture feature image extraction in step (2) to step (3) is performed by linearly combining, pixel counting and screening the texture feature images of the same organic pollutant at different concentrations to obtain a comprehensive texture feature image that can represent the three-dimensional fluorescence spectral characteristics of this type of organic pollutant.

5. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 1, characterized in that: The threshold selected in step (3) can be adjusted according to different recognition algorithms. Here, selecting 1 is to remove sample points with no commonality.

6. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 1, characterized in that: The overall texture similarity in step (4.3) is calculated by the following formula: Among them, eq is a function that determines the equality of two parameters within a function, and returns a logical value of 1 when the two values ​​are equal; S(m,n) represents the value of the texture feature image of the unknown organic pollutant solution obtained in step (4.2) at the pixel coordinate (m,n); Y(m,n) represents the value of the comprehensive texture feature image of each category of monomer organic pollutant samples obtained in step (3) at the pixel coordinate (m,n).

7. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 1, characterized in that: The multi-region weighted similarity in step (4.3) is calculated according to the following formula: Among them, eq is a function that determines the equality of two parameters within a function, and returns a logical value of 1 when the two values ​​are equal; x i represents the weight of the i-th sub-image after the texture feature image of the unknown pollutant solution is segmented in step (4.3); S i (j, k) represents the value of the pixel coordinate (j, k) of the i-th sub-image after image segmentation of the texture feature image of the unknown pollutant solution in step (4.3); Y i (j, k) represents the value of the pixel coordinate (j, k) of the i-th sub-image after image segmentation of the comprehensive texture feature image of the sample pollutant in step (4.3).

8. The method for extracting and qualitatively identifying the concentration comprehensive texture features of three-dimensional fluorescence spectrum images according to claim 1, characterized in that: Monomeric organic pollutants are monocyclic aromatic hydrocarbons.

Images