Image data processing method for neomycin spectrogram based on deep learning model

Abstract

The invention relates to the technical field of image processing, in particular to an image data processing method for neomycin spectrograms based on a deep learning model, and the method comprises the steps: obtaining initial neomycin spectrogram data, inputting each neomycin spectrogram into a target deep learning model, and obtaining a corresponding image feature vector; obtaining classification cluster data of neomycin according to the image feature vector; it can be known that through the time sequence relation degree between the image features of any two initial neomycin images and then based on the time sequence relation degree, classification processing is performed on the initial neomycin images, the time relation between the neomycin images can be analyzed, data of neomycin changing along with time is established, and the data size of the neomycin images is increased.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method for processing image data of a neomycin spectrum based on a deep learning model. Background Art

[0002] Neomycin is an aminoglycoside antibiotic obtained through fermentation. It consists of three main components: neomycin A, neomycin B, and neomycin C, with neomycin B being the primary component. Pharmacopoeias worldwide specify requirements for the content of each neomycin component, with neomycin A content no greater than 2% and neomycin C content no greater than 15%. In products produced through current fermentation processes, neomycin A rarely exceeds the standard range. However, because neomycin C and neomycin B are stereoisomers, and neomycin B is synthesized from neomycin C via epimerase, neomycin C is a necessary precursor for the synthesis of neomycin B, and technological means cannot be used to bypass neomycin C in the synthesis pathway.

[0003] Biological activity testing found that the antibacterial activity of neomycin C is only one-third of that of neomycin B, and it has severe toxic side effects, making it one of the controlled impurities in the quality standards of various countries.

[0004] With the rapid development of deep learning technology in various fields, deep learning has also further advanced proteomic analysis in the field of chemomics. Some studies use tools such as MSGF+ and MaxQuant to analyze raw mass spectra to identify markers. After obtaining differential markers through statistical analysis, they use traditional machine learning methods such as decision trees and support vector machines to establish disease diagnosis and classification models. Some studies use convolutional neural networks to extract features from mass spectrometry data in the form of graphs, and then perform marker identification and establish disease diagnosis models. These classification results are overly dependent on the accuracy of marker identification. Since marker identification is prone to incomplete identification or poor correlation with the disease, the accuracy of downstream diagnostic models is reduced.

[0005] However, deep learning has not been used for the neomycin component before. At the same time, the amount of neomycin data is relatively small. When a certain amount of neomycin image data is obtained, it is impossible to accurately analyze the time relationship between neomycin images and establish data on neomycin changes over time. Summary of the Invention

[0006] According to the present invention, a method for processing image data of a neomycin spectrum based on a deep learning model is provided, the method comprising the following steps: S100, obtain the initial neomycin spectrum data A={A1, ..., A i ,……,A m}, Ai is the i-th initial neomycin spectrum, i ranges from 1 to m, and m is the number of initial neomycin spectra.

[0007] S200, each A i Input into the target deep learning model to obtain A i The corresponding image feature vector B i ={B i1 ,……,B ij ,……,B in}, B ij It's A i The jth image feature in the image, the value range of j is 1 to n, and n is A i The corresponding number of image features.

[0008] S300, according to each B i , get the classification cluster data of neomycin U={U1,……,U t ,……,U β},U t is the t-th neomycin classification cluster, t ranges from 1 to β, and β is the number of neomycin classification clusters.

[0009] The step S300 further includes the following steps: S301, according to B i With B except B i The temporal relationship degree F constructed between other image feature vectors other than i ={F i1 ,……,F iu ,……,F iη}, F iu It's B i With B except B i The temporal relationship between the u-th image feature vectors, F iu Meet the following conditions: , B 0 uj It is B except B i The jth image feature in the uth image feature vector among the other image feature vectors except B, where u ranges from 1 to η, and η is the value of B except B. i The total amount of other image feature vectors other than η=n-1, where △B g is the classification feature threshold of the g-th acquisition time, where △B gThe acquisition is to first obtain the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image in a certain neomycin component, and then obtain the average of the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image of all neomycin components, and max() is the maximum value function.

[0010] S302: Generate a target vector temporal relationship matrix F based on the initial vector temporal relationship matrix F and a preset temporal relationship threshold ΔF. 0 , F meets the following conditions: .

[0011] S303, according to the temporal relationship matrix F between target vectors 0 , generate the classification cluster data U={U1,...,U t ,……,U β}.

[0012] Compared with the prior art, the present invention has at least the following beneficial effects: The present invention provides a method for processing image data of neomycin spectrum based on a deep learning model, the method comprising: S100, obtaining initial neomycin spectrum data A={A1, ..., A i ,……,A m}, A i is the i-th initial neomycin spectrum, i ranges from 1 to m, and m is the number of initial neomycin spectra; S200, each A i Input into the target deep learning model to obtain A i The corresponding image feature vector B i ={B i1 ,……,B ij ,……,B in}, B ij It's A i The jth image feature in the image, the value range of j is 1 to n, and n is A i The number of corresponding image features; S300, according to each B i , get the classification cluster data of neomycin U={U1,……,U t ,……,U β},U t is the t-th neomycin classification cluster, the value range of t is 1 to β, β is the number of neomycin classification clusters; wherein, the step S300 also includes the following steps: S301, according to B i With B except B i The temporal relationship degree F constructed between other image feature vectors other than i ={Fi1 ,……,F iu ,……,F iη}, F iu It's B i With B except B i The temporal relationship between the u-th image feature vectors, F iu Meet the following conditions: , B 0 uj It is B except B i The jth image feature in the uth image feature vector among the other image feature vectors except B, where u ranges from 1 to η, and η is the value of B except B. i The total amount of other image feature vectors other than η=n-1, where △B g is the classification feature threshold of the g-th acquisition time, where △B g The acquisition is to first obtain the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image in a certain neomycin component, and then calculate the average of the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image of all neomycin components, where max() is the maximum value function; S302, based on the initial temporal relationship matrix F between vectors and the preset temporal relationship threshold △F, generate the target temporal relationship matrix F between vectors 0 ; S303, according to the temporal relationship matrix F between target vectors 0 , generate the classification cluster data U={U1,...,U t ,……,U β It can be seen that by analyzing the temporal relationship between the image features of any two initial neomycin images and then classifying the initial neomycin images based on the temporal relationship, the temporal relationship between the neomycin images can be analyzed, and the data of neomycin changes over time can be established, thereby increasing the data volume of neomycin images. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A flowchart of a method for processing image data of a neomycin spectrum based on a deep learning model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] like Figure 1 As shown, this embodiment provides a method for processing image data of neomycin spectrum based on a deep learning model, the method comprising the following steps: S100, obtain the initial neomycin spectrum data A={A1, ..., A i ,……,A m}, A i is the i-th initial neomycin spectrum, i ranges from 1 to m, and m is the number of initial neomycin spectra.

[0017] Specifically, the initial neomycin spectrum is an initial neomycin mass spectrum image, wherein the neomycin mass spectrum image is obtained by a mass spectrometer. Those skilled in the art are aware of any method for obtaining a neomycin mass spectrum image by a mass spectrometer in the prior art, which will not be repeated here.

[0018] Further, A1 to A m Each initial neomycin mass spectrometry image was inconsistent.

[0019] S200, each A i Input into the target deep learning model to obtain A i The corresponding image feature vector B i ={B i1 ,……,B ij ,……,B in}, B ij It's A i The jth image feature in the image, the value range of j is 1 to n, and n is A i The corresponding number of image features.

[0020] Specifically, step S200 also includes the following steps: S201, obtain neomycin sample image data C={C1, ..., C r ,……,C s}, C r ={C x1 ,……,C rg ,……,C rz}, C rgis the g-th sample image in the r-th neomycin component, where r ranges from 1 to s, s is the total number of neomycin components, and g ranges from 1 to z, where z is the total number of sample images in each neomycin component; further, the sample image is a mass spectrometry image of any neomycin component that changes with time, wherein the sample image collected at the g-th time node and the sample image collected at the g+1-th time node in different neomycin components remain unchanged, which can be understood as: the g-th sample image is the sample image acquired at the g-th acquisition time.

[0021] S202, for each C rg Process and get C rg The corresponding intermediate image C 0 rg , where C 0 rg Including several element points, C 0 rg Each element point (x0, y0) in (x0, y0) meets the following conditions: ; , where kh is the horizontal convolution kernel size, kw is the vertical convolution kernel size, μh is the horizontal Gaussian distribution mean, μw is the vertical Gaussian distribution mean, e is the horizontal offset within the kernel, f is the vertical offset within the kernel, bh is the horizontal position offset, bw is the vertical position offset, vh is the horizontal scaling factor, vw is the vertical scaling factor, σh is the horizontal Gaussian distribution standard deviation, σw is the vertical Gaussian distribution standard deviation, x is the horizontal coordinate point in the sample image, y is the vertical coordinate point in the sample image, x0 is the horizontal coordinate point in the intermediate image, and y0 is the vertical coordinate point in the intermediate image.

[0022] S203, according to C rg For each element point (x0, y0), get C rg The image feature vector D in rg =(D 1 rg ,……,D j rg ,……,D n rg ), D j rg It is C rg The jth image feature in .

[0023] Furthermore, step S203 further includes the following steps: S2031, according to C rg For each element point (x0, y0), select Crg Corresponding key element point data G rg ={(XG 1 rg , YG 1 rg ),……,(XG a rg , YG a rg ),……,(XG b rg , YG b rg )},(XG a rg , YG a rg ) is the ath key element point, XG a rg It is C rg The key coordinate point on the X axis of the a-th point, YG a rg It is C rg The key coordinate point on the Y axis of the a-th point, where each key element point (XG a rg , YG a rg ) is the element point representing the component of neomycin in each element point (x0, y0), the value range of a is 1 to b, b is the number of key element points, which can be understood as: C rg The corresponding key element point is C rg The element points that characterize the components of neomycin, C rg Other element points and C rg The key element points have obvious differences and can be distinguished by features recognized by the image, such as texture features, RGB value features and other image features. Those skilled in the art are aware of any method of extracting features from images in the prior art, which will not be described here.

[0024] S2032, from C rg Filter out several peak points and change degrees, insert each peak point and change degree as image features into the image feature vector of the empty set to obtain D rg , can be understood as: D rg The image features are peak points or degree of change.

[0025] S204, all D rg A preset deep learning model is used as a sample for training to obtain a target deep learning model. Those skilled in the art are familiar with any method for training a deep learning model in the prior art, which will not be described here; for example, the preset deep learning model is a CNN-based model.

[0026] As described above, the image features of the neomycin sample can be optimized using a suitable kernel to generate a target deep learning model, which is beneficial for extracting accurate image features from the initial neomycin spectrum, and then accurately classifying the relationship between the initial neomycin spectra.

[0027] S300, according to each B i , get the classification cluster data of neomycin U={U1,……,U t ,……,U β},U t is the t-th neomycin classification cluster, t ranges from 1 to β, and β is the number of neomycin classification clusters.

[0028] Specifically, step S300 also includes the following steps: S301, according to B i With B except B i The temporal relationship degree F constructed between other image feature vectors other than i ={F i1 ,……,F iu ,……,F iη}, F iu It's B i With B except B i The temporal relationship between the u-th image feature vectors, F iu Meet the following conditions: , B 0 uj It is B except B i The jth image feature in the uth image feature vector among the other image feature vectors except B, where u ranges from 1 to η, and η is the value of B except B. i The total amount of other image feature vectors other than η=n-1, where △B g is the temporal relationship threshold corresponding to the sample images collected at the g-th time node for different neomycin components, and max() is the maximum value function.

[0029] Furthermore, △B g The method is to first obtain the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image in a certain neomycin component, and then calculate the average of the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image of all neomycin components. For example, the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image in the r-th neomycin component meets the following conditions: , D j rgis the jth image feature in the gth sample image of the rth neomycin component, D j rg+1 is the jth image feature in the g+1th sample image in the rth neomycin component.

[0030] S302: Generate a target vector temporal relationship matrix F based on the initial vector temporal relationship matrix F and a preset temporal relationship threshold ΔF. 0 , F meets the following conditions:

[0031] As mentioned above, when F iu ≥△F, at F 0 Keep F iu When F iu <△F, at F 0 Set F iu =0, wherein those skilled in the art can set the timing relationship threshold according to actual needs, which will not be described in detail here.

[0032] S303, according to the temporal relationship matrix F between target vectors 0 , generate the classification cluster data U={U1,...,U t ,……,U β}; can be understood as: F 0 Each initial neomycin spectrum with the corresponding horizontal temporal relationship degree of 0 is regarded as a single neomycin classification cluster, and F 0 All the initial neomycin spectra whose corresponding horizontal temporal relationship degrees are not all 0 are regarded as the classification clusters of single neomycin. For example, when F i1 to F iη When both are 0, the corresponding initial neomycin spectrum is regarded as a single neomycin classification cluster. At the same time, when F m1 to F mη When both are 0, the corresponding initial neomycin spectrum is also a classification cluster of a single neomycin; and when F i1 to F iη The corresponding initial neomycin spectrum and when F m1 to F mη All other initial neomycin spectra before the corresponding initial neomycin spectra are regarded as the classification cluster of a single neomycin.

[0033] The present embodiment provides a method for processing image data of neomycin spectrum based on a deep learning model, the method comprising: S100, obtaining initial neomycin spectrum data A={A1, ..., A i ,……,A m}, A iis the i-th initial neomycin spectrum, i ranges from 1 to m, and m is the number of initial neomycin spectra; S200, each A i Input into the target deep learning model to obtain A i The corresponding image feature vector B i ={B i1 ,……,B ij ,……,B in}, B ij It's A i The jth image feature in the image, the value range of j is 1 to n, and n is A i The number of corresponding image features; S300, according to each B i , get the classification cluster data of neomycin U={U1,……,U t ,……,U β},U t is the t-th neomycin classification cluster, the value range of t is 1 to β, β is the number of neomycin classification clusters; wherein, the step S300 also includes the following steps: S301, according to B i With B except B i The temporal relationship degree F constructed between other image feature vectors other than i ={F i1 ,……,F iu ,……,F iη}, F iu It's B i With B except B i S302, based on the initial inter-vector temporal relationship matrix F and the preset temporal relationship threshold △F, generate the target inter-vector temporal relationship matrix F 0 ; S303, according to the temporal relationship matrix F between target vectors 0 , generate the classification cluster data U={U1,...,U t ,……,U β It can be seen that by analyzing the temporal relationship between the image features of any two initial neomycin images and then classifying the initial neomycin images based on the temporal relationship, the temporal relationship between the neomycin images can be analyzed, and the data of neomycin changes over time can be established, thereby increasing the data volume of neomycin images.

[0034] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A method for processing image data of neomycin spectra based on a deep learning model, characterized in that: The method comprises the following steps: S100, obtain the initial neomycin spectrum data A={A1, ..., A i ,……,A m }, A i is the i-th initial neomycin spectrum, i ranges from 1 to m, and m is the number of initial neomycin spectra; S200, each A i Input into the target deep learning model to obtain A i The corresponding image feature vector B i ={B i1 ,……,B ij ,……,B in }, B ij It's A i The jth image feature in the image, the value range of j is 1 to n, and n is A i The number of corresponding image features; S300, according to each B i , get the classification cluster data of neomycin U={U1,……,U t ,……,U β }, U t is the t-th neomycin classification cluster, the value of t ranges from 1 to β, and β is the number of neomycin classification clusters; wherein, step S300 also includes the following steps: S301, obtain B i With B except B i The temporal relationship degree F constructed between other image feature vectors other than i ={F i1 ,……,F iu ,……,F iη }, F iu It's B i With B except B i The temporal relationship between the u-th image feature vectors other than iu Meet the following conditions: , B 0 uj It is B except B i The jth image feature in the uth image feature vector among the other image feature vectors except B, where u ranges from 1 to η, and η is the value of B except B. i The total amount of other image feature vectors other than g is the temporal relationship threshold corresponding to the sample images collected at the g-th time node of different neomycin components. The value range of g is 1 to z, z is the number of sample images in each neomycin component, and max() is the maximum value function; S302: Generate a target vector temporal relationship matrix F based on the initial vector temporal relationship matrix F and a preset temporal relationship threshold ΔF. 0 , F 0 Meet the following conditions: ； S303, according to the temporal relationship matrix F between target vectors 0 , generate the classification cluster data U={U1,...,U t ,……,U β }.

2. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 1, characterized in that: The initial neomycin spectrum is an initial neomycin mass spectrum image.

3. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 2, characterized in that: Neomycin mass spectrum images were acquired by mass spectrometer.

4. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 2, characterized in that: A1 to A m Each initial neomycin mass spectrometry image was inconsistent.

5. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 1, characterized in that: η=n-1.

6. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 1, characterized in that: △B g The acquisition process includes the following: Obtaining a temporal relationship between an image feature vector of the gth sample image and an image feature vector of the g+1th sample image in the rth neomycin component, where r ranges from 1 to s, and s is the total number of neomycin components, wherein the temporal relationship between the sample image collected at the gth time node and the sample image collected at the g+1th time node in different neomycin components remains unchanged; Obtain the average value of the temporal relationship between the image feature vector of the g-th sample image and the image feature vector of the g+1-th sample image of all neomycin components.

7. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 6, characterized in that: The temporal relationship between the image feature vector of the gth sample image and the image feature vector of the g+1th sample image in the rth neomycin component meets the following conditions: , D j rg is the jth image feature in the gth sample image of the rth neomycin component, D j rg+1 is the jth image feature in the g+1th sample image in the rth neomycin component.

8. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 1, characterized in that: The step S303 includes: 0 Each initial neomycin spectrum with the corresponding horizontal temporal relationship degree of 0 is regarded as a single neomycin classification cluster, and F 0 All initial neomycin spectra whose corresponding horizontal temporal relationship degrees are not all 0 are regarded as classification clusters of single neomycin.

9. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 8, characterized in that: In step S302, when F iu ≥△F, at F 0 Keep F iu .

10. The method for processing image data of neomycin spectrum based on a deep learning model according to claim 9, characterized in that: In step S302, when F iu <△F, at F 0 Set F iu =0.

Images