Method and system for microplastic particle profiling based on hyperspectral image big data

Abstract

The present invention relates to a method and a system for microplastic particle profiling based on hyperspectral image big data, in which: an optimal profiling model may be derived by learning hyperspectral image data obtained by measuring field samples of various media and a standard spectrum for each microplastic material; and the position, material, and size of microplastic particles may be profiled in an ultra-precise, rapid, and large-scale manner from hyperspectral images acquired for unknown field samples.

Description

Method and system for profiling microplastic particles based on hyperspectral imaging big data

[0001] The present invention relates to a method and system for profiling microplastic particles based on hyperspectral imaging big data, and more particularly, to a method and system for profiling microplastic particles based on hyperspectral imaging big data, which can rapidly and precisely detect the location, material, and size of microplastic particles included in three-dimensional hyperspectral imaging data of a sample containing various microplastics using probabilistic and artificial neural network models, and visualize the results.

[0002] Recently, reports have surfaced of unintentional contamination of microplastics in bottled water, sea salt, and seafood-based snacks. This has raised concerns about public health, as well as the potential for significant economic losses in the related industries due to safety controversies. Therefore, it is urgent to establish countermeasures.

[0003] Microplastics are 'solid plastic particles with a length or diameter of 5 mm or less'. They are created by the erosion of waste and then enter the aquatic and other environmental ecosystems. They are known to be exposed to humans, the top predators, through various media (fishery products, drinking water, salt, etc.) and cause negative health effects such as tissue inflammation.

[0004] Since the material, type, and size of microplastics are each decisive factors in the cause of diseases (carcinogenesis, endocrine disorders, etc.) and the effects of absorption and movement in the body, microplastic detection technology that can quickly and accurately analyze them is very important.

[0005] Microplastic analysis is typically performed by software in three steps: sample preparation, hyperspectral image measurement, and profiling. In the first step, sample preparation, preprocessing and filtration of non-plastic particles are performed after sample collection. In the second step, hyperspectral image measurement, the prepared sample is measured using a spectroscopic device such as FT-IR (Fourier-transformed infrared spectroscopy) to obtain a three-dimensional hyperspectral image. In the third step, profiling, the raw hyperspectral image data is processed to detect the location and size of microplastics in the hyperspectral image and classify their materials.

[0006] In particular, the profiling technology for microplastics has issues in terms of performance, such as the accuracy of the profiling model and the processing speed of the algorithm, as hyperspectral image data is subject to variability due to factors such as the condition of the field microplastic sample (degree of erosion, overlapping, etc.), and the measurement spectra for each hyperspectral image matrix are organized in three dimensions and are large-scale data.

[0007] Accordingly, there is a need to research and develop a method and system for profiling microplastic particles capable of hyper-precise high throughput processing that can rapidly process large-scale hyperspectral imaging data while effectively adapting to the polymorphism of microplastic field sample states.

[0008] Accordingly, the present invention has been devised to solve the above problems, and the purpose of the present invention is to provide a method and system for profiling microplastic particles based on hyperspectral image big data, which can derive an optimal profiling model by learning the standard spectrum of microplastics by material and hyperspectral image data measured from field samples of various media, and can profile the location, material, and size of microplastic particles with high precision, speed, and mass production from hyperspectral images acquired from unknown field samples.

[0009] In addition, the purpose of the present invention is to provide a method and system for profiling microplastic particles based on hyperspectral imaging big data, which can quickly and precisely detect the location, material, and size of microplastic particles included in three-dimensional hyperspectral imaging data of a sample containing various microplastics using probabilistic and artificial neural network models, and visualize the results.

[0010] However, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present invention belongs from the description below.

[0011] As a technical method for achieving the above object, according to one embodiment of the present invention, a method for profiling microplastic particles based on hyperspectral image big data comprises the steps of: a) registering microplastic hyperspectral image data for each sample in a project folder created by a user using software installed on a terminal, and preprocessing the microplastic hyperspectral image data by an artificial neural network model installed on the terminal; b) performing a profiling operation by operating the artificial neural network model in a profiling analysis mode to calculate the presence or absence of microplastic particles and the material classification results of microplastic particles in grid units from the preprocessed microplastic hyperspectral image data; c) grouping clustered microplastic particles of the same material from the microplastic hyperspectral image data subjected to the profiling operation by the software into a single event, calculating a microplastic particle analysis result, which is the location, material, and size of the microplastic particles, from the microplastic hyperspectral image data, and visualizing the result in the software; d) generating an analysis result report based on the microplastic particle analysis result visualized in the software; and e) a step of generating result data, which is a histogram and probability distribution analysis result by size of microplastic particles, by the software;

[0012] In addition, the software can display the location, material, and size of microplastic particles, which are the results of microplastic particle analysis of microplastic hyperspectral image data registered for each sample in the folder of the project, in the form of a dashboard on the software.

[0013] And in the step a), when the user inputs the hyperspectral image file and the video image file of the sample into the folder of the project, the registration of the microplastic hyperspectral image data can be completed based on the software aligning the image of the hyperspectral image file and the image of the video image file to generate the microplastic hyperspectral image data.

[0014] In addition, the microplastic hyperspectral image data can be implemented in a form in which the location of microplastic particles detected from the image of the hyperspectral image file is displayed on the image of the video image file.

[0015] And in the step a), the software adjusts the wavelength range of each spectrum so that the unknown spectrum in the wavelength vector for the unknown plastic of the microplastic hyperspectral image data and the material-specific standard spectrum of the microplastic particles for a field sample containing the same microplastic particles as the sample of the microplastic hyperspectral image data among the material-specific standard spectra of the microplastic particles stored in the database have the same range, and then determines the value of the wavelength element of the wavelength vector for the unknown plastic, thereby completing the preprocessing of the microplastic hyperspectral image data.

[0016] In addition, the step a) above can perform a preprocessing algorithm according to the following formula for an unknown spectrum matrix composed of wavelength and intensity vectors for each grid of the hyperspectral image file image.

[0017]

[0018] In the above formula, S is a vector of the standard spectrum for each material of the microplastic particles, U is a vector of the unknown spectrum, k is the index of the vector, N is the dimension of the vector, k, w are wavelength values ​​of the vector k index, n is an index for traversal within the spectrum vector, and nsin means a normalized sine function.

[0019] And the step a) above can be performed by comparing the values ​​of the wavelength vector for the unknown plastic based on the resolution used in deep learning using the standard spectrum for each material of the microplastic particle according to the above formula, and if the intervals match, preprocessing of the microplastic hyperspectral image data is omitted, and if the intervals do not match, preprocessing the microplastic hyperspectral image data based on matching the dimensions through interpolation using the nisn function.

[0020] In addition, the step a) above allows the software to remove non-microplastic regions that do not contain microplastic particles that are unnecessary for profiling operations from the hyperspectral image file of the preprocessed microplastic hyperspectral image data.

[0021] And in the step b), at least one of the first artificial neural network model, which is a library search model, and the second artificial neural network model, which is a deep learning model, among the artificial neural network models can perform a profiling operation on microplastic hyperspectral image data.

[0022] In addition, the step b) includes: b-1) a step in which, in the profiling operation process of the library search model, an unknown spectrum from a wavelength vector for an unknown plastic of the microplastic hyperspectral image data and a material-specific standard spectrum of microplastic particles stored in the database are input into the library search model; b-2) a step in which the library search model corrects an intensity value for each wavelength of the unknown spectrum by correcting a mismatched baseline between the unknown spectrum and the material-specific standard spectrum of the microplastic particle; b-3) a step in which the library search model scales and normalizes the unknown spectrum whose wavelength-specific intensity value has been corrected; b-4) a step in which the library search model performs a correlation analysis between the normalized unknown spectrum and the material-specific standard spectrum of the microplastic particle, thereby calculating a material-specific correlation coefficient of the microplastic particle and outputting a vector of grid and material-specific correlation scores; b-5) a step in which the library search model calculates a probability that the microplastic particle of the unknown spectrum is a material of a specific microplastic particle by using the grid and material-specific correlation scores; And b-6) the library search model may include a step of selecting or rejecting a microplastic particle of the unknown spectrum as a candidate for the specific microplastic particle, depending on whether the probability that the microplastic particle of the unknown spectrum is a material of the specific microplastic particle exceeds a threshold value.

[0023] And in the step b-2), the library search model divides the entire wavelength range of the unknown spectrum and the standard spectrum of the material of the microplastic particles into four regions, derives candidate linear models using the least squares method of the first and second orders for each region, and selects the optimal linear model for each region among the candidate linear models based on the following formula, and then corrects the intensity value for each wavelength of the unknown spectrum based on the optimal linear model.

[0024]

[0025] In the above formula, r is the idex of a specific grid in the microplastic hyperspectral image data, U is the measured value of the absorption (reflectance) value of the ith wavelength of the unknown spectrum, is the predicted value of the above candidate linear model, and SE means the root mean square error.

[0026] In addition, the step b) above can calculate the probability that the microplastic particle of the unknown spectrum is a material of a specific microplastic particle based on the following formula using the grid and material-specific correlation score.

[0027]

[0028] In the above formula, mp represents the material of the microplastic particle, N represents the dimension of the candidate microplastic, S represents the match score of the microplastic material, which is the correlation score for each grid and material, and P represents the match probability of the microplastic particle.

[0029] And the step b) above is, b-1) a step in which, in the profiling operation process of the deep learning model, an unknown spectrum from a wavelength vector for an unknown plastic of the microplastic hyperspectral image data and a material-specific standard spectrum of microplastic particles stored in the database are input into the deep learning model; b-2) a step in which the deep learning model differentiates the unknown spectrum and extracts parameter features for the differentiated value of the unknown spectrum; b-3) a step in which the deep learning model normalizes the parameter features to scale and normalize the unknown spectrum; b-4) a step in which the deep learning model drives a material-specific classification model for classifying the material of the microplastic particle of the normalized unknown spectrum and outputs a vector of grid and material-specific energy scores; b-5) a step in which the deep learning model calculates a probability that the microplastic particle of the unknown spectrum is a material of a specific microplastic particle using the grid and material-specific energy scores; And b-6) the deep learning model may include a step of selecting or rejecting a microplastic particle of the unknown spectrum as a candidate for the specific microplastic particle, depending on whether the probability that the microplastic particle of the unknown spectrum is a material of the specific microplastic particle exceeds a threshold value.

[0030] In addition, the step b) above can calculate the probability that the microplastic particle of the unknown spectrum is a material of a specific microplastic particle based on the following formula using the grid and material-specific energy score.

[0031]

[0032] In the above formula, mp represents the material of the microplastic particle, N represents the dimension of the candidate microplastic, S represents the match score of the microplastic material, which is the energy score for each grid and material, and P represents the match probability of the microplastic particle.

[0033] And in the step c), when the software groups clustered microplastic particles of the same material in the microplastic hyperspectral image data into one event, the clustered microplastic particles of the same material in the microplastic hyperspectral image data can be expressed as one microplastic particle in the software.

[0034] The present invention derives an optimal profiling model by learning the standard spectrum of microplastics by material and hyperspectral image data measured from field samples of various media, and can profile the location, material, and size of microplastic particles with high precision, speed, and large quantities from hyperspectral images acquired from unknown field samples.

[0035] In addition, the present invention can quickly and precisely detect the location, material, and size of microplastic particles included in three-dimensional hyperspectral image data of a sample containing various microplastics by utilizing probabilistic and artificial neural network models, and visualize the results.

[0036] However, the effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects that are not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present invention belongs from the description below.

[0037] FIG. 1 is a block diagram illustrating components of a microplastic particle profiling system based on hyperspectral imaging big data according to one embodiment of the present invention.

[0038] FIG. 2 is a flowchart illustrating a process of a method for profiling microplastic particles based on hyperspectral imaging big data according to one embodiment of the present invention, performed by a profiling system for microplastic particles based on hyperspectral imaging big data.

[0039] FIG. 3 is a diagram for explaining a material-specific standard spectrum registration step according to one embodiment of the present invention.

[0040] FIG. 4 is a diagram for explaining the project and sample creation steps and the relationship among the project, sample, and hyperspectral image data according to one embodiment of the present invention.

[0041] Figure 5 is a diagram illustrating an example of a status statistics dashboard displayed when a project is selected.

[0042] FIG. 6 and FIG. 7 are diagrams for explaining a microplastic hyperspectral image data input step according to one embodiment of the present invention.

[0043] FIG. 8 is a diagram for explaining a processing hyperspectral image region reduction step according to one embodiment of the present invention.

[0044] FIG. 9 is a diagram illustrating a detailed process of a profiling analysis step according to one embodiment of the present invention.

[0045] FIGS. 10 to 12 are diagrams for explaining the particle analysis result output and visualization steps according to one embodiment of the present invention.

[0046] FIG. 13 is a diagram for explaining an analysis result report generation step according to one embodiment of the present invention.

[0047] Figure 14 is a drawing for explaining a step of analyzing the distribution of fine particles by size according to a material according to one embodiment of the present invention.

[0048] Figure 15 is a table showing the results of evaluating the performance of an artificial neural network model installed in a terminal according to one embodiment of the present invention.

[0049] Figure 16 is a table showing the results of a comparative evaluation of the performance of an artificial neural network model according to one embodiment of the present invention and a conventional model.

[0050] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily practice the present invention. However, the description of the present invention is merely an embodiment for structural and functional explanation, and therefore the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, since the embodiments can be modified in various ways and can have various forms, the scope of the present invention should be understood to include equivalents that can realize the technical idea. In addition, the purposes or effects presented in the present invention do not mean that a specific embodiment must include all of them or only such effects, and therefore the scope of the present invention should not be construed as being limited thereby.

[0051] The meanings of terms described in the present invention should be understood as follows.

[0052] Terms such as "first" and "second" are intended to distinguish one component from another, and the scope of the rights should not be limited by these terms. For example, a first component could be referred to as a second component, and similarly, a second component could also be referred to as a first component. When a component is referred to as being "connected" to another component, it should be understood that it may be directly connected to that other component, but there may also be other components in between. Conversely, when a component is referred to as being "directly connected" to another component, it should be understood that there are no other components in between. Meanwhile, other expressions describing the relationship between components, such as "between" and "immediately between" or "adjacent to" and "directly adjacent to", should be interpreted similarly.

[0053] Singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as "comprises" or "has" should be understood to specify the presence of stated features, numbers, steps, operations, components, parts, or combinations thereof, but not to exclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0054] Unless otherwise defined, all terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted to be consistent with their meaning within the context of the relevant technology, and should not be interpreted as having ideal or overly formal meanings unless explicitly defined herein.

[0055]

[0056] Microplastic Particle Profiling System Based on Hyperspectral Image Big Data

[0057] Hereinafter, the configuration of a preferred embodiment will be described in detail with reference to the attached drawings.

[0058] FIG. 1 is a block diagram illustrating components of a microplastic particle profiling system based on hyperspectral imaging big data according to one embodiment of the present invention.

[0059] Referring to FIG. 1, a microplastic particle profiling system (1) based on hyperspectral imaging big data can be implemented through a terminal (10) equipped with software (11), a database (12), and an artificial neural network model (13) in one embodiment to detect the location, material, and size of microplastic particles included in three-dimensional hyperspectral imaging data (microplastic hyperspectral imaging data) and visualize the results.

[0060] In the present invention, 3D hyperspectral image data and microplastic hyperspectral image data mean the same data, and accordingly, even if they are used interchangeably in the present invention, it is preferable to understand them as the same data.

[0061] In one embodiment, the software (11) may be a program or application that executes a window (1100) for visualizing the results of profiling operations of microplastic particles and outputs the executed window (1100) on the screen of the terminal (10).

[0062] In the present invention, a window (1100) is a focal program for profiling microplastic particles, and refers to a window form or dialog box for providing a window that configures an interface to a user.

[0063] In one embodiment, the database (12) is a storage device in which data for profiling operations of microplastic particles is stored, and may be composed of first to third databases (12a, 12b, 12c) as illustrated in FIG. 2.

[0064] In one embodiment, the first database (12a) may be a material-specific standard spectrum library database that stores material-specific standard spectra of microplastic particles extracted by the software (11) from a spreadsheet file composed of wavelengths and measurement values ​​(absorption rate, etc.) for the material of microplastic particles registered (or uploaded) by the user to a window (1100) executed through the software (11).

[0065] Here, the material-specific standard spectrum library database (12a) may be a database that stores the material-specific standard spectrum of the microplastic particles so that correlation analysis can be performed between the baseline-corrected unknown spectrum and the material-specific standard spectrum of the microplastic particles in the library model (13a), which is a probabilistic model to be described later.

[0066] Additionally, the standard spectrum library database (12a) by material may be a database used for learning the deep learning model (13b) described later.

[0067] In one embodiment, the second database (12b) may be a hypercube object database that stores three-dimensional hyperspectral image data (210) of field samples containing microplastic particles that a user inputs into a window (1100) executed through software (11).

[0068] Here, the hypercube object database (12b) may be a database including a multidimensional data management function for storing three-dimensional hyperspectral image data (210) including spatial (2D) and spectral (1D) information.

[0069] In one embodiment, the third database (12c) may be a field sample particle analysis result database that stores the microplastic particle analysis results detected by the software (11) by analyzing the three-dimensional hyperspectral image data (210) of the field sample containing microplastic particles.

[0070] Here, the field sample particle analysis result database (12c) may be a database that stores the location, material, and size of microplastic particles, which are the results of the microplastic particle analysis of field samples containing microplastic particles, after converting them into a relational database.

[0071] Additionally, the field sample particle analysis result database (12c) may be a database used for learning the deep learning model (13b) described later.

[0072] In one embodiment, the artificial neural network model (13) is a model learned to perform classification of the presence or absence of microplastic particles and materials for each grid from three-dimensional hyperspectral image data (210), and may be composed of a library search model (13a), which is a first artificial neural network model, and a deep learning model (13b), which is a second artificial neural network model, as shown in FIG. 2.

[0073] In one embodiment, the library search model (13a) can output a grid and material-specific correlation score (S) vector based on calculating a material-specific correlation coefficient through correlation analysis between an unknown spectrum extracted by the software (11) from the 3D hyperspectral image data (210) of a field sample containing microplastic particles input by the user in a window (1100) executed through the software (11) and the material-specific standard spectra of microplastic particles having the same material as the microplastic particles contained in the 3D hyperspectral image data (210) among the material-specific standard spectra of microplastic particles stored in the first database (12a).

[0074] In one embodiment, the deep learning model (13b) can output a grid and material-specific energy score (S) vector based on extracting parameter features for values ​​obtained by differentiating an unknown spectrum extracted by the software (11) from three-dimensional hyperspectral image data (210) of a field sample containing microplastic particles input by a user in a window (1100) executed through the software (11).

[0075] In this way, the profiling system (1) according to one embodiment of the present invention may be a system for quickly and precisely detecting the location, material, and size of microplastic particles included in three-dimensional hyperspectral image data (210) of a field sample containing various microplastic particles by utilizing a probabilistic and artificial neural network model (13), and visualizing the results.

[0076]

[0077] A method for profiling microplastic particles based on hyperspectral imaging big data

[0078] Hereinafter, the process of a microplastic particle profiling method (S100) based on hyperspectral image big data according to one embodiment of the present invention performed by the profiling system (1) will be described in detail with reference to the drawings.

[0079] FIG. 2 is a flowchart illustrating a process of a method for profiling microplastic particles based on hyperspectral imaging big data according to one embodiment of the present invention, performed by a profiling system for microplastic particles based on hyperspectral imaging big data.

[0080] Referring to Fig. 2, before the process of the method for profiling microplastic particles based on hyperspectral image big data (S100) is performed, the method for constructing an artificial neural network (S10) for implementing correlation analysis of a library search model (13a) and generating a deep learning model (13b) may be performed first.

[0081] The above artificial neural network construction method (S10) can be performed in the following order: a microplastic material selection step (S11), a standard spectrum registration step for each material (S12), and a deep learning model creation step (S13).

[0082] In the above microplastic material selection step (S11), the user can select a specific microplastic particle material to be registered as a standard spectrum by uploading a spreadsheet file among the microplastic particle materials presented in the window (1100) executed through the software (11).

[0083] In the above material-specific standard spectrum registration step (S12), the user can upload a spreadsheet file composed of wavelengths and measurement values ​​(absorption rate, etc.) for the material of the microplastic particles selected in the window (1100) through software (11), as shown in FIG. 3.

[0084] FIG. 3 is a diagram for explaining a material-specific standard spectrum registration step according to one embodiment of the present invention.

[0085] Referring to FIG. 3, the window (1100) of the present invention includes a first information category (1101) that displays the material of microplastic particles that can be registered as a standard spectrum.

[0086] Additionally, the window (1100) of the present invention includes a second information category (1102) that displays information of a spreadsheet file uploaded by a user.

[0087] And the window (1100) of the present invention includes a third information category (1103) in which the software (11) displays the standard spectrum for each material of the selected microplastic particles and the microplastic particles extracted from the spreadsheet file after the user completes the selection of microplastic particles to be registered as a standard spectrum in the first information category (1101) and the upload of the spreadsheet file, and the information of the spreadsheet file is displayed in the second information category (1102).

[0088] At this time, it is preferable that the first database (12a) pre-store the standard spectra of microplastic particles by material that can be selected from the first information category (1101) so that the software (11) can display the standard spectra of microplastic particles by material in the third information category (1103).

[0089] In the above deep learning model generation step (S13), the deep learning model (13b) can be generated based on the material-specific standard spectrum of microplastic particles stored in the first database (12a).

[0090] In one embodiment, the deep learning model (13b) is an optimal classification model selected from a material-specific classification model for classifying the material of microplastic particles of an unknown spectrum by learning (MLP, Multi-Layer Perceptron) based on the material-specific standard spectrum of microplastic particles pre-stored in the first database (12a), and can perform classification of the presence or absence of microplastic particles and the material for each grid.

[0091] At this time, the optimal classification model selected as the deep learning model (13b) means a model that can classify microplastic particles of the same material as microplastic particles of an unknown spectrum among multiple material-specific classification models, and more specifically, means a classification model with the highest accuracy in classifying microplastic particles.

[0092] In addition, a material-specific classification model that can be generated using a deep learning model (13b) can be learned based on the material-specific standard spectrum of microplastic particles pre-stored in the first database (12a) and the microplastic particle analysis results when the microplastic particle analysis results are stored in the third database (12c).

[0093] After the process of the artificial neural network model construction method (S10) described above is performed, the software (11) of the present invention can perform a microplastic particle profiling method (S100) based on hyperspectral image big data to detect and visualize the location, material, and size of microplastic particles from microplastic hyperspectral image data (210).

[0094] Referring back to FIG. 2, the method for profiling microplastic particles based on hyperspectral imaging big data (S100) may proceed in the following order: data input and preprocessing step (S110), profiling operation step (S120), particle synthesis and physical quantity calculation step (S130), particle analysis result output and visualization step (S140), analysis result report generation step (S150), and particle size distribution analysis step for microplastics by material (S160).

[0095] The above data input and preprocessing step (S110) may be performed in the following order: project and sample creation step (S111), microplastic hyperspectral image data input step (S112), preprocessing step (S113), and processing hyperspectral image area reduction step (S114).

[0096] In the above project and sample creation step (S111), the user can create a project (20) and a sample (200) in a window (1100) executed through software (11) as shown in FIG. 4.

[0097] FIG. 4 is a diagram for explaining the project and sample creation steps and the relationship among the project, sample, and hyperspectral image data according to one embodiment of the present invention.

[0098] Referring to (A) of FIG. 4, the window (1100) includes a fourth information category (1104) to provide a user with the ability to create a project (20) and a sample (200).

[0099] Referring to (B) of FIG. 4, a project (20) means a management unit of a sample (200), the sample (200) means a management unit of microplastic hyperspectral image data (210), and the hyperspectral image particle analysis data (220) means a microplastic particle analysis result analyzed through the microplastic hyperspectral image data (210).

[0100] That is, it is preferable that the microplastic particle analysis result referred to in the present invention be understood as hyperspectral imaging particle analysis data (220).

[0101] The software (11) of the present invention causes a folder of a project (20) assigned with a project name to be created in the fourth information category (1104) when a user creates a project (20) with a project name on a window (1100) executed through the software (11) using the fourth information category (1104).

[0102] That is, it is preferable to understand that the folder of the project (20) of the present invention is created through software (11).

[0103] At this time, the folder of the project (20) created in the 4th information category (1104) can display the number of microplastic hyperspectral image data (210) for each sample (200) and the analysis status of the microplastic hyperspectral image data (210) for each sample (200).

[0104] In addition, the number of microplastic hyperspectral image data (210) can be displayed on a window (1100) based on the form of a fraction, such as the number of microplastic hyperspectral image data (210) for which analysis of microplastic particles has been completed for each sample (200) / the number of microplastic hyperspectral image data (210) registered in the folder of the project (20).

[0105] Additionally, the analysis status of microplastic hyperspectral image data (210) can be displayed on a window (1100) in the form of a bar.

[0106] And the analysis status of the microplastic hyperspectral image data (210) may be displayed in different forms depending on whether the number of microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) is analyzed, or whether the number of microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) is not analyzed.

[0107] As a specific example, the analysis status of the microplastic hyperspectral image data (210) may be displayed on the window (1100) through a bar of a first color (e.g., blue) when the number of microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) is equal to the number of microplastic hyperspectral image data (210) analyzed, and on the other hand, when the number of microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) is not equal to the number of microplastic hyperspectral image data (210) analyzed, the analysis status may be displayed on the window (1100) through a bar of a second color (e.g., red).

[0108] The folder of the project (20) of the present invention not only provides the number and analysis status of microplastic hyperspectral image data (210), but also provides a function for a user to additionally register microplastic hyperspectral image data (210) to a sample (200) of the project (20).

[0109] The window (1100) of the present invention can generate a fifth information category (1105) that displays hyperspectral image particle analysis data (220), which is a result of analyzing microplastic particles through microplastic hyperspectral image data (210) registered for each sample (200) in the project (20), when a user selects a folder of all or an individual project (20) of the fourth information category (1104).

[0110] Figure 5 is a diagram illustrating an example of a status statistics dashboard displayed when a project is selected.

[0111] Referring to FIG. 5, the software (11) receives the location, material, and size of microplastic particles, which are the results of microplastic particle analysis of microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) selected by the user, from the third database (12c), in conjunction with the third database (12c), and then can display the results of microplastic particle analysis of the microplastic hyperspectral image data (210) in the form of a dashboard such as a chart, graph, or table on the fifth information category (1105).

[0112] In this way, the process in which the software (11) displays the microplastic hyperspectral image data (210) in the form of a dashboard is preferably performed after the profiling operation (analysis) of the microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) is completed, since the microplastic hyperspectral image data (210) must be stored in the third database (12c).

[0113] That is, the software (11) may omit the function of displaying the fifth information category (1105) in the window (1100) before the profiling operation (analysis) of the microplastic hyperspectral image data (210) registered for each sample (200) in the folder of the project (20) is completed.

[0114] In the above microplastic hyperspectral image data input step (S112), after the folder of the project (20) is created, the user can input the microplastic hyperspectral image data (210) to be subjected to profiling calculation into the folder of the project (20).

[0115] FIG. 6 and FIG. 7 are diagrams for explaining a microplastic hyperspectral image data input step according to one embodiment of the present invention.

[0116] Referring to FIG. 6, when a user inputs a hyperspectral image file (210a) of a sample (200) to register microplastic hyperspectral image data (210) of a sample (200) in a folder of a project (20) created in the fourth information category (1105), the software (11) can display an image for a specific wavelength of the hyperspectral image file (210a) on a window (1100).

[0117] At this time, the window (1100) of the present invention can preferentially display an image for the first wavelength (Band=0) of the hyperspectral image file (210a) whenever the hyperspectral image file (210a) is input.

[0118] Additionally, the window (1100) includes a first setting category (1106) in the form of a gauge bar for setting the wavelength for the image of the microplastic hyperspectral image file (210a).

[0119] And the window (1100) includes a second setting category (1107) for enlarging / reducing a microplastic hyperspectral image file (210a) image when a user inputs a signal, measuring the horizontal and vertical lengths of the microplastic hyperspectral image file (210a) image, and displaying the spectrum of a specific point (pixel) from the microplastic hyperspectral image file (210a) image.

[0120] The software (11) of the present invention can display an image of a microplastic hyperspectral image file (210a) for a wavelength set through a gauge bar of a first setting category (1106) on a window (1100).

[0121] Additionally, the software (11) can enlarge or reduce the image of the microplastic hyperspectral image file (210a) when the user requests enlargement or reduction of the image of the microplastic hyperspectral image file (210a) through the second setting category (1107).

[0122] And the software (11) can measure the horizontal and vertical lengths of the microplastic hyperspectral image file (210a) image when the user requests measurement of the horizontal and vertical lengths of the microplastic hyperspectral image file (210a) image through the second setting category (1107).

[0123] In addition, when a user selects a specific point by inputting the coordinates of a specific point through the second setting category (1107) or by inputting a signal such as touching an area of ​​a specific point from an image of a microplastic hyperspectral image file (210a), the software (11) can display the characteristic point spectrum of the microplastic hyperspectral image file (210a) image at the bottom of the microplastic hyperspectral image file (210a) image on a window (1100).

[0124] Additionally, in the microplastic hyperspectral image data input step (S112), the user can input (register) not only the hyperspectral image file (210a) of the sample (200), but also the video image file (210b) of the sample (200) to register the microplastic hyperspectral image data (210) of the sample (200) in the folder of the project (20) of the fourth information category (1105), as shown in FIG. 7.

[0125] Referring to FIG. 7, the video image file (210b) is complementary data of the hyperspectral image file (210a), and refers to image data that captures changes in a sample (200) containing microplastic particles over time in real time, and can be displayed in the form of an image in a window (1100).

[0126] When the software (11) of the present invention registers a video image file (210b) in the fourth information category (1105), the software corrects the image of the video image file (210b) so that the horizontal and vertical lengths are the same and the same point is captured while enlarging / reducing it at the same ratio as the image of the hyperspectral image file (210a), and then aligns the image of the hyperspectral image file (210a) with the image of the video image file (210b) to generate microplastic hyperspectral image data (210).

[0127] That is, the microplastic hyperspectral image data (210) of the present invention is generated through alignment between the image of the hyperspectral image file (210a) and the image of the video image file (210b), and then is registered in the folder of the project (20), and the location of the microplastic particle detected from the image of the hyperspectral image file (210a) can be implemented in a form in which it is displayed on the image of the video image file (210a).

[0128] In addition, microplastic hyperspectral image data (210) can be generated through alignment between the image of the hyperspectral image file (210a) and the image of the video image file (210b), and then registered in the folder of the project (20) for each sample (200).

[0129] In the above preprocessing step (S113), the software (11) can preprocess the microplastic hyperspectral image data (210) registered in the folder of the project (20).

[0130] The software (11) of the present invention can perform a preprocessing algorithm described in [Mathematical Formula 1] below on a matrix of an unknown spectrum composed of wavelength and intensity vectors for each grid of a hyperspectral image file (210a) constituting microplastic hyperspectral image data (210).

[0131]

[0132] In the above [Mathematical Formula 1], S is a vector of a standard spectrum for each material of the microplastic particles, U is a vector of the unknown spectrum, k is an index of the vector, N is a dimension of the vector, k, w are wavelength values ​​of the vector k index, n is an index for traversal within the spectrum vector, and nsin means a normalized sine function.

[0133] The software (11) of the present invention compares the values ​​of the wavelength vector for an unknown plastic based on the resolution used in deep learning using the standard spectrum for each material of the microplastic particle according to the above [Mathematical Formula 1], and if the intervals match, preprocessing of the microplastic hyperspectral image data (210) can be omitted. On the other hand, if the intervals do not match, the microplastic hyperspectral image data (210) can be preprocessed based on matching the dimensions through interpolation using the nisn function.

[0134] In order to perform the preprocessing algorithm of the above [Mathematical Formula 1], the software (11) of the present invention can adjust the wavelength range of each spectrum so that the unknown spectrum in the wavelength vector for the unknown plastic of the microplastic hyperspectral image data (210) and the material-specific standard spectrum of the microplastic particles for a field sample containing the same microplastic particles as the sample (200) of the microplastic hyperspectral image data (210) among the material-specific standard spectra of the microplastic particles stored in the first database (12a) have the same range.

[0135] More specifically, the software (11) can match the upper and lower limits of the wavelength range by removing elements less than the minimum of the unknown spectrum and the standard spectrum for each material and elements exceeding the maximum wavelength from the wavelength vector for the unknown plastic during the wavelength range adjustment process.

[0136] In addition, the software (11) of the present invention can complete preprocessing of microplastic hyperspectral image data (210) by adjusting the wavelength range of each spectrum and then determining the value of the wavelength element of the wavelength vector for unknown plastic.

[0137] And the microplastic hyperspectral image data (210) preprocessed from the software (11) can be stored in the second database (12b).

[0138] In the above processing hyperspectral image area reduction step (S114), the software (11) can remove a non-microplastic area (210a') that does not contain microplastic particles unnecessary for profiling operations from the hyperspectral image file (210a) of the preprocessed microplastic hyperspectral image data (210), as illustrated in FIG. 8.

[0139] FIG. 8 is a diagram for explaining a processing hyperspectral image region reduction step according to one embodiment of the present invention.

[0140] Referring to FIG. 8, the software (11) of the present invention can reduce the processing target range by removing non-microplastic areas (210a') unnecessary for profiling operations from the hyperspectral image file (210a) of preprocessed microplastic hyperspectral image data (210) based on the distribution and discriminant function of the spectral average.

[0141] In this way, the software (11) removes the non-microplastic region (210a') unnecessary for the profiling operation, because in spectrum analysis, the number of times the profiling operation is required to be performed is the number of times the horizontal × vertical grid (typically, 500 × 500), which takes a considerable amount of time, and thus, the number of profiling operations and the time are reduced.

[0142] Accordingly, the preprocessed microplastic hyperspectral image data (210) may be data on which a profiling operation is performed on the remaining area excluding the area corresponding to the non-microplastic area (210a') of the hyperspectral image file (210) from the artificial neural network model (13a, 13b).

[0143] Referring again to FIG. 2, the profiling operation step (S120) may be a step in which the software (11) and the artificial neural network model (13) operate in a profiling analysis mode (S121) from the preprocessed microplastic hyperspectral image data (210) to perform a profiling operation on the microplastic hyperspectral image data (210).

[0144] After this profiling analysis mode operation (S121), the process of the profiling operation step (S120) is as shown in Fig. 9.

[0145] FIG. 9 is a diagram illustrating a detailed process of a profiling analysis step according to one embodiment of the present invention.

[0146] Referring to FIG. 9, in the profiling operation step (S120), at least one of the library search model (13a) and the deep learning model (13b) can perform a profiling operation on microplastic hyperspectral image data (210).

[0147] Below, the profiling operation process of the library search model (13a) and the profiling operation process of the deep learning model (13b) will be described in detail.

[0148] In addition, when both the library search model (13a) and the deep learning model (13b) are driven to perform profiling operations, the profiling operation process of the library search model (13a) and the profiling operation process of the deep learning model (13b) may be performed simultaneously or sequentially.

[0149] In the profiling operation process of the library search model (13a), the unknown spectrum of the unknown plastic of the microplastic hyperspectral image data (210) and the standard spectrum by material of the microplastic particles pre-stored in the first database (12a) used for learning the deep learning model (13b) can be input into the library search model (13a) of the present invention (S122).

[0150] After the above input step (S122), the library search model (13a) can correct the inconsistent baseline between the unknown spectrum and the standard spectrum for each material of the microplastic particles due to overlapping of multiple microplastic particles or differences in measurement conditions (S123a).

[0151] In the above baseline correction step (S123a), the library search model (13a) divides the entire wavelength range of the input unknown spectrum and the standard spectrum by material of the microplastic particles into four regions, derives a candidate linear model using the least square method of the first order (linear) and second order (quadratic) for each region, and selects the optimal linear model for each region based on the RMSE (Root mean square deviation) of [Mathematical Formula 2] below among the candidate linear models, and then corrects the intensity value for each wavelength of the unknown spectrum based on the optimal linear model.

[0152]

[0153] In the above [Mathematical Formula 2], r is the idex of a specific grid in the microplastic hyperspectral image data, U is the measured value of the absorption (reflectivity) value of the i-th wavelength of the unknown spectrum, is the predicted value of the above candidate linear model, and SE means the root mean square error.

[0154] After the baseline correction step (S123a), the library search model (13a) can scale the unknown spectrum whose wavelength-specific intensity values ​​have been corrected (S124).

[0155] In the above data normalization step (S124), the library search model (13a) can normalize the unknown spectrum based on Min-Max Normalization, which scales the value of the unknown spectrum to between 0 and 1, or Z-Score Normalization, which scales the mean of the unknown spectrum to 0 and the standard deviation to 1.

[0156] After the above data normalization step (S124), the library search model (13a) performs a correlation analysis between the normalized unknown spectrum and the standard spectrum of each material of the microplastic particles stored in the first database (12a) (S125a), thereby calculating the correlation coefficient of each material of the microplastic particles and outputting a vector of grid and material-specific correlation scores (S) (S126a).

[0157] After the above correlation score vector output step (S126), the library search model (13a) can calculate the probability (P) that a microplastic particle of an unknown spectrum is a material of a specific microplastic particle based on the following [Mathematical Formula 3] using the grid and material-specific correlation score (S) (S127).

[0158]

[0159] In the above [Mathematical Formula 3], mp represents the material of the microplastic particle, N represents the dimension of the candidate microplastic, S represents the match score of the microplastic material (correlation score or energy score by grid and material), and P represents the match probability of the microplastic particle.

[0160] In the above multi-class classification probability calculation step (S127), the standard spectrum for each material of the microplastic particles has characteristic wavelength and intensity values ​​for each material of the microplastic particles, and the library search model (13a) can calculate the probability (P) that the unknown spectrum belongs to each class (material of the microplastic particle).

[0161] After the multi-class classification probability calculation step (S127), the library search model (13a) can determine whether the probability (P) that a microplastic particle of an unknown spectrum is a material of a specific microplastic particle exceeds a threshold value (S128).

[0162] At this time, if the probability (P) that the microplastic particle of the unknown spectrum is the material of a specific microplastic particle exceeds the threshold (S128-NO), the library search model (13a) can select the microplastic particle of the unknown spectrum exceeding the threshold as a candidate for the specific microplastic particle (S129a).

[0163] In contrast, if the probability (P) that a microplastic particle of an unknown spectrum is the material of a specific microplastic particle does not exceed a threshold (S128-YES), the library search model (13a) can reject microplastic particles of an unknown spectrum that do not exceed the threshold as candidates for the specific microplastic particle (S129b).

[0164] Meanwhile, in the profiling operation process of the deep learning model (13b), the unknown spectrum in the wavelength vector for the unknown plastic of the microplastic hyperspectral image data (210) and the standard spectrum by material of the microplastic particles pre-stored in the first database (12a) used for learning the deep learning model (13b) can be input into the deep learning model (13b) of the present invention (S122).

[0165] After the above input step (S122), the deep learning model (13b) can differentiate the unknown spectrum (S123b).

[0166] In the above differentiation step (S123b), the deep learning model (13b) can extract parameter features for the differentiated values ​​of the unknown spectrum.

[0167] At this time, the parameter features extracted from the differential value of the unknown spectrum by the deep learning model (13b) are not limited in type, but may include the median (mean), variability (standard deviation), minimum, maximum, skewness and kurtosis of the unknown spectrum intensity, etc.

[0168] After the above differentiation step (S123b), the deep learning model (13b) can normalize the parameter features to scale the unknown spectrum (S124).

[0169] In the above data normalization step (S124), the deep learning model (13b) can normalize the unknown spectrum based on Min-Max Normalization, which scales the value of the unknown spectrum to between 0 and 1, or Z-Score Normalization, which scales the mean of the unknown spectrum to 0 and the standard deviation to 1.

[0170] After the above data normalization step (S124), the deep learning model (13b) can run a material-specific classification model to classify the material of the microplastic particles of the normalized unknown spectrum (S125b), thereby outputting a vector of grid and material-specific energy scores (S) (S126b).

[0171] After the above energy score vector output step (S126b), the deep learning model (13b) can calculate the probability (P) that the microplastic particle of an unknown spectrum is a material of a specific microplastic particle based on the above [Mathematical Formula 3] using the energy score (S) for each grid and material (S127).

[0172] That is, the multi-class classification probability calculation step (S127) may be a step of calculating the probability (P) that the unknown spectrum belongs to each class (material of microplastic particles) based on at least one of the grid and material-specific correlation score (S) and the grid and material-specific energy score (S).

[0173] After the multi-class classification probability calculation step (S127), the deep learning model (13b) can determine whether the probability (P) that a microplastic particle of an unknown spectrum is a material of a specific microplastic particle exceeds a threshold (S128).

[0174] At this time, if the probability (P) that the microplastic particle of the unknown spectrum is the material of a specific microplastic particle exceeds the threshold (S128-NO), the deep learning model (13b) can select the microplastic particle of the unknown spectrum exceeding the threshold as a candidate for the specific microplastic particle (S129a).

[0175] In contrast, if the probability (P) that a microplastic particle of an unknown spectrum is the material of a specific microplastic particle does not exceed a threshold (S128-YES), the deep learning model (13b) can reject a microplastic particle of an unknown spectrum that does not exceed the threshold as a candidate for a specific microplastic particle (S129b).

[0176] In the above profiling operation step (S120), when microplastic hyperspectral image data (210) is input to at least one of the library search model (13a) and the deep learning model (13b) after the data input and preprocessing step (S110) is completed, at least one of the library search model (13a) and the deep learning model (13b) that has received the microplastic hyperspectral image data (210) is automatically driven to perform a profiling operation on the microplastic hyperspectral image data (210).

[0177] However, the above profiling operation step (S120) may be performed as a modified example when the user requests the operation of at least one of the library search model (13a) and the deep learning model (13b) through the window (1100) in the particle synthesis and physical quantity calculation step (S130).

[0178] In this modified example, it is preferable that the profiling method (S100) of microplastic particles based on hyperspectral imaging big data is performed with the profiling calculation step (S120) after the particle synthesis and physical quantity calculation step (S130).

[0179] Referring again to FIG. 2, the particle synthesis and physical quantity calculation step (S130) may be a step in which software (11) synthesizes microplastic particles of microplastic hyperspectral image data (210) and calculates the location, material, and size of microplastic particles of microplastic hyperspectral image data (210) in conjunction with an artificial neural network model (13a, 13b).

[0180] At this time, the reason why the particle synthesis and physical quantity calculation step (S130) must be performed is that the presence or absence of microplastic particles and the material classification results are output in a window (1100) in grid units for the microplastic particles profiled by at least one of the library search model (13a) and the deep learning model (13b), and in order to evaluate the possibility of body injection, detection of microplastic particles in which identical grids are clustered is required.

[0181] Therefore, a synthesis process of clustered microplastic particles (microplastic lattices) with identical grids is required, and smoothing is required because actual physical microplastic particles are composed of round (circular) particles rather than rectangular ones.

[0182] For the microplastic grids detected for grid synthesis, hyperspectral image data (210) for each material of microplastic particles are reconstructed, and through image scanning, adjacent microplastic grids are collected to construct microplastic particle events, and the microplastic grids belonging to the microplastic grids are aligned and stored.

[0183] Afterwards, the location, material, and size of the microplastic particles can be calculated using the microplastic particle event.

[0184] In addition, the particle analysis result output and visualization step (S140) is a step performed after the particle synthesis and physical quantity calculation step (S130), and may be a step of outputting and visualizing the location, material, and size of microplastic particles of the microplastic hyperspectral image data (210) produced in the particle synthesis and physical quantity calculation step (S130) and providing them to the user.

[0185] In order to proceed with the particle synthesis and physical quantity calculation step (S130) and the particle analysis result output and visualization step (S140), the user can request the operation of at least one of the library search model (13a) and the deep learning model (13b) through the window (1100).

[0186] The window (1100) of the present invention includes a drive request category (1108) for requesting the drive of at least one of a library search model (13a) and a deep learning model (13b), as illustrated in FIG. 10, so that a user can request the drive of an artificial neural network model (13a, 13b).

[0187] FIGS. 10 to 12 are diagrams for explaining the particle analysis result output and visualization steps according to one embodiment of the present invention.

[0188] Referring to FIG. 10, when a user requests the operation of at least one of a library search model (13a) and a deep learning model (13b) through a drive request category (1108), the software (11) of the present invention may receive the result of a profiling operation of microplastic hyperspectral image data (210) performed in at least one of the library search model (13a) and the deep learning model (13b) requested by the user from at least one of the library search model (13a) and the deep learning model (13b), or may perform a profiling operation in at least one of the library search model (13a) and the deep learning model (13b) and then receive the result of a profiling operation from at least one of the library search model (13a) and the deep learning model (13b) in which the profiling operation has been performed.

[0189] The window (1100) of the present invention includes an image window (1109) and an analysis result window (1110) for displaying the profiling operation results of microplastic hyperspectral image data (210) received from at least one of a library search model (13a) and a deep learning model (13b) by the software (11), as illustrated in FIG. 10.

[0190] Referring to FIG. 10, the image window (1109) can display microplastic hyperspectral image data (210) on which a profiling operation is performed to select microplastic particles of an unknown spectrum as candidates for specific microplastic particles in at least one of a library search model (13a) and a deep learning model (13b) on a window (1100).

[0191] At this time, the microplastic hyperspectral image data (210) displayed in the image window (1109) means an image of the microplastic hyperspectral image data (210) in which the hyperspectral image file (210a) and the video image file (210b) are aligned, and can be stored in the second database (12b).

[0192] And the image of the microplastic hyperspectral image data (210) can be displayed in an overlapping manner by material of microplastic particles adopted from at least one of the library search model (13a) and the deep learning model (13b), which are the results of the profiling operation, in the image window (1109).

[0193] Referring to FIG. 10, the analysis result window (1110) can display the microplastic particle analysis results in the form of a list (table) on a window (1100) that synthesizes the location (center coordinate), material and size (area, long axis, short axis, etc.) of the microplastic particles detected from the microplastic hyperspectral image data (210) and the degree of agreement of the microplastic particles detected through at least one profiling operation among the library search model (13a) and the deep learning model (13b) driven through the drive request category (1108).

[0194] The window (1100) of the present invention includes an event row (1111) for synthesizing a specific microplastic particle among the microplastic particles sorted in the analysis result window (1110) or displaying the spectrum of a specific microplastic particle in the window (1100).

[0195] Additionally, the window (1100) includes a spectrum display window (1112) for displaying the spectrum of a specific microplastic particle selected through an event row (1111).

[0196] The software (11) of the present invention, when a user inputs a signal into an event row (1111) to select at least one microplastic particle among a plurality of microplastic particles arranged in an analysis result window (1110) and requests spectrum display, causes the spectrum of the microplastic particle selected through the event row (1111) to be displayed on a window (1100) through a spectrum display window (1112).

[0197] Referring to FIG. 11, the software (11) of the present invention, when a user wants to synthesize clustered microplastic particles of the same material but with differences in pixels in the microplastic hyperspectral image data (210) displayed from the image window (1109) for the synthesis of microplastic particles and group them into one event (A), by selecting an adjacent microplastic grid of the same material through the event row (1111) and requesting particle synthesis, the synthesis of the corresponding microplastic particles selected through the event row (1111) can proceed.

[0198] Referring to FIG. 12, clustered microplastic particles of the same material in the microplastic hyperspectral image data (210) are displayed as being separated in the image window (1109) due to differences in pixels before synthesis, but when synthesis is performed through software (11), they can be composed of one event (A) and displayed as one microplastic particle in the image window (1109).

[0199] Referring back to FIG. 2, the analysis result report generation step (S150) may be a step of generating an analysis result report (30) based on a microplastic particle analysis result that synthesizes the location (center coordinate), material and size (area, long axis, short axis, etc.) of the microplastic particles detected from the microplastic hyperspectral image data (210) displayed in the analysis result window (1110) by the software (11) of the present invention and the degree of agreement of the microplastic particles detected through the profiling operation of at least one of the library search model (13a) and the deep learning model (13b) driven through the drive request category (1108), and then displaying the analysis result report on a window (1100) as illustrated in FIG. 13.

[0200] FIG. 13 is a diagram for explaining an analysis result report generation step according to one embodiment of the present invention.

[0201] Referring to FIG. 13, when the software (11) of the present invention displays the microplastic particle analysis results in the analysis results window (1110) through the particle analysis results output and visualization step (S140), and the user requests the creation of an analysis results report (30) in the window (1100), the software can create an analysis results report (30) that analyzes the microplastic particle analysis results displayed in the analysis results window (1110) through a table, graph, or image.

[0202] At this time, the analysis result report (30) can be generated by the software (11) and stored in the terminal (10) in PDF format.

[0203] In addition, when the user selects an area to be extracted from the window (1100), the analysis result report (30) can be downloaded as a spreadsheet file and a high-resolution image file and stored in the terminal (10) after extracting only the data of the selected area.

[0204] Referring again to FIG. 2, the step (S160) of analyzing the particle size distribution of microplastics by material may be a step in which the software (11) of the present invention generates result data (40), which is a histogram and probability distribution analysis result of microplastic particles by size, as shown in FIG. 14, and then displays it in a window (1100).

[0205] Figure 14 is a drawing for explaining a step of analyzing the distribution of fine particles by size according to a material according to one embodiment of the present invention.

[0206] Referring to FIG. 14, the software (11) of the present invention can display result data (40) through a window (1100) after the analysis result report generation step (S150).

[0207] In the step (S160) of analyzing the distribution of microplastic particles by size for each material, the result data (40) can be filtered to display the histogram and probability distribution analysis results by size of the selected specific microplastic particle when the user selects the specific microplastic particle through the window (1100).

[0208] When a user inputs a signal into a frequency distribution chart of result data (40) through a window (1100), the software (11) of the present invention displays microplastic particles in the frequency distribution chart in which the signal is input, and moves to and tracks the microplastic particles.

[0209]

[0210] Performance Evaluation

[0211] Below, the performance evaluation results of the artificial neural network model (13a, 13b) of the present invention will be described in detail with reference to the drawings.

[0212] Figure 15 is a table showing the results of evaluating the performance of an artificial neural network model installed in a terminal according to one embodiment of the present invention.

[0213] Referring to Fig. 15, the artificial neural network model (13a, 13b) performed a profiling operation on microplastic hyperspectral image data (210) of a field sample containing various microplastic particles.

[0214] At this time, the microplastic particles include polyamide, etc., as shown in FIG. 15, and the microplastic hyperspectral image data (210) can be set in multiple sets for learning and testing of the artificial neural network model (13a, 13b).

[0215] As a result of performing profiling operations on microplastic hyperspectral image data (210) of field samples containing microplastic particles, the artificial neural network model (13a, 13b) of the present invention was confirmed to have a profiling operation performance of 100% for most microplastic particles.

[0216]

[0217] Performance Comparison Evaluation

[0218] Below, with reference to the drawings, the evaluation results comparing the performance of the artificial neural network model (13a, 13b) of the present invention with that of a conventional model will be described in detail.

[0219] Figure 16 is a table showing the results of a comparative evaluation of the performance of an artificial neural network model according to one embodiment of the present invention and a conventional model.

[0220] Referring to Fig. 16, the artificial neural network model (13a, 13b) and the conventional model each performed profiling operations on microplastic hyperspectral image data (210) of field samples containing various microplastic particles.

[0221] The artificial neural network model (13a, 13b) of the present invention took an average of 26 seconds to complete the profiling operation, while the conventional model took an average of 1 minute and 10 seconds (70 seconds). Accordingly, it was confirmed that the artificial neural network model (13a, 13b) of the present invention had a profiling operation performance 2.6 times higher than that of the conventional model.

[0222]

[0223] The detailed description of the preferred embodiments of the present invention disclosed above has been provided to enable those skilled in the art to implement and practice the present invention. While the above description has been made with reference to preferred embodiments of the present invention, those skilled in the art will appreciate that various modifications and variations can be made to the present invention without departing from the scope of the present invention. For example, those skilled in the art can utilize the individual components described in the above-described embodiments in combination with each other. Accordingly, the present invention is not intended to be limited to the embodiments described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0224] The present invention may be embodied in other specific forms without departing from the technical spirit and essential characteristics thereof. Therefore, the above detailed description should not be construed as limiting in all respects but should be considered illustrative. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all changes coming within the equivalent scope of the present invention are intended to be included therein. The present invention is not intended to be limited to the embodiments set forth herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein. Furthermore, claims that are not explicitly cited in the claims may be combined to form an embodiment or incorporated into a new claim by post-application amendment.

[0225] The method and system for profiling microplastic particles based on hyperspectral imaging big data of the present invention learns the standard spectrum of microplastics by material and hyperspectral imaging data measured from field samples of various media to derive an optimal profiling model, and can profile the location, material, and size of microplastic particles from hyperspectral images acquired from unknown field samples with high precision, quickly, and in large quantities, and can quickly and precisely detect the location, material, and size of microplastic particles included in three-dimensional hyperspectral imaging data of samples containing various microplastics using probabilistic and artificial neural network models, and visualize the results, and therefore has industrial applicability.

Claims

1. a) A step in which a user registers microplastic hyperspectral image data for each sample in a project folder created using software installed on a terminal, and an artificial neural network model installed on the terminal preprocesses the microplastic hyperspectral image data; b) a step of performing a profiling operation by operating the artificial neural network model in a profiling analysis mode to produce the presence or absence of microplastic particles and the material classification results of microplastic particles in grid units from the preprocessed microplastic hyperspectral image data; c) A step of grouping clustered microplastic particles of the same material into one event by synthesizing the clustered microplastic particles of the same material from the microplastic hyperspectral image data profiled by the software, and then calculating the microplastic particle analysis results, which are the location, material, and size of the microplastic particles, from the microplastic hyperspectral image data and visualizing them in the software; d) a step of generating an analysis result report based on the microplastic particle analysis results visualized in the above software; and e) A method for profiling microplastic particles based on hyperspectral imaging big data, characterized in that it comprises a step of generating result data which is a histogram and probability distribution analysis result of the size of the microplastic particles by the software.

2. In paragraph 1, The above software, A method for profiling microplastic particles based on hyperspectral imaging big data, characterized in that the location, material, and size of microplastic particles, which are the results of microplastic particle analysis of microplastic hyperspectral imaging data registered for each sample in the folder of the above project, are displayed in the form of a dashboard on the software.

3. In paragraph 1, Step a) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that when the user inputs a hyperspectral image file and a video image file of the sample into a folder of the project, the software aligns the image of the hyperspectral image file and the image of the video image file to generate the microplastic hyperspectral image data, thereby completing the registration of the microplastic hyperspectral image data.

4. In paragraph 3, The above microplastic hyperspectral imaging data is, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the location of microplastic particles detected from the image of the hyperspectral image file is implemented in a form displayed on the image of the video image file.

5. In paragraph 3, Step a) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the software adjusts the wavelength range of each spectrum so that the unknown spectrum in the wavelength vector for the unknown plastic of the microplastic hyperspectral image data and the material-specific standard spectrum of the microplastic particles for a field sample containing the same microplastic particles as the sample of the microplastic hyperspectral image data among the material-specific standard spectra of the microplastic particles stored in the database have the same range, and then determines the value of the wavelength element of the wavelength vector for the unknown plastic to complete the preprocessing of the microplastic hyperspectral image data.

6. In paragraph 5, Step a) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the software performs a preprocessing algorithm according to the following formula on an unknown spectrum matrix composed of wavelength and intensity vectors for each grid of the hyperspectral image file image. In the above formula, S is a vector of the standard spectrum for each material of the microplastic particles, U is a vector of the unknown spectrum, k is the index of the vector, N is the dimension of the vector, k, w are wavelength values ​​of the vector k index, n is an index for traversal within the spectrum vector, and nsin means a normalized sine function.

7. In paragraph 6, Step a) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the software compares the values ​​of the wavelength vector for the unknown plastic based on the resolution used in deep learning using the standard spectrum for each material of the microplastic particle according to the formula, and if the intervals match, preprocessing of the microplastic hyperspectral image data is omitted, and if the intervals do not match, the dimensions are matched through interpolation using the nisn function, thereby preprocessing the microplastic hyperspectral image data.

8. In paragraph 5, Step a) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the software removes non-microplastic regions that do not contain microplastic particles unnecessary for profiling operations from a hyperspectral image file of the preprocessed microplastic hyperspectral image data.

9. In paragraph 5, Step b) above, A method for profiling microplastic particles based on hyperspectral imaging big data, characterized in that at least one of the first artificial neural network model, which is a library search model, and the second artificial neural network model, which is a deep learning model, among the above artificial neural network models performs a profiling operation on microplastic hyperspectral imaging data.

10. In paragraph 9, Step b) above, b-1) In the profiling operation process of the library search model, a step of inputting an unknown spectrum from a wavelength vector for an unknown plastic of the microplastic hyperspectral image data and a standard spectrum by material of the microplastic particles stored in the database into the library search model; b-2) A step of correcting the intensity value of each wavelength of the unknown spectrum by correcting the inconsistent baseline of the library search model and the material-specific standard spectrum of the microplastic particles; b-3) A step of normalizing the unknown spectrum by scaling the wavelength-specific intensity values ​​corrected by the library search model; b-4) A step of performing a correlation analysis of the normalized unknown spectrum of the library search model and the standard spectrum of the microplastic particles by material, calculating a correlation coefficient by material of the microplastic particles, and outputting a vector of grid and material-specific correlation scores; b-5) A step in which the library search model calculates the probability that the microplastic particle of the unknown spectrum is a material of a specific microplastic particle using the grid and material-specific correlation scores; and b-6) A method for profiling microplastic particles based on hyperspectral imaging big data, characterized in that it comprises a step of selecting or rejecting microplastic particles of the unknown spectrum as candidates for the specific microplastic particles, depending on whether the probability that the microplastic particles of the unknown spectrum are the material of the specific microplastic particles exceeds a threshold value, by the library search model.

11. In paragraph 10, Step b-2) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the library search model divides the entire wavelength range of the unknown spectrum and the standard spectrum by material of the microplastic particles into four regions, derives candidate linear models using the least squares method of the first and second orders for each region, selects the optimal linear model for each region among the candidate linear models based on the following equation, and then corrects the intensity value for each wavelength of the unknown spectrum based on the optimal linear model. In the above formula, r is the idex of a specific grid in the microplastic hyperspectral image data, U is the measured value of the absorption (reflectance) value of the ith wavelength of the unknown spectrum, is the predicted value of the above candidate linear model, and SE means the root mean square error.

12. In paragraph 10, Step b) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the probability that the microplastic particles of the unknown spectrum are of a specific microplastic particle material is calculated based on the following formula using the above grid and material-specific correlation scores. In the above formula, mp represents the material of the microplastic particle, N represents the dimension of the candidate microplastic, S represents the match score of the microplastic material, which is the correlation score for each grid and material, and P represents the match probability of the microplastic particle.

13. In paragraph 9, Step b) above, b-1) In the profiling operation process of the deep learning model, a step in which an unknown spectrum from a wavelength vector for an unknown plastic of the microplastic hyperspectral image data and a standard spectrum by material of microplastic particles stored in the database are input into the deep learning model; b-2) A step in which the deep learning model differentiates the unknown spectrum and extracts parameter features for the differentiated value of the unknown spectrum; b-3) A step in which the deep learning model normalizes the parameter features and scales and normalizes the unknown spectrum; b-4) A step in which the deep learning model drives a material-specific classification model to classify the material of the microplastic particles of the normalized unknown spectrum, thereby outputting a vector of grid and material-specific energy scores; b-5) A step in which the deep learning model calculates the probability that the microplastic particles of the unknown spectrum are of a specific microplastic particle material using the grid and material-specific energy scores; and b-6) A method for profiling microplastic particles based on hyperspectral image big data, characterized in that it comprises a step of selecting or rejecting microplastic particles of the unknown spectrum as candidates for the specific microplastic particles, depending on whether the probability that the microplastic particles of the unknown spectrum are the material of the specific microplastic particles exceeds a threshold value.

14. In paragraph 13, Step b) above, A method for profiling microplastic particles based on hyperspectral image big data, characterized in that the probability that the microplastic particles of the unknown spectrum are of a specific microplastic particle material is calculated based on the following formula using the grid and material-specific energy score. In the above formula, mp represents the material of the microplastic particle, N represents the dimension of the candidate microplastic, S represents the match score of the microplastic material, which is the energy score for each grid and material, and P represents the match probability of the microplastic particle.

15. In paragraph 1, Step c) above, A method for profiling microplastic particles based on hyperspectral imaging big data, characterized in that when the software groups clustered microplastic particles of the same material in the microplastic hyperspectral imaging data into one event, the clustered microplastic particles of the same material in the microplastic hyperspectral imaging data are expressed as one microplastic particle in the software.

Images