Cell culture monitoring method and system based on infrared-raman image recovery model

Abstract

The present application relates to the technical field of bioengineering, in particular to a cell culture monitoring method and system based on infrared-Raman image recovery model, the method comprising: 1, building a metabolic activity comprehensive score model, and building a metabolic active region set using the metabolic activity comprehensive score model; 2, selecting the first N metabolic active regions in the metabolic active region set as selected regions, dividing the selected regions into low change regions and key regions, and solving compressed measurement Raman images according to the key regions; 3, constructing a data set; 4, training a pre-built infrared-Raman image recovery model using the data set, and transmitting the trained infrared-Raman image recovery model to a device end; 5, monitoring cell culture using the infrared-Raman image recovery model on the device end. The present application solves the problems of insufficient global information capture and low visual and language information fusion efficiency in traditional single-mode classification methods, and achieves higher classification accuracy and task generalization ability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of bioengineering technology, in particular to a cell culture monitoring method and system based on infrared-Raman image recovery model. BACKGROUND

[0002] The rapid development of the biopharmaceutical industry puts forward higher requirements for real-time monitoring and precise control of cell culture process. In the culture process of cell therapy products such as monoclonal antibodies and CAR-T, real-time detection of cell metabolic state and product quality is crucial to ensure product quality and improve production efficiency. Traditional cell culture monitoring mainly relies on offline sampling analysis, which not only has a large workload, but also has a lag, making it difficult to meet the needs of biopharmaceutical process control.

[0003] Raman spectroscopy technology has become an effective means for cell culture monitoring due to its high sensitivity and specificity to cell metabolites. However, existing Raman spectral imaging has the following technical difficulties: first, the imaging speed is slow, it usually takes several hours to complete a full-field Raman scan, making it difficult to capture the dynamic changes of cell metabolism; second, the sampling efficiency is low, traditional uniform sampling strategies do not consider the uneven distribution of information in the sample, resulting in a large amount of resources wasted in low-information areas; third, it is difficult to align infrared and Raman data, the collaborative application of these two complementary spectral technologies is limited by technical bottlenecks.

[0004] Therefore, there is an urgent need for an efficient and accurate cell culture monitoring method to realize online monitoring of cell culture process and provide scientific support for real-time monitoring and precise control technology of biopharmaceutical processes. SUMMARY

[0005] The present application provides a cell culture monitoring method and system based on infrared-Raman image recovery model to solve the technical problems mentioned in the background art.

[0006] To achieve the above-mentioned purposes, the technical solution of the present application is as follows:

[0007] The present application provides a cell culture monitoring method based on infrared-Raman image recovery model, comprising the following steps:

[0008] S1, multiple infrared images of cell metabolic activity are taken, then the difference images of infrared images at consecutive time points are calculated by infrared image analysis, and a metabolic activity comprehensive scoring model is built based on the difference images, and a metabolic active region set is built using the metabolic activity comprehensive scoring model ;

[0009] S2, selecting a metabolic active region set before the metabolic active region set The selected region is divided into a low change region and a key region, and then a compressed measurement Raman image is solved according to the key region;

[0010] S3, a range of the compressed measurement Raman image is generated by combining a linear interpolation algorithm, and an infrared spectrum image of a corresponding region is obtained according to the range of the compressed measurement Raman image; then a complete Raman image is acquired by collecting the corresponding region, and a data set is constructed by using the complete Raman image, the adjusted infrared spectrum image and the compressed measurement Raman image;

[0011] S4, the pre-built infrared-Raman image recovery model is trained by using the data set, and a trained infrared-Raman image recovery model is obtained, and the trained infrared-Raman image recovery model is transmitted to the device end;

[0012] S5, the cell culture is monitored by using the infrared-Raman image recovery model on the device end.

[0013] The application also provides a cell culture monitoring system based on an infrared-Raman image recovery model, comprising a device end configured to execute the cell culture monitoring method.

[0014] The application has the following beneficial effects:

[0015] 1. The application provides a cell culture monitoring method based on an infrared-Raman image recovery model, which identifies a metabolically active region by infrared pre-scanning, combines a priority sampling strategy, and concentrates limited Raman sampling resources in the most information valuable region, so that the imaging time is shortened by more than 80%, the Raman spectrum imaging efficiency is greatly improved, and the real-time monitoring of the cell culture process is realized.

[0016] In addition, the application breaks through the technical bottleneck of traditional Raman spectrum imaging and provides an innovative solution for cell culture process monitoring, which is of great significance for promoting the development of the biopharmaceutical industry.

[0017] 2. The cell culture monitoring method based on the infrared-Raman image recovery model provided by the application uses an infrared-Raman image recovery model in the internal design, fully utilizes the complementary characteristics of infrared and Raman images, realizes the effective fusion of multi-modal information through a self-attention mechanism, greatly improves the signal-to-noise ratio of the reconstructed image, and ensures the high quality of the reconstructed Raman image.

[0018] 3. The infrared-Raman image recovery model provided by the application solves the problem of spatial alignment of two kinds of spectral data, realizes the accurate alignment of Raman spectrum and infrared spectrum, and provides a reliable foundation for multi-modal spectrum analysis.

[0019] 4. The present invention discloses a cell culture monitoring system based on an infrared-Raman image restoration model. The cell culture monitoring system can automatically detect and track metabolically active areas without manual intervention, greatly reducing operational complexity and human errors, while also significantly improving the automation level of cell culture monitoring.

[0020] 5. The cell culture monitoring system of the present invention can monitor the cell metabolism status and product quality in real time, detect abnormalities in a timely manner, provide a scientific basis for culture parameter optimization and product quality control, and also provide a reliable quality control method for the biopharmaceutical process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a framework diagram of the infrared-Raman image restoration model of the present invention. DETAILED DESCRIPTION

[0022] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The accompanying drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many other forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present disclosure.

[0023] Reference Figure 1 The present invention provides a cell culture monitoring method based on an infrared-Raman image restoration model, comprising the following steps:

[0024] S1. Take multiple infrared images of cell metabolic activities, then use infrared image analysis to calculate the difference map of infrared images at consecutive time points, and build a metabolic activity comprehensive scoring model based on the difference map. Use the metabolic activity comprehensive scoring model to build a metabolic jump region set. ;

[0025] S2. Gather in metabolically active areas Select metabolically active The selected area is divided into a low-variation area and a key area, and then the compressed measurement Raman image is solved according to the key area;

[0026] S3. Generate the range of the compressed measurement Raman image using a linear interpolation algorithm, and acquire an infrared spectrum image of the corresponding area based on the range of the compressed measurement Raman image; then acquire a complete Raman image of the corresponding area, and construct a data set using the complete Raman image, the adjusted infrared spectrum image, and the compressed measurement Raman image;

[0027] S4, train the pre-built infrared-Raman image recovery model using the data set, obtain the trained infrared-Raman image recovery model, and transmit the trained infrared-Raman image recovery model to the device end;

[0028] S5, monitor the cell culture using the infrared-Raman image recovery model on the device end.

[0029] The cell culture monitoring method in the application identifies the metabolic active area through infrared pre-scanning, combines the priority sampling strategy, concentrates the limited Raman sampling resources in the area with the most information value, shortens the imaging time by more than 80%, greatly improves the Raman spectrum imaging efficiency, and realizes real-time monitoring of the cell culture process.

[0030] In addition, the application breaks through the technical bottleneck of traditional Raman spectrum imaging and provides an innovative solution for cell culture process monitoring, which is of great significance for promoting the development of the biological pharmaceutical industry.

[0031] In some embodiments, S1 specifically comprises the following steps:

[0032] S11, multiple infrared images of cell metabolic activity are photographed; based on the heat change and pH change generated by cell metabolic activity, the changes will be shown as temperature gradient and absorption intensity change of specific waveband in the infrared image;

[0033] S12, the temperature gradient map in each infrared image is extracted through the radiation-to-temperature conversion algorithm built in the photographing system , and the area where the local temperature is higher than the environment is identified in the temperature gradient map , to obtain the local physical characteristics, which are used as physical characteristics for subsequent metabolic activity score of each pixel point in the infrared image;

[0034] S13, the characteristics of the set metabolic product (such as lactic acid) are identified in the infrared image to obtain the local biological characteristics, and then the infrared absorption intensity map of the local biological characteristic absorption waveband (such as area) is obtained, which is used as biological characteristics for subsequent metabolic activity score of each pixel point in the infrared image;

[0035] S14, and based on the time series change rate analysis, the difference map of the adjacent time point infrared image is calculated;

[0036] S15, the first threshold is used to segment the difference map, and the significant area with a change degree reaching a set degree is extracted , the significant area The significantly changed region, which usually represents high metabolic activity, can be used for further calculation of the metabolic activity comprehensive score; then the time series change rate curve of the significantly changed region C is calculated;

[0037] S16、In the extracted significant region , the temperature gradient map including local physical features , the infrared absorption intensity map of the local biological feature absorption band, and the time series change rate curve are combined to construct a metabolic activity comprehensive score model, which can quantify the metabolic activity degree of each pixel point; the metabolic activity score of each pixel point in the infrared image is calculated using the metabolic activity comprehensive score model, and a metabolic activity score map is constructed accordingly;

[0038] S17, the metabolic activity score map is position segmented using a second threshold, which is preferably 20%; that is, the top 20% of the metabolic activity score is taken as the region of interest, and the metabolic activity score map is position segmented using the region of interest;

[0039] Morphological operations (such as opening and closing operations) are used to denoise and connect the broken regions of the segmentation result, then the obtained regions are counted, and a metabolic activity region set is constructed based on the counting result , , where n represents the total number of obtained regions.

[0040] In some embodiments, the calculation formula of the difference map in S14 is as follows:

[0041] ;

[0042] , wherein, represents the difference map of the adjacent time point infrared image; represents the average value of all wavebands of the infrared image at the moment; represents the average value of all wavebands of the infrared image at the moment; , , and , respectively represent the position of the infrared image on the x-axis and y-axis.

[0043] In some embodiments, the expression of the significantly changed region in S15 is as follows:

[0044] ;

[0045] , wherein, represents the first threshold.

[0046] In some embodiments, the expression of the metabolic activity comprehensive score model in S16 is specifically as follows:

[0047] ;

[0048] wherein, represents the metabolic activity score; , , respectively represent the weight coefficients of the temperature gradient map , the infrared absorption intensity map of the local biological feature absorption waveband, and the difference map of the adjacent time point infrared image; represents the infrared absorption intensity map of the local biological feature absorption waveband.

[0049] In some embodiments, S2 specifically comprises the following steps:

[0050] S21, selecting a metabolic active region before the metabolic active region set as a selected region; preferably, the value of k is 5;

[0051] S22, for each selected region, calculating the mean and variance of all pixel points in the selected region in the selected region, defining the selected region with a variance less than the mean by a first set percentage as a low change region; defining the selected region with a variance greater than or equal to the mean by a first set percentage as a key scanning region , wherein m represents the number of key scanning regions; the first set percentage is preferably 10%;

[0052] Then randomly selecting pixel point positions in each low change region for single point scanning, and calculating the mean of the Raman signals of single point scanning to represent the characteristic curve (i.e. Raman spectrum curve) of the low change region, which is used for subsequent analysis of the material composition of the low change region; preferably, the value of n is 10; the low change region does not need to be scanned and imaged in large quantities, and a few randomly sampled points can represent the characteristics in the region. Then this characteristic can be analyzed through the analysis of the Raman spectrum curve in chemical and physical measurement to determine what substances are in it and what the corresponding composition is.

[0053] For each key region , a current key region The maximum outer rectangle of the first set of pixels is obtained, and the maximum outer rectangle is taken as a region to be finely imaged, then a global sampling scan is performed in the imaging device, and in the process of the global sampling scan, a sampling matrix is generated by using a random function, and a second set percentage of pixel points in the sampling matrix are randomly sampled as a compressed measurement Raman image , the compressed measurement Raman image is used for subsequent reconstruction, and the scanning time of Raman in the key region is greatly shortened. The second set percentage is preferably 20%.

[0054] In some embodiments, the S3 specifically comprises the following steps:

[0055] S31, for each key region , the infrared spectrum image collected by the infrared spectrometer corresponding to each region is obtained respectively ; the infrared spectrum image has dimensions of length H, width W and band number B respectively;

[0056] S32, the band of the infrared spectrum image is mapped to the same size as the band of the compressed measurement Raman image by using a linear interpolation algorithm ; an adjusted infrared spectrum image is obtained, and the adjusted infrared spectrum image has dimensions of length H, width W and band number respectively;

[0057] The step S32 in the application solves the problem of alignment of two kinds of spectral data spaces, realizes accurate alignment of Raman spectrum and infrared spectrum, and provides a reliable foundation for multi-modal spectral analysis.

[0058] A complete Raman image in each region is collected , and the complete Raman image has dimensions of length H, width W and band number ; a data set is constructed by using the complete Raman image , the corresponding adjusted infrared spectrum image and the corresponding compressed measurement Raman image .

[0059] In some embodiments, the infrared-Raman image recovery model comprises a near-infrared image encoder, a Raman image encoder, an image interaction attention model and a reconstruction decoder.

[0060] The near-infrared image encoder and the Raman image encoder are connected in parallel, and the output ends of the two are connected with the image interaction attention model and the reconstruction decoder in sequence.

[0061] The infrared-Raman image restoration model in the application fully utilizes the complementary characteristics of infrared and Raman images, realizes effective fusion of multi-modal information through a self-attention mechanism, and makes the signal-to-noise ratio of the reconstructed image increased by more than 40%, thereby ensuring the high quality of the reconstructed Raman image.

[0062] In some embodiments, the S4 specifically comprises the following steps:

[0063] S41, selecting complete Raman images from a data set , corresponding adjusted infrared spectral images , and corresponding compressed measured Raman images are input into a pre-built infrared-Raman image restoration model;

[0064] S42, the adjusted infrared spectral images , the compressed measured Raman images are respectively input into a near-infrared image encoder and a Raman image encoder to obtain near-infrared features and Raman features , which are expressed by the following formula:

[0065] ;

[0066] ;

[0067] wherein, represents a feature extraction operation of the near-infrared image encoder; represents a feature extraction operation of the Raman image encoder; the dimensions of the near-infrared features and the Raman features are all length , width , and band number ;

[0068] S43, the order of the spatial dimensions (i.e., length and width) and the band number in the near-infrared features and the Raman features are adjusted respectively, and then expanded in the band number dimension to obtain feature-vectored Raman signals and infrared spectral signals ; at this time, the dimensions of the feature-vectored Raman signals and the infrared spectral signals are ; the spatial dimension refers to the length and the width in the dimension;

[0069] S44, generating token vectors and sequence vectors using the feature-vectored infrared spectral signals, and generating value vectors using the feature-vectored Raman signals, which are expressed by the following formula:

[0070] ;

[0071] ;

[0072] ;

[0073] wherein, , , denote the weights of the token vector , the sequence vector and the value vector ;

[0074] S45, then calculate the attention weight according to the token vector and the sequence vector and apply it to the value vector to obtain the output feature , which is expressed by the formula as follows:

[0075] ;

[0076] ;

[0077] wherein, denotes the attention weight; denotes the value under the square root of the numerical size, which is used to scale the dot product to prevent gradient disappearance; denotes the length of the last dimension of the token vector , normalization exponential function; at this time, the dimension of the output feature is ;

[0078] S46, the spatial dimension of the output feature and the number of bands are reconverted, so that the output feature is converted from a vector feature to an image feature, to obtain the converted output feature , the dimension of the converted output feature is ; and the converted output feature is input into the reconstruction decoder, which maps the feature back to the original dimension (H, W, B) to obtain the reconstructed Raman image ;

[0079] S47, the total loss function is constructed by using the reconstructed Raman image and the corresponding complete Raman image , and the total loss function is as follows:

[0080] ;

[0081] wherein, denotes the total loss value, denotes the number of a batch in the training process; denotes the Fourier transform; denotes the square of the L2 norm, denotes the L1 norm, and denotes the loss weight distribution coefficient in the frequency and spatial domain; denotes the i-th complete Raman image; denotes the i-th reconstructed Raman image ;

[0082] S48, circulate S41 to S47, minimize the total loss function until the total loss function converges, and update the weight of the infrared-Raman image recovery model to obtain the trained infrared-Raman image recovery model.

[0083] Another aspect of the present application also provides a cell culture monitoring system based on an infrared-Raman image recovery model, comprising a device end configured to perform the above-mentioned cell culture monitoring method.

[0084] The cell culture monitoring system can automatically detect and track the metabolically active area without manual intervention, greatly reducing the operation complexity and human error, and significantly improving the automation level of cell culture monitoring.

[0085] The cell culture monitoring system in the present application can monitor the cell metabolic state and product quality in real time, discover abnormalities in time, and provide a scientific basis for culture parameter optimization and product quality control, and also provide a reliable quality control means for biopharmaceutical processes.

[0086] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Moreover, the technical solutions of each embodiment of the present application can be combined with each other, but it must be based on the realization of the ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the protection scope required by the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A cell culture monitoring method based on infrared-Raman image restoration model, characterized in that: The steps include: S1. Take multiple infrared images of cell metabolic activities, then use infrared image analysis to calculate the difference map of infrared images at consecutive time points, and build a metabolic activity comprehensive scoring model based on the difference map. Use the metabolic activity comprehensive scoring model to build a metabolic jump region set. ; S2. Gather in metabolically active areas Select metabolically active The selected area is divided into a low-variation area and a key area, and then the compressed measurement Raman image is solved according to the key area; S3. Generate a range of a compressed measurement Raman image using a linear interpolation algorithm, and acquire an infrared spectrum image of a corresponding area based on the range of the compressed measurement Raman image; Then, the corresponding area is collected to obtain a complete Raman image, and a data set is constructed using the complete Raman image, the adjusted infrared spectrum image, and the compressed measurement Raman image; S4. Using the data set to train the pre-built infrared-Raman image restoration model to obtain a trained infrared-Raman image restoration model, and transmitting the trained infrared-Raman image restoration model to the device end; S5. Monitor cell culture using the infrared-Raman image restoration model on the device side.

2. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 1, characterized in that: The S1 specifically includes the following steps: S11. Take multiple infrared images of cell metabolic activities; S12. Extract the temperature gradient map in each infrared image through the built-in radiation-to-temperature conversion algorithm of the shooting system , and in the temperature gradient diagram Identify the areas where the local temperature is higher than the environment and obtain the local physical characteristics; S13. In the infrared image, for the absorption band of the local biological characteristics, calculate and obtain an infrared absorption intensity map of the absorption band of the local biological characteristics; S14, and calculating a difference image of the infrared images at adjacent time points based on the time series change rate analysis; S15: Segment the difference image using the first threshold value and extract the significant areas where the degree of change reaches the set degree. , and calculate the significant change area Time series change rate curve; S16. In the extracted salient area Incorporating temperature gradient maps including local physical characteristics , constructing a metabolic activity comprehensive scoring model based on the infrared absorption intensity graph of the local biological characteristic absorption band and the time series change rate curve, using the metabolic activity comprehensive scoring model to calculate the metabolic activity score of each pixel in the infrared image, and constructing a metabolic activity score graph based on this; S17, using the second threshold to segment the metabolic activity score map, using morphological operations to denoise the segmentation results and connect the broken areas, then statistically analyze the obtained areas, and construct a metabolic jump area set based on the statistical results. , , where n represents the total number of obtained regions.

3. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 2, characterized in that: The calculation formula of the difference map in S14 is as follows: ； in, A difference image representing infrared images at adjacent time points; Indicates the The average value of all bands of the infrared image at the moment; Indicates the The infrared image at each moment is targeted at the absorption band of local biological features. 、 Represent the position of the infrared image on the x-axis and y-axis respectively.

4. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 3, characterized in that: The significant changes in S15 The expression is as follows: ； in, Indicates the first threshold.

5. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 4, characterized in that: The expression of the metabolic activity comprehensive scoring model in S16 is as follows: ； in, represents the metabolic activity score; 、 、 Represents the temperature gradient diagram , the weight coefficient of the infrared absorption intensity map of the local biological characteristic absorption band and the difference map of the infrared images at adjacent time points; Infrared absorption intensity map showing local biogenic characteristic absorption bands.

6. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 5, characterized in that: The S2 specifically includes the following steps: S21. Gather in metabolically active areas Select metabolically active areas as selected areas; S22. For each selected area, calculate the mean and variance of all pixels in the selected area, define the selected area with a variance less than a first set percentage of the mean as a low-variance area, and define the selected area with a variance greater than or equal to the first set percentage of the mean as a key scanning area. , , where m represents the number of key scanning areas; S23. For each key area , establish current key areas The maximum circumscribed rectangle is used as the area to be finely imaged, and then a global sampling scan is performed in the imaging device. During the global sampling scan, a sampling matrix is ​​generated using a random function, and a second set percentage of pixels are randomly sampled within the sampling matrix as a compressed measurement Raman image. .

7. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 6, characterized in that: The S3 specifically includes the following steps: S31. For each key area Positioning will be performed to obtain the infrared spectrum images collected by the infrared spectrometer corresponding to each area. ; Infrared spectral image The dimensions are length H, width W, and number of bands B; S32, use linear interpolation algorithm to convert infrared spectrum image Band mapping and compression of measured Raman images The bands are the same size ; Get the adjusted infrared spectrum image , adjusted infrared spectrum image The dimensions are length H, width W, number of bands ; S33, collect complete Raman images in each area , complete Raman image The dimensions are length H, width W, and the number of bands is ; Using the complete Raman image , the corresponding adjusted infrared spectrum image X HK And the corresponding compressed measurement Raman image Build a dataset.

8. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 7, characterized in that: The infrared-Raman image restoration model includes a near-infrared image encoder, a Raman image encoder, an image interaction attention model and a reconstruction decoder; Among them, the near-infrared image encoder and the Raman image encoder are connected in parallel, and the output ends of both are connected to the image interaction attention model and the reconstruction decoder in turn.

9. The cell culture monitoring method based on the infrared-Raman image restoration model according to claim 8, characterized in that: The S4 specifically includes the following steps: S41. Select complete Raman image from the data set , the corresponding adjusted infrared spectrum image And the corresponding compressed measurement Raman image Input into a pre-built infrared-Raman image restoration model; S42, the adjusted infrared spectrum image , compressed measurement Raman image Input them into the near infrared image encoder and Raman image encoder respectively to obtain the near infrared features and Raman characteristics , expressed by the formula, as follows: ； ； in, represents the feature extraction operation of the near infrared image encoder; represents the feature extraction operation of the Raman image encoder; S43, adjust the near infrared characteristics separately and Raman characteristics The order of the spatial dimension and the number of bands is then expanded in the dimension of the number of bands to obtain the eigenvectorized Raman signal and infrared spectral signals ;Spatial dimensions refer to the length and width of the dimensions; S44. Generate token vector using feature vectorized infrared spectrum signal and sequence vector , using the eigenvectorized Raman signal to generate a value vector , expressed by the formula, as follows: ； ； ； in, 、 、 Represents token vectors , sequence vector and the value vector The weight of S45, then based on the token vector and sequence vector Calculate attention weights and apply to the value vector , and get the output features , expressed by the formula, as follows: ； ； in, represents the attention weight; Indicates that according to the square root The numerical size of is used to scale the dot product to prevent the gradient from disappearing; Represents a token vector The length of the last dimension, Normalized exponential function; S46. Output features The spatial dimension and number of bands are reconverted to make the output features Convert vector features into image features and obtain the converted output features , and the converted output features Input to the reconstruction decoder, the reconstruction decoder remaps the features back to the original dimension to obtain the reconstructed Raman image ; S47. Using the reconstructed Raman image and the corresponding complete Raman image Construct the total loss function, which is as follows: ； in, represents the total loss value, Indicates the number of batches in the training process; represents Fourier transform; represents the square of the L2 norm, represents the L1 norm, and Represents the loss weight distribution coefficient in frequency and spatial domains; Indicates the A complete Raman image; Reconstruction Zhang Raman image ; S48. Loop S41 to S47 to minimize the total loss function until the total loss function converges, and update the weights of the infrared-Raman image restoration model to obtain the trained infrared-Raman image restoration model.

10. A cell culture monitoring system based on infrared-Raman image restoration model, characterized in that: The device comprises a device end configured to execute the cell culture monitoring method according to any one of claims 1 to 9.

Images