Microscopy measurement data processing method and system

Abstract

The application provides a microscopic measurement data processing method and system. In the scheme, the microscopic measurement data processing system comprises a microscopic measurement device, a microscopic data processing platform and a microscopic measurement database. The microscopic data processing platform first receives microscopic measurement data collected by the microscopic measurement device, performs normalization processing on the microscopic measurement data based on an adaptive specification method to obtain first measurement microscopic data, then classifies noise types and noise quantities in the first measurement microscopic data by using a noise classification model to obtain at least one noise type and a noise quantity corresponding to each noise type, then removes the noise in the first measurement microscopic data by using a noise removal model to obtain second measurement microscopic data after noise removal, and stores the second measurement microscopic data in the microscopic measurement database. The microscopic measurement data processing method and system provided by the application can effectively improve the quality of the microscopic measurement data and provide protection for the generation of subsequent high-quality microscopic images.

Description

Technical Field

[0001] This application relates to digital processing technology, and more particularly to a method and system for processing microscopic measurement data. Background Technology

[0002] Driven by emerging technologies, the manufacturing and development of microstructures such as basic components place high demands on the precision and reliability of microstructure surface measurements. Confocal microscopy plays a crucial role in precision measurement and testing. However, confocal microscopes, as precision testing instruments, suffer from low imaging resolution.

[0003] To address these issues, current technical solutions primarily focus on expanding the imaging range and improving the resolution of confocal microscopes. Both expanding the imaging range and improving the resolution involve modifying the hardware components of the confocal microscope to achieve higher imaging resolution.

[0004] However, while the above-mentioned technical solutions have achieved improved imaging resolution, they have come at the cost of reduced performance in other key technical indicators. Therefore, how to improve imaging resolution without compromising other key technical indicators has become a pressing issue. Summary of the Invention

[0005] This application provides a method and system for processing microscopic measurement data to solve the technical problem of low imaging resolution at present.

[0006] In a first aspect, this application provides a method for processing microscopic measurement data, applied to a microscopic measurement data processing system, the microscopic measurement data processing system including a microscopic measurement device, a microscopic data processing platform, and a microscopic measurement database, the method comprising:

[0007] The microscopic measuring device acquires microscopic measurement data and uploads the microscopic measurement data to the microscopic data processing platform;

[0008] The microscopic data processing platform normalizes the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data.

[0009] The microscopic data processing platform uses a noise classification model in a pre-trained noise processing large model to classify the noise type and noise quantity in the first microscopic measurement data, and obtain at least one noise type and the noise quantity corresponding to each noise type.

[0010] The microscopic data processing platform removes noise from the first microscopic measurement data using the noise removal model in the large noise processing model based on the at least one noise type and the noise quantity corresponding to each noise type, thereby obtaining the denoised second microscopic measurement data.

[0011] The second microscopic measurement data is stored in the microscopic measurement database.

[0012] In one possible design of the first aspect, the noise removal model removes noise from the first microscopic measurement data to obtain denoised second microscopic measurement data, including:

[0013] The denoising processing order for the at least one noise type is determined based on the number of noises corresponding to each noise type.

[0014] According to the denoising processing order, each noise type in the first microscopic measurement data is denoised using a variety of denoising algorithms corresponding to the pre-set noise type. The best result is selected from the results of different denoising algorithms as the input of the denoising algorithm for the next noise type, until all noise of at least one noise type in the first microscopic measurement data is denoised, and the second microscopic measurement data is obtained.

[0015] In one possible design of the first aspect, selecting the optimal processing result from the results corresponding to different denoising algorithms includes:

[0016] Calculate the peak signal-to-noise ratio and structural similarity of the first microscopic measurement data with the results corresponding to each denoising algorithm;

[0017] Based on peak signal-to-noise ratio and structural similarity, the processing result with the best effect is selected from the results corresponding to different denoising algorithms.

[0018] In one possible design of the first aspect, the noise classification model is a neural network model for classifying data corresponding to microscopic images, which is obtained by pre-training an initial model constructed based on a convolutional neural network model.

[0019] The initial model includes a first microscopic data classification sub-model, a second microscopic data classification sub-model, a third microscopic data classification sub-model, and an output unit. The output unit is used to calculate the average value of the classification results of the first microscopic data classification sub-model, the second microscopic data classification sub-model, and the third microscopic data classification sub-model and then output it.

[0020] The first microscopic data classification sub-model includes one input layer, three output layers, and 20 hidden layers. The 20 hidden layers include 10 convolutional layers, 6 pooling layers, 3 fully connected layers, and one activation layer.

[0021] The second microscopic data classification sub-model includes one input layer, three output layers, and 30 hidden layers. The 30 hidden layers include 17 convolutional layers, 7 pooling layers, 5 fully connected layers, and one activation layer.

[0022] The third microscopic data classification sub-model includes one input layer, three output layers, and 25 hidden layers. The 25 hidden layers include 12 convolutional layers, eight pooling layers, four fully connected layers, and one activation layer.

[0023] In one possible design of the first aspect, the number of pixels in the microscopic image corresponding to the microscopic measurement data is M*N*3, where M represents the number of row pixels, N represents the number of column pixels, and 3 represents the number of channels;

[0024] Accordingly, the kernel size of the convolutional layer in the first microscopic data classification sub-model is a*a*3, where the first a represents the length of the kernel, the second a represents the width of the kernel, 3 represents the number of channels, and a takes any integer value between 1 and n, where n is the minimum value of M and N.

[0025] In one possible design of the first aspect, the size of the convolution kernel of the convolutional layer in the second microscopic data classification sub-model is b*1*3, where b represents the length of the convolution kernel, the width of the convolution kernel is 1, the number of channels of the convolution kernel is 3, and the value of b is any integer between 1 and M.

[0026] In one possible design of the first aspect, the size of the convolution kernel of the convolutional layer in the third microscopic data classification sub-model is 1*c*3, the length of the convolution kernel is 1, c represents the width of the convolution kernel, the number of channels of the convolution kernel is 3, and the value of c is any integer between 1 and N.

[0027] In one possible design of the first aspect, in the first microscopic data classification sub-model, the size of the convolutional kernel in the convolutional layer decreases sequentially from input to output;

[0028] In the second microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output.

[0029] In the third microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output.

[0030] In one possible design of the first aspect, the at least one noise type includes at least one of the following:

[0031] Device noise, thermal noise, electronic noise, scattering noise, fluorescent dye noise, frosted glass noise, and analog-to-digital conversion noise.

[0032] Secondly, this application relates to a microscopic measurement data processing system, comprising: a microscopic measurement device, a microscopic data processing platform, and a microscopic measurement database, wherein...

[0033] The microscopic measuring device is used to: acquire microscopic measurement data and upload the microscopic measurement data to the microscopic data processing platform;

[0034] The microscopic data processing platform is used for:

[0035] The microscopic measurement data are normalized using an adaptive normalization method to obtain the first microscopic measurement data.

[0036] The noise classification model in the pre-trained noise processing large model is used to classify the noise type and noise quantity in the first microscopic measurement data, so as to obtain at least one noise type and the noise quantity corresponding to each noise type.

[0037] Based on the at least one noise type and the number of noises corresponding to each noise type, the noise removal model in the large noise processing model is used to remove the noise in the first microscopic measurement data to obtain the denoised second microscopic measurement data.

[0038] The microscopic measurement database is used to store the second microscopic measurement data.

[0039] Thirdly, this application provides a microscopic measurement data processing apparatus, comprising:

[0040] The acquisition module is used by the microscopic measuring device to collect and acquire microscopic measurement data and upload the microscopic measurement data to the microscopic data processing platform;

[0041] The processing module is used by the microscopic data processing platform to normalize the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data.

[0042] The processing module is also used by the microscopic data processing platform to classify the noise type and noise quantity in the first microscopic measurement data using a noise classification model in a pre-trained noise processing large model, so as to obtain at least one noise type and the noise quantity corresponding to each noise type.

[0043] The processing module is further configured to use the noise removal model in the noise processing large model to remove noise from the first microscopic measurement data according to at least one noise type and the noise quantity corresponding to each noise type, so as to obtain the noise-removed second microscopic measurement data.

[0044] The storage module is used to store the second microscopic measurement data into the microscopic measurement database.

[0045] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the microscopic measurement data processing method as described in any of the first aspects.

[0046] Fifthly, this application provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, is used to implement the microscopic measurement data processing method described in any one of the first aspects.

[0047] The microscopic measurement data processing method and system provided in this application can be used in the field of data processing technology. In this scheme, the microscopic measurement data processing system includes a microscopic measurement device, a microscopic data processing platform, and a microscopic measurement database. The noise classification model in the microscopic data processing platform takes the microscopic measurement data collected by the microscopic measurement device as input and outputs the noise type and its corresponding quantity, providing a prerequisite for subsequent accurate removal of different types of noise. The noise removal model in the microscopic processing platform accurately removes noise from the microscopic measurement data based on the noise type and quantity, and stores the processed data in the microscopic measurement database, providing a reliable guarantee for generating high-quality microscopic images. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0049] Figure 1 A schematic diagram illustrating the application scenario of the microscopic measurement data processing method provided in this application;

[0050] Figure 2 This is a flowchart illustrating an embodiment of the microscopic measurement data processing method provided in this application.

[0051] Figure 3 This is a flowchart illustrating Embodiment 2 of the microscopic measurement data processing method provided in this application;

[0052] Figure 4 This application provides a schematic diagram of a noise removal model structure.

[0053] Figure 5 A flowchart illustrating Embodiment 3 of the microscopic measurement data processing method provided in this application;

[0054] Figure 6 This application provides a schematic diagram of the structure of a noise classification model.

[0055] Figure 7 A schematic diagram of the structure of a first microscopic data classification sub-model provided in this application

[0056] Figure 8 A schematic diagram of the structure of a microscopic measurement data processing system provided in this application;

[0057] Figure 9 This is a schematic diagram of the structure of a microscopic measurement data processing device provided in this application.

[0058] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0059] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0060] With the rapid development of microelectronics technology, high precision and high reliability in microstructure surface measurement have become indispensable in modern scientific research and industrial manufacturing. Confocal microscopy is increasingly used in precision measurement and testing, but the problem of low imaging resolution still exists.

[0061] To address the aforementioned issues, current technical solutions primarily focus on expanding the imaging range and improving the resolution of confocal microscopes. Expanding the imaging range aims to cover a larger sample surface area, enabling more comprehensive and accurate surface information in industrial manufacturing. Improving the resolution of confocal microscopes means being able to observe finer structures and features, thus providing more refined measurement results. Current improvements to expand the imaging range mainly involve optimizing optical path design and increasing detector sensitivity and response range. Optimizing the optical path design reduces light loss during imaging, improving image brightness and clarity. Simultaneously, increasing detector sensitivity and response range allows for the capture of weaker signals and expands the imaging range. Improvements to confocal microscope resolution primarily involve optimizing optical components. Using high-quality optical components, such as high numerical aperture objectives and efficient optical filters, can improve the resolution of the imaging system.

[0062] However, while the aforementioned technical solutions achieve improved imaging resolution, they come at the cost of reduced performance in other areas. For example, expanding the imaging range may slow down the imaging process, while increasing resolution may increase the complexity and cost of the imaging system. Therefore, how to improve imaging resolution without compromising other performance characteristics has become a pressing issue.

[0063] To address the aforementioned problems, the inventors, while researching methods to improve imaging resolution, discovered that to observe whether quality issues exist within fine structures, it is necessary to construct a three-dimensional microscopic image based on microscopic measurement data acquired by a microscopic measuring device. However, currently constructed three-dimensional images suffer from low imaging resolution and poor quality. To solve these problems, the hardware components of the microscopic measuring device have undergone varying degrees of improvement, including increased resolution and improved optical path design. However, while improvements to the hardware have enhanced the imaging resolution of the microscopic image, they have also resulted in a decrease in other technical indicators, such as slower imaging speed and higher improvement costs. Therefore, the inventors considered whether high-resolution imaging could be achieved at the software level, rather than relying on hardware improvements to increase imaging resolution. Specifically, a microscopic data processing platform normalizes the microscopic measurement data acquired by the microscopic measuring device and uses a noise classification model to identify the type and quantity of noise. Based on the obtained noise information, the noise removal model can accurately remove different types of noise and store the processed data in the microscopic measurement database, thereby improving data quality and accuracy and laying the foundation for constructing high-quality three-dimensional microscopic images.

[0064] Figure 1 This is a schematic diagram illustrating an application scenario for the microscopic measurement data processing method provided in this application. For example... Figure 1 As shown, the application scenario of the solution provided in this application includes a microscopic measurement device 100 and a server 101. The server 101 is used to deploy a microscopic data processing platform 1011 and a microscopic measurement database 1012. The microscopic measurement device 100 is connected to the server 101 via a communication network.

[0065] A microscopic measuring device 100 is a tool or apparatus used to observe and measure minute dimensions, shapes, and surface features. Common microscopic measuring devices 100 include optical microscopes, electron microscopes, atomic force microscopes, laser scanning confocal microscopes, X-ray diffraction microscopes, and laser interferometer microscopes. Each common microscopic measuring device 100 has its specific application area and measurement principle. The specific device selected depends on the actual needs and sample characteristics. Although... Figure 1 Only one microscopic measuring device 100 is shown, but it should be understood that two or more microscopic measuring devices 100 may exist.

[0066] Server 101 is primarily used to deploy the microscopic data processing platform 1011 and the microscopic measurement database 1012. The microscopic data processing platform 1011 contains a pre-trained large-scale noise processing model used to perform denoising processing on the microscopic measurement data acquired by the microscopic measurement device 100. The microscopic measurement database 1012 is primarily used to store the denoised microscopic measurement data. Although... Figure 1 Only one server 101 is shown, but it should be understood that there may be two or more servers 101.

[0067] The microscopic data processing platform 1011 in server 101 receives microscopic measurement data acquired from the microscopic measuring device 100 in real time. First, an adaptive normalization method is used to normalize the microscopic measurement data, obtaining first microscopic measurement data. Next, the microscopic data processing platform 1011 uses a noise classification model from a large noise processing model to classify the noise type and quantity in the first microscopic measurement data. Based on the obtained noise type and quantity, and combined with a noise removal model from the large noise processing model, noise is removed from the microscopic measurement data, resulting in second microscopic measurement data after noise removal. Finally, the microscopic data processing platform 1011 stores the noise-removed second microscopic measurement data in the microscopic measurement database 1012. Through this process, the microscopic data processing platform 1011 can effectively process and store microscopic measurement data, providing an accurate and reliable data foundation for subsequent analysis and applications.

[0068] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0069] Figure 2 This is a schematic flowchart of an embodiment of the microscopic measurement data processing method provided in this application. Figure 2 As shown, the workflow of this microscopic measurement data processing method may include:

[0070] S201: The microscopic measuring device acquires microscopic measurement data and uploads the microscopic measurement data to the microscopic data processing platform.

[0071] In this step, three-dimensional microscopic imaging requires microscopic measurement data, therefore, the acquisition of this data is crucial. The acquisition of this microscopic measurement data primarily relies on the microscopic measurement apparatus.

[0072] Specifically, a microscopic measuring device is an instrument or device used to observe and measure minute objects. It typically includes a microscope to magnify the sample so that it can be observed with the naked eye, and is equipped with measuring scales or graduations to measure and analyze the object's size, shape, position, etc. Microscopic measuring devices are widely used in scientific research, industrial manufacturing, medical diagnostics, and other fields, providing strong support for studying microstructures and performing precise measurements and analyses. Common microscopic measuring devices include optical microscopes, electron microscopes, atomic force microscopes, laser scanning confocal microscopes, X-ray diffraction microscopes, and laser interferometer microscopes. In practical applications, the appropriate device is selected for microscopic observation and measurement based on specific application needs and measurement requirements.

[0073] After identifying the sample to be observed, the microscopic measuring device performs microscopic measurements and acquires microscopic measurement data in real time. Because the microscopic measurement data acquired by the device may have issues such as low quality, it is necessary to further upload the obtained microscopic measurement data to a microscopic data processing platform in real time to enable further processing of the microscopic measurement data, thereby providing high-quality microscopic measurement data for subsequent three-dimensional microscopic imaging.

[0074] S202: The microscopic data processing platform normalizes the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data.

[0075] In this step, based on step S201, after the microscopic data processing platform receives the microscopic measurement data collected and acquired by the microscopic measurement device, it first normalizes the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data.

[0076] Specifically, the microscopic data processing platform is deployed on a server and connected to the microscopic measurement device via a communication network. The adaptive normalization method is pre-built into the platform as an algorithm, enabling dynamic normalization of the real-time transmitted microscopic measurement data. Normalization refers to converting microscopic measurement data of different scales and distribution ranges into a unified standard value, thereby eliminating the influence of different value ranges of microscopic measurement data on subsequent model training. Commonly used normalization methods include Min-Max normalization, Z-Score normalization, and the Sigmoid function. To adapt to the real-time and massive nature of the microscopic measurement data acquisition process, a normalization method suitable for streaming data needs to be designed, namely, an adaptive normalization method.

[0077] Optionally, the adaptive normalization method can adaptively adjust the window size based on the transmission rate of the microscopic measurement data. If the microscopic measurement data rate increases, the window size decreases; if the microscopic measurement data rate decreases, the window size increases, ensuring that the microscopic data processing platform processes the microscopic measurement data at an ideal rate. In one possible implementation, a pre-set window is first used as the current window, and its maximum and minimum values ​​are used for Min-Max normalization. Then, when the next window arrives, it replaces the previous window as the current window, and the data rate of the current window is compared with that of the previous window to predict the size of the next window. Here, the data rate... The calculation formula is as follows:

[0078]

[0079] in, This refers to the data rate. This indicates the transmission rate of the microscopic measurement data in the current window. This indicates the transmission rate of the microscopic measurement data from the previous window.

[0080] In this process, an ideal data rate is set in advance. As a threshold, real-time comparison and The size relationship. When If this happens, a fixed constant will be added to the current window length. .when If this happens, the current window length will be reduced by a fixed constant. Among them, fixed constants Predefined. (By) and By comparing the sizes, the current window size can be adjusted to the ideal rate. Based on this, the microscopic data processing platform uses the maximum and minimum values ​​in the modified window to perform Min-Max normalization on the microscopic measurement data in the window.

[0081] After the microscopic data processing platform normalizes the microscopic measurement data using an adaptive normalization method, the first microscopic measurement data can be obtained.

[0082] S203: The microscopic data processing platform uses the noise classification model in the pre-trained noise processing large model to classify the noise type and noise quantity in the first microscopic measurement data, and obtain at least one noise type and the noise quantity corresponding to each noise type.

[0083] In this step, based on step S202, the microscopic data processing platform normalizes the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data. Then, it uses the noise classification model in the pre-trained noise processing large model to classify the noise type and noise quantity in the first microscopic measurement data.

[0084] Specifically, the pre-trained noise processing model comprises two sub-models, each performing different functions. One of these sub-models is a noise classification model, built from a multi-layer neural network structure. Its main function is to identify the types and quantities of noise present in the first microscopic measurement data. The noise classification model takes the first microscopic measurement data as input and outputs at least one noise type and the quantity of noise corresponding to each type. Through this model, the types and quantities of noise present in the first microscopic measurement data can be accurately distinguished. The noise types can be any one or more of environmental noise, power supply noise, sensor noise, analog circuit noise, and scattering noise, etc. The quantity corresponding to each noise type will also vary; for example, environmental noise may appear 3 times, while sensor noise may appear 7 times. By classifying and statistically analyzing the noise, the distribution and influence of different noise types in the first microscopic measurement data can be better understood.

[0085] S204: The microscopic data processing platform removes noise from the first microscopic measurement data by using the noise removal model in the large noise processing model according to at least one noise type and the noise quantity corresponding to each noise type, and obtains the denoised second microscopic measurement data.

[0086] In this step, based on step S203, the microscopic data processing platform uses the noise classification model in the pre-trained noise processing large model to classify the noise type and noise quantity in the first microscopic measurement data, and obtains at least one noise type and the noise quantity corresponding to each noise type. Then, the microscopic data processing platform uses the noise removal model in the noise processing large model to remove the noise in the first microscopic measurement data according to the at least one noise type and the noise quantity corresponding to each noise type, and obtains the denoised second microscopic measurement data.

[0087] The noise removal model is the second sub-model within the larger noise processing model. Its function is to remove different types of noise from the first set of microscopic measurement data. Specifically, the noise removal model takes the first set of microscopic measurement data as input and uses a built-in algorithm to remove noise from it, effectively reducing or eliminating its influence, resulting in noise-removed second set of microscopic measurement data as output. Through the processing of the noise removal model, the output second set of microscopic measurement data will be more accurate and clearer, and will restore the characteristics of the original microscopic measurement data as much as possible.

[0088] S205: Store the second microscopic measurement data in the microscopic measurement database.

[0089] In this step, based on step S204, the microscopic data processing platform removes noise from the first microscopic measurement data using a noise removal model in the noise processing large model according to at least one noise type and the number of noises corresponding to each noise type. After obtaining the denoised second microscopic measurement data, the second microscopic measurement data is stored in the microscopic measurement database.

[0090] The microscopic measurement database is deployed on a server and is primarily built using a relational database management system, such as Oracle Database, MySQL, or SQL Server. The database is designed to support microscopic measurement data of various sample types, and it predefines corresponding database tables and names for different types of microscopic measurement data. This predefined structure effectively improves the efficiency of data querying and retrieval.

[0091] The microscopic measurement data processing method provided in this embodiment mainly illustrates how to preprocess microscopic measurement data using a microscopic data processing platform to obtain high-quality, clearer microscopic measurement data, thereby providing a more accurate and reliable data foundation for the subsequent construction of three-dimensional microscopic images. Specifically, the large-scale noise processing model built into the microscopic data processing platform can accurately determine the type and quantity of noise present in the microscopic measurement data, and can perform denoising based on different noise types, thus obtaining high-quality microscopic measurement data and helping to improve the resolution and quality of microscopic imaging.

[0092] Figure 3 This is a schematic flowchart of Embodiment 2 of the microscopic measurement data processing method provided in this application. Figure 3 As shown, based on the above embodiments, the noise removal model removes noise from the first microscopic measurement data to obtain the denoised second microscopic measurement data. The process of this microscopic measurement data processing method may include:

[0093] S301: Determine the denoising processing order for at least one noise type based on the number of noises corresponding to each noise type.

[0094] In this step, the microscopic data processing platform uses the noise classification model in the pre-trained noise processing large model to classify the noise type and noise quantity in the first microscopic measurement data, and obtains at least one noise type and the noise quantity corresponding to each noise type. Then, the noise removal model in the noise processing large model is used to remove the noise in the microscopic measurement data.

[0095] In the process of removing noise from microscopic measurement data, the noise removal model needs to determine the denoising order for at least one noise type based on the amount of noise corresponding to each noise type.

[0096] Optionally, the noise reduction processing order can be determined based on the quantity of noise corresponding to each noise type, the severity of the noise type, or a custom order based on the specific circumstances.

[0097] Based on the noise classification model, the microscopic measurement data was processed and found to contain three types of noise: noise type A, noise type B, and noise type C. Noise type A corresponds to 8 noise items, noise type B corresponds to 2 noise items, and noise type C corresponds to 4 noise items.

[0098] If the noise removal process order is determined according to the quantity of noise corresponding to each noise type, then the noise removal model determines the noise processing order as noise type A, noise type C, and noise type B.

[0099] If the noise removal process order is determined according to the severity of different noise types, the noise removal model may determine the noise processing order as noise type C, noise type A, and noise type B. Among them, noise type C has the greatest impact on the subsequent microscopic imaging resolution, followed by noise type A, and noise type B last.

[0100] If the noise removal process order is determined using a custom method, the noise removal model may determine the noise processing order as noise type A, noise type B, and noise type C. The specific order needs to be determined based on the actual situation.

[0101] S302: According to the denoising processing order, each noise type in the first microscopic measurement data is denoised using a variety of denoising algorithms corresponding to the pre-set noise type. The best result is selected from the results of different denoising algorithms as the input of the denoising algorithm corresponding to the next noise type. This process continues until all noise of at least one noise type in the first microscopic measurement data is denoised, and the second microscopic measurement data is obtained.

[0102] In this step, based on step S301, after the microscopic data processing platform determines the denoising processing order for at least one noise type according to the number of noises corresponding to each noise type, it sequentially performs denoising processing on each noise type in the first microscopic measurement data using multiple denoising algorithms corresponding to the pre-set noise type. The best processing result is selected from the results of different denoising algorithms as the input of the denoising processing algorithm corresponding to the next noise type, until all noise of at least one noise type in the first microscopic measurement data is denoised, and the second microscopic measurement data is obtained.

[0103] The noise removal model pre-configures multiple denoising algorithms. In one possible implementation, confocal microscopy measurement data obtained using a confocal microscope contains three types of noise: device noise, thermal noise, and electronic noise. For device noise, algorithms such as mean filtering, median filtering, or Gaussian filtering can be used for denoising. For thermal noise, algorithms such as averaging filters, median filters, or Gaussian filters can be used for denoising. For electronic noise, algorithms such as frequency domain filters, median filters, or wavelet transforms can be used for denoising. According to the order of denoising, the noise removal model calls the corresponding multiple denoising algorithms in parallel to denoise the same type of noise, obtaining at least one denoising result. The noise removal model selects the optimal denoising result as the input for the next noise type's denoising algorithm, until all noise is removed, obtaining the second set of denoised microscopic measurement data.

[0104] Figure 4 This is a schematic diagram of a noise removal model structure provided in this application. Figure 4 As shown, the noise removal model 400 includes an input module 401, a noise removal module 402, and an output module 403. The input module 401 receives first microscopic measurement data, noise type, and corresponding noise quantity, and sends this data to the noise removal model 402. Following a pre-defined denoising process, the corresponding noise type is determined, and noise is removed. Finally, the denoised microscopic measurement data is output to the output module 403 as the second microscopic measurement data. The noise removal module 402 includes denoising algorithms 1-1, 1-2, and 1-3 for noise type 1; denoising algorithms 2-1 and 2-2 for noise type 2; and denoising algorithms 3-1, 3-2, and 3-3 for noise type 3. Based on the determined denoising process order, the noise removal model 400 can locate the noise processing module corresponding to different noise types, achieving the removal of the corresponding noise type.

[0105] The microscopic measurement data processing method provided in this embodiment aims to detail how a noise removal model removes noise from the first microscopic measurement data. The noise removal model sequentially applies multiple algorithms to each noise type according to the denoising processing order, selecting the optimal result as the input for the next noise type, until at least one noise type is completely removed. By determining the denoising processing order and selecting the optimal algorithm, interference from different types of noise can be effectively removed, resulting in clearer and more accurate second microscopic measurement data. Furthermore, this processing method is flexible, adaptable to various noise types and levels, and significantly improves the denoising effect.

[0106] Figure 5 This is a flowchart illustrating Embodiment 3 of the microscopic measurement data processing method provided in this application. Figure 5 As shown, based on the above embodiments, the process of this microscopic measurement data processing method can include selecting the optimal processing result from the results corresponding to different denoising algorithms, and the flow can also include:

[0107] S501: Calculate the peak signal-to-noise ratio and structural similarity of the first microscopic measurement data and the results corresponding to each denoising algorithm.

[0108] In this step, the noise removal model removes noise from the first microscopic measurement data to obtain the denoised second microscopic measurement data. During this process, the optimal processing result needs to be selected from the results of different denoising algorithms. The selection of the optimal processing result is based on peak signal-to-noise ratio and structural similarity.

[0109] Specifically, for each type of noise, the noise removal model calls the corresponding denoising algorithm for parallel processing. This results in multiple denoising outcomes. For example, based on noise type A, the noise removal model might call denoising algorithm 1, denoising algorithm 2, and denoising algorithm 3. After denoising, denoising outcome 1, denoising outcome 2, and denoising outcome 3 are obtained. Then, the peak signal-to-noise ratio (PSNR) and structural similarity for each denoising outcome need to be calculated separately.

[0110] In image processing, Peak Signal-to-Noise Ratio (PSNR) is used to evaluate the degree of distortion between the original image and the reconstructed image after compression or other processing. It quantifies image quality by comparing the mean squared error between the original and reconstructed images, essentially calculating the average difference at the pixel level. A higher PSNR indicates less distortion and better image quality.

[0111] In the process of microscopic measurement data processing, the acquired microscopic measurement data is usually presented in the form of two-dimensional images. Assuming the two-dimensional image before denoising is labeled Image_A, and the two-dimensional images after denoising using two denoising algorithms corresponding to noise type A are labeled Image_A1 and Image_A2, then the peak signal-to-noise ratios (PSNRs) of Image_A and Image_A1 are... The calculation formula is:

[0112]

[0113] in, The maximum pixel value in Image_A The mean squared error is calculated using the following formula:

[0114]

[0115] Where Image_A has M×N pixels, M represents the length of Image_A, and N represents the height of Image_A. This represents the pixel in the i-th column. This represents the pixel in the j-th row.

[0116] Structural similarity between Image_A and Image_A1 The calculation formula is:

[0117]

[0118] in, This represents the average value of all pixels in Image_A. This represents the average value of all pixels in Image_A1. This represents the variance of all pixels in Image_A. This represents the variance of all pixels in Image_A1. This represents the covariance of all pixels in Image_A and Image_A1. and This is a user-defined constant to avoid situations where the denominator is 0.

[0119] Optionally, after calculating the peak signal-to-noise ratio... and structural similarity Then, using peak signal-to-noise ratio and structural similarity weighted sum The final evaluation metric for judging the quality of the denoising result is calculated using the following formula:

[0120]

[0121] in, and This is an adjustment coefficient, conforming to... The calculation formula, Peak signal-to-noise ratio, For structural similarity.

[0122] Similarly, the peak signal-to-noise ratios corresponding to Image_A and Image_A2 The calculation formula is:

[0123]

[0124] in, The maximum pixel value in Image_A The mean squared error is calculated using the following formula:

[0125]

[0126] Where Image_A has M×N pixels, M represents the length of Image_A, and N represents the height of Image_A. This represents the pixel in the i-th column. This represents the pixel in the j-th row. The structural similarity between Image_A and Image_A2. The calculation formula is:

[0127]

[0128] in, This represents the average value of all pixels in Image_A. This represents the average value of all pixels in Image_A2. This represents the variance of all pixels in Image_A. This represents the variance of all pixels in Image_A2. This represents the covariance of all pixels in Image_A and Image_A2. and This is a user-defined constant to avoid situations where the denominator is 0.

[0129] After calculating the peak signal-to-noise ratio and structural similarity Then, using peak signal-to-noise ratio and structural similarity weighted sum The final evaluation metric for judging the quality of the denoising result is calculated using the following formula:

[0130]

[0131] in, and This is an adjustment coefficient, conforming to... The calculation formula, Peak signal-to-noise ratio, For structural similarity.

[0132] After calculating the evaluation indicators and Then, the optimal denoising result can be determined.

[0133] S502: Based on the peak signal-to-noise ratio and structural similarity, select the processing result with the best effect from the results corresponding to different denoising algorithms.

[0134] In this step, based on step S502, after calculating the peak signal-to-noise ratio and structural similarity of the first microscopic measurement data with the results corresponding to each denoising algorithm, the weighted sum of the peak signal-to-noise ratio and structural similarity is used as the final evaluation index, and then the processing result with the best effect is selected from the results corresponding to different denoising algorithms.

[0135] For a given type of noise, different denoising algorithms produce different denoising results. The optimal denoising algorithm is determined by an evaluation metric: the weighted sum of peak signal-to-noise ratio (PSNR) and structural similarity. A higher weighted sum of PNR and structural similarity indicates better denoising performance. For example, for denoising type R, the weighted sum of PNR and structural similarity for denoising algorithm R_1 is 18, for R_2 it is 5, and for R_3 it is 19.1. Based on these weighted sums of PNR and structural similarity for different denoising algorithms, algorithm R_3 is determined to have the best denoising effect. Therefore, the denoising result of algorithm R_3 is selected as the input for the next noise type's denoising algorithm, and subsequent processing follows a similar pattern.

[0136] This embodiment of the microscopic measurement data processing method mainly illustrates how to select the optimal processing result from the results corresponding to different denoising algorithms. Specifically, for the first microscopic measurement data, different denoising algorithms are used to process the results, the peak signal-to-noise ratio (PSNR) and structural similarity are calculated, the results are compared and analyzed, and the processing result with the optimal PSNR and structural similarity is selected. By calculating the PSNR and structural similarity, the processing effect of different denoising algorithms on microscopic measurement data can be objectively evaluated. Furthermore, selecting the processing result with higher PSNR and structural similarity can obtain microscopic measurement data that is closer to the original data, reducing noise interference and laying the foundation for extracting more accurate and clear features and details in subsequent microscopic image construction.

[0137] Based on any of the above embodiments, in one possible implementation, the noise classification model is a neural network model for classifying data corresponding to microscopic images, which is obtained by pre-training an initial model constructed based on a convolutional neural network model.

[0138] The initial model includes a first microscopic data classification sub-model, a second microscopic data classification sub-model, a third microscopic data classification sub-model, and an output unit. The output unit is used to calculate the average value of the classification results of the first microscopic data classification sub-model, the second microscopic data classification sub-model, and the third microscopic data classification sub-model and then output it.

[0139] The first microscopic data classification sub-model includes one input layer, three output layers, and 20 hidden layers. The 20 hidden layers include 10 convolutional layers, 6 pooling layers, 3 fully connected layers, and one activation layer.

[0140] The second microscopic data classification sub-model includes one input layer, three output layers, and 30 hidden layers. The 30 hidden layers include 17 convolutional layers, 7 pooling layers, 5 fully connected layers, and one activation layer.

[0141] The third microscopic data classification sub-model includes one input layer, three output layers, and 25 hidden layers. The 25 hidden layers include 12 convolutional layers, eight pooling layers, four fully connected layers, and one activation layer.

[0142] Specifically, the microscopic data processing platform normalizes the microscopic measurement data using an adaptive normalization method to obtain the first set of microscopic measurement data. Then, it uses a pre-trained noise classification model from a large noise processing model to classify the noise type and quantity in the first set of microscopic measurement data. The noise classification model is a neural network model that is pre-trained on an initial model built based on a convolutional neural network model to classify the data corresponding to the microscopic images.

[0143] The initial model corresponding to this noise classification model includes a first microscopic data classification sub-model, a second microscopic data classification sub-model, a third microscopic data classification sub-model, and an output unit. The output unit calculates the average of the classification results from the first, second, and third microscopic data classification sub-models and outputs the average. The first, second, and third microscopic data classification sub-models are all built using convolutional neural networks, and their function is to obtain the noise type and quantity corresponding to the first microscopic measurement data. Considering the instability of the output results of the convolutional neural network structure, multiple classification sub-models are designed to process the same first microscopic measurement data in parallel, and the processing results are averaged to ensure a more accurate classification result.

[0144] Figure 6 This is a schematic diagram of the structure of a noise classification model provided in this application. Figure 6 As shown, the noise classification model 600 includes a first microscopic data classification sub-model 601, a second microscopic data classification sub-model 602, a third microscopic data classification sub-model 603, and an output unit 604. The first microscopic data classification sub-model 601, the second microscopic data classification sub-model 602, and the third microscopic data classification sub-model 603 all take the first microscopic measurement input as input, and after processing by an internal multi-layer convolutional neural network, output the corresponding noise type and the corresponding noise quantity.

[0145] Figure 7 This is a schematic diagram of the structure of a first microscopic data classification sub-model provided in this application. Figure 7As shown, the first microscopic data classification sub-model 700 includes an input module 701, a convolution module 702, a pooling module 703, a fully connected module 704, an activation module 705, and an output module 706. Specifically, the input module 701 includes one input layer, the convolution module 702 includes 10 convolutional layers, the pooling module 703 includes 6 pooling layers, the fully connected module 704 includes 3 fully connected layers, the activation module 705 includes one activation layer, and the output module 706 includes 3 output layers.

[0146] The input layer is mainly used to receive the two-dimensional microscopic image formed by the first microscopic measurement data. The value of each pixel is passed to the convolutional layer for processing.

[0147] Convolutional layers are crucial for feature extraction. In a convolutional layer, different features of a two-dimensional microscopic image are extracted by convolving the input data with pre-defined kernels. Each kernel scans a small region of the input data, calculates the convolution result between that region and the kernel, and generates a feature map. Through multiple convolutional layers, each kernel can extract different image features.

[0148] Pooling layers follow convolutional layers and are used to reduce the dimensionality of data and reduce computational cost. Common pooling operations include max pooling and average pooling. Pooling operations sample the feature maps output by the convolutional layers, retain the most salient features, and reduce the size of the feature maps, which helps to extract more abstract and stable features.

[0149] A fully connected layer connects all neurons in the previous layer to all neurons in the current layer. In a fully connected layer, each neuron is connected to all neurons in the previous layer, forming a fully connected network structure. The function of a fully connected layer is to further combine and transform the previously extracted features in order to better distinguish different types and amounts of noise.

[0150] Activation layers increase the expressive power of neural networks by introducing nonlinear transformations. Common activation functions include ReLU, Sigmoid, and Tanh. These functions can perform nonlinear mapping on the output of fully connected layers, enabling the network to learn more complex feature representations.

[0151] The output layer consists of three layers, defined as output layer 1, output layer 2, and output layer 3. Output layer 1 is used to output the quantity of various types of noise, output layer 2 is used to output the types of noise present in the first microscopic measurement data, and output layer 3 is used to integrate the output results of output layer 1 and output layer 2 to obtain the types of noise and their corresponding quantities present in the first microscopic measurement data.

[0152] After the first microscopic data classification sub-model structure is completed, the pre-prepared training set and validation set are used as inputs to train the first microscopic data classification sub-model, thereby determining the noise type and noise quantity.

[0153] The training set and validation set include two-dimensional microscopic images corresponding to massive amounts of microscopic measurement data, as well as the types and quantities of noise included in each two-dimensional microscopic image.

[0154] Optionally, training and validation sets can be obtained through websites such as The Cell Image Library, The Allen Cell Explorer, The Broad Bioimage Benchmark Collection, and The Image Data Resource. The Cell Image Library is an online image library containing various biological cell images, including confocal microscopy images. The Allen Cell Explorer is a database provided by the Allen Institute for Brain Science, containing a large number of confocal microscopy images covering various tissues, cells, and subcellular structures in mice and humans. The Broad Bioimage Benchmark Collection is an image database provided by the Broad Bioimage Library, containing a large number of confocal microscopy images. The Image Data Resource is an online platform for sharing and managing life science image data, including many confocal microscopy image datasets. Alternatively, large amounts of microscopic measurement data can be obtained through manual labeling, or training and validation sets can be generated for the first microscopic data classification sub-model using generative adversarial networks (GANs).

[0155] Optionally, considering the pixel differences in the two-dimensional microscopic images corresponding to the acquired microscopic measurement data, when inputting the microscopic measurement data into the first microscopic data classification sub-model, it is necessary to compare the size relationship between the two-dimensional microscopic image to be processed and the two-dimensional microscopic image that the first microscopic data classification sub-model can process. If the size of the two-dimensional microscopic image to be processed is equal to the size of the two-dimensional microscopic image that the first microscopic data classification sub-model can process, no processing is performed, and it can be directly input into the convolutional layer for subsequent processing; if the size of the two-dimensional microscopic image to be processed is greater than the size of the two-dimensional microscopic image that the first microscopic data classification sub-model can process, then the two-dimensional microscopic image to be processed is compressed; if the size of the two-dimensional microscopic image to be processed is smaller than the size of the two-dimensional microscopic image that the first microscopic data classification sub-model can process, then the confocal microscopic image to be processed is zero-padding.

[0156] Among them, the first microscopic data classification sub-model can process two-dimensional microscopic images of various sizes. It can be obtained by setting the number of neurons and the size of the convolution kernel in advance, and the calculation formula is:

[0157]

[0158] in, This indicates the size of the two-dimensional microscopic images that the first microscopic data classification sub-model can process. This indicates the number of neurons, i.e., the size of the feature map output by the convolutional layer. This indicates the kernel size, i.e., the dimensions of the convolution kernel. This represents the stride, which is the sliding stride of the convolution kernel when performing a convolution operation.

[0159] loss function The calculation formula is:

[0160]

[0161] in, The loss function is defined during pre-training of the first microscopic data classification sub-model. and This is an adjustment coefficient, which meets the conditions. , To determine the loss function corresponding to the noise type, To determine the loss function corresponding to the amount of noise. Specifically, what is the loss function corresponding to the type of noise? The calculation formula is as follows:

[0162]

[0163] in, This represents the loss function corresponding to the noise type. Represents the total number of samples. This represents the true label, i.e., the true type of noise. This represents the actual label, i.e., the noise type obtained from the first microscopic data classification sub-model.

[0164] Loss function corresponding to the amount of noise The calculation formula is as follows:

[0165]

[0166] in, The loss function represents the amount of noise. Represents the total number of samples. This represents the true label, i.e., the actual amount of noise. This represents the actual label, i.e., the amount of noise obtained from the first microscopic data classification sub-model.

[0167] When the current loss function of the first microscopic data classification sub-model Value and the previous loss function When the absolute value of the difference between the values ​​is less than or equal to a set threshold, it indicates that the training of the first microscopic data classification sub-model is complete. In this embodiment, the indication that the training of the first microscopic data classification sub-model is complete is that the absolute value of the difference between adjacent loss function values ​​is less than or equal to a preset threshold, or it could be that the ratio between adjacent loss function values ​​is less than or equal to other preset thresholds, etc.

[0168] The second microscopic data classification sub-model has the same function as the first microscopic data classification sub-model: to obtain the type and quantity of noise present in the microscopic measurement data. To achieve this function, the second microscopic data classification sub-model requires pre-training. The pre-training process is similar to that of the first microscopic data classification sub-model and will not be described in detail here.

[0169] The framework of the second microscopic data classification sub-model differs from that of the first. The second sub-model includes one input layer, three output layers, and 30 hidden layers. The 30 hidden layers comprise 17 convolutional layers, 7 pooling layers, 5 fully connected layers, and one activation layer. The functions of each layer are consistent with their corresponding functions in the first sub-model, and will not be elaborated further here.

[0170] The third microscopic data classification sub-model has the same function as the first and second microscopic data classification sub-models: to obtain the type and quantity of noise present in the microscopic measurement data. To achieve this function, the third microscopic data classification sub-model also requires pre-training, the process of which is similar to that of the first microscopic data classification sub-model and will not be described in detail here.

[0171] The framework of the third microscopic data classification sub-model differs from that of the first and second microscopic data classification sub-models. The third sub-model includes one input layer, three output layers, and 25 hidden layers. These 25 hidden layers comprise 12 convolutional layers, 8 pooling layers, 4 fully connected layers, and one activation layer. The functions of each layer are consistent with their corresponding functions in the first microscopic data classification sub-model, and will not be elaborated further here.

[0172] This embodiment details the internal structure of the noise classification model. The noise classification model is a neural network model trained on an initial model built from a convolutional neural network, used to classify noise in microscopic measurement data corresponding to microscopic images. The model includes first, second, and third microscopic data classification sub-models and an output unit. The output unit calculates the average classification results of the three sub-models and outputs the result. Each sub-model includes an input layer, an output layer, and a hidden layer containing various structures such as convolutional layers, pooling layers, fully connected layers, and activation layers. By using three sub-models and averaging their classification results, the overall stability and accuracy of noise classification can be improved, thereby reducing the randomness introduced by a single model, increasing the model's robustness, and making the results more reliable and consistent, providing a foundation for subsequent noise removal.

[0173] Based on any of the above embodiments, in one possible implementation, the number of pixels in the microscopic image corresponding to the microscopic measurement data is M*N*3, where M represents the number of row pixels, N represents the number of column pixels, and 3 represents the number of channels.

[0174] Correspondingly, the kernel size of the convolutional layer in the first microscopic data classification sub-model is a*a*3, where the first a represents the length of the kernel, the second a represents the width of the kernel, 3 represents the number of channels, and a takes any integer value between 1 and n, where n is the minimum value of M and N.

[0175] Specifically, microscopic images are built upon microscopic measurement data, and there is a correspondence between them. Specifically, microscopic measurement data can be used to generate microscopic images, where each pixel contains information about the measured object. A pixel in a microscopic image can be represented as M*N*3, where M represents the number of rows of pixels, N represents the number of columns of pixels, and 3 represents the number of channels. This means that each pixel consists of three values, representing the intensity of the red, green, and blue channels, respectively. By combining these three channels, a corresponding color microscopic image can be obtained.

[0176] Before removing noise from microscopic images, it's necessary to first determine the type and quantity of noise in the corresponding image based on a noise classification model. This model uses multiple convolutional layers to extract feature maps corresponding to the microscopic image. Convolutional layers are an important component of neural networks, effectively extracting local features from microscopic images through convolution operations. In the noise classification model, each convolutional layer takes the input microscopic image as input and performs convolution operations on it using a set of convolutional kernels. Each convolutional kernel can be considered a filter, detecting specific features in the microscopic image, such as edges and textures. By using multiple convolutional layers, the noise classification model can gradually extract more abstract and high-level feature representations.

[0177] In the first microscopic data classification sub-model, the convolutional kernel size of the convolutional layer is a*a*3, where the first 'a' represents the length of the kernel, the second 'a' represents the width of the kernel, and 3 represents the number of channels. The value of 'a' can be any integer between 1 and n, where n is the minimum of M and N. Specifically, in the first microscopic data classification sub-model, the convolutional kernel is square, meaning its length and width are the same. This ensures that during convolution, each kernel covers a local square region, allowing for better capture of structural features in the microscopic image.

[0178] This embodiment mainly describes how the convolutional kernel in the first microscopic data classification sub-model is designed based on the pixel size of the microscopic image corresponding to the microscopic measurement data. In the first microscopic data classification sub-model, the size of the convolutional kernel is set to a*a*3, that is, a square convolutional kernel. This square convolutional kernel can capture more features at different scales in the microscopic image, and convolutional kernels of different sizes can capture different levels of detail and structural information, thereby improving the feature extraction capability. At the same time, since the size of the convolutional kernel is the same across the entire microscopic image, the number of model parameters can be effectively reduced. This parameter sharing mechanism makes the model more lightweight, reduces the computational burden, and improves the training speed and the model's generalization ability.

[0179] Based on any of the above embodiments, in one possible implementation, the size of the convolution kernel of the convolutional layer in the second microscopic data classification sub-model is b*1*3, where b represents the length of the convolution kernel, the width of the convolution kernel is 1, the number of channels of the convolution kernel is 3, and the value of b is any integer between 1 and M.

[0180] Specifically, in determining the type and quantity of noise, the applied neural network model is a noise classification model, which includes three sub-models. The second microscopic data classification sub-model incorporates multiple convolutional layers to extract features from the microscopic images corresponding to the microscopic measurement data. Each convolutional layer contains a convolutional kernel of different size than the kernel in the first microscopic data classification sub-model. This kernel has a size of b*1*3, where b represents the kernel length, 1 represents the kernel width, and 3 represents the number of channels. The value of b can be any integer between 1 and M. For example, if the pixel representation of a microscopic image is 6*5*3, the kernel length can be any integer between 1 and 6, meaning the kernel length can be any one of 1, 2, 3, 4, 5, or 6.

[0181] This embodiment primarily describes how to design the convolutional kernel in the second microscopic data classification sub-model based on the pixel size of the microscopic image corresponding to the microscopic measurement data. In the second microscopic data classification sub-model, the convolutional kernel size is set to b*1*3, i.e., a horizontally rectangular convolutional kernel. This rectangular convolutional kernel can focus more on extracting vertical features. When processing microscopic images, due to the significant differences between different microscopic images, vertical feature information cannot be ignored. Therefore, using a rectangular convolutional kernel can better capture this feature, thereby improving the classification accuracy of the model.

[0182] Based on any of the above embodiments, in one possible implementation, the size of the convolution kernel of the convolutional layer in the third microscopic data classification sub-model is 1*c*3, the length of the convolution kernel is 1, c represents the width of the convolution kernel, the number of channels of the convolution kernel is 3, and the value of c is any integer between 1 and N.

[0183] Specifically, in determining the type and quantity of noise, the applied neural network model is a noise classification model, which includes three sub-models. The third microscopic data classification sub-model incorporates multiple convolutional layers to extract features from the microscopic images corresponding to the microscopic measurement data. Each convolutional layer contains a convolutional kernel, whose size differs from the kernels in the first and second microscopic data classification sub-models. This kernel has a size of 1*c*3, where 1 represents the kernel's length, c represents its width, and 3 represents the number of channels. The value of c can be any integer between 1 and N. For example, if the pixel representation of a microscopic image is 13*24*3, the width of the convolutional kernel could be any integer between 1 and 24.

[0184] This embodiment mainly describes how to design the convolution kernel in the third microscopic data classification sub-model based on the pixel size of the microscopic image corresponding to the microscopic measurement data. In the third microscopic data classification sub-model, the size of the convolution kernel is set to 1*c*3, that is, a vertically rectangular convolution kernel. The width of this vertically rectangular convolution kernel can take different values, which can more flexibly capture lateral features of different widths, providing strong support for the subsequent determination of noise type and noise quantity.

[0185] Based on any of the above embodiments, in one possible implementation, in the first microscopic data classification sub-model, the size of the convolution kernel in the convolutional layer decreases sequentially from input to output.

[0186] In the second microscopic data classification sub-model, the size of the convolutional kernel in the convolutional layer decreases sequentially from input to output.

[0187] In the third microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output.

[0188] Specifically, in the process of applying the noise classification model to determine the type and quantity of noise in microscopic measurement data, the multiple convolutional layers in the first, second, and third microscopic data classification sub-models of the noise classification model all correspond to convolutional kernels of different sizes. In the first microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output. For example, if the first microscopic data classification sub-model has three convolutional layers: convolutional layer 1, convolutional layer 2, and convolutional layer 3, and the kernel size of convolutional layer 1 is 5*5*3, then the kernel size of convolutional layer 2 is 4*4*3, and the kernel size of convolutional layer 3 is 3*3*3. Alternatively, the kernel size of convolutional layer 2 could be 3*3*3, and the kernel size of convolutional layer 3 could be 2*2*3. The key is to ensure that the kernel size decreases sequentially from input to output.

[0189] In the second microscopic data classification sub-model, the kernel size in the convolutional layers decreases sequentially from input to output. For example, in the first microscopic data classification sub-model, there are four convolutional layers: convolutional layer 1, convolutional layer 2, convolutional layer 3, and convolutional layer 4. The kernel size for convolutional layer 1 is 8*1*3, then the kernel size for convolutional layer 2 is 6*1*3, the kernel size for convolutional layer 3 is 4*1*3, and the kernel size for convolutional layer 4 is 2*1*3. Alternatively, the kernel size for convolutional layer 2 could be 7*1*3, the kernel size for convolutional layer 3 could be 5*1*3, and the kernel size for convolutional layer 4 could be 3*1*3. The key is to ensure that the kernel size decreases sequentially from input to output.

[0190] In the third microscopic data classification sub-model, the kernel size in the convolutional layers decreases sequentially from input to output. For example, in the first microscopic data classification sub-model, there are three convolutional layers: convolutional layer 1, convolutional layer 2, and convolutional layer 3. The kernel size for convolutional layer 1 is 1*8*3, then the kernel size for convolutional layer 2 is 1*6*3, and the kernel size for convolutional layer 3 is 1*4*3. Alternatively, the kernel size for convolutional layer 2 could be 1*7*3, and the kernel size for convolutional layer 3 could be 1*5*3. The key is to ensure that the kernel size decreases sequentially from input to output.

[0191] This embodiment mainly illustrates the design of different convolutional layers in each microscopic classification sub-model, where the kernel size gradually decreases from input to output. This setting allows for the capture of detailed information in microscopic images at lower levels, while capturing more abstract features at higher levels. Simultaneously, reducing the kernel size also reduces the number of model parameters, thereby avoiding overfitting and laying the foundation for determining the type and quantity of noise.

[0192] Based on any of the above embodiments, in one possible implementation, at least one noise type includes at least one of the following: device noise, thermal noise, electronic noise, scattering noise, fluorescent dye noise, frosted glass noise, and analog-to-digital conversion noise.

[0193] Specifically, through observation and analysis of microscopic measurement data, the microscopic measurement data includes at least one or more of the following: device noise, thermal noise, electronic noise, scattering noise, fluorescent dye noise, frosted glass noise, and analog-to-digital conversion noise.

[0194] Device noise is introduced by the electronic and optical components of the instrument itself. Examples include the instability of the light source, the dark current of the detector, and the noise of the amplifier.

[0195] Thermal noise is caused by ambient temperature and random thermal motion generated by electronic components, which can lead to random brightness fluctuations in images.

[0196] Electronic noise is caused by the inherent noise of electronic components, and this noise usually manifests as random brightness variations in an image.

[0197] Scattering noise is caused by light scattering within the sample, which leads to background blurring and noise in the image.

[0198] Fluorescent dye-related noise refers to the noise that fluorescent dyes may introduce into fluorescence microscope images. This may include the spontaneous emission, photobleaching, or flickering of fluorescent dyes.

[0199] Ground glass noise is caused by irregular textures on the surface of the sample or microscope lens, which produces uniform noise and patterns in the image.

[0200] Analog-to-digital conversion noise is the error and noise that may be introduced when converting analog signals to digital signals.

[0201] Based on different noise types, the noise removal model employs an adaptive denoising algorithm to process the microscopic measurement data, obtaining denoised second microscopic measurement data. This data is then stored in a microscopic database for subsequent use in constructing microscopic images.

[0202] This embodiment primarily lists several different noise types. Since microscopic measurement data may contain one or more different types of noise, based on the prior acquisition and identification of these noise types, corresponding denoising algorithms are set in the noise removal model to achieve accurate removal of different noises. This adaptive processing provides a high-quality data foundation for imaging resolution, ensuring that the obtained second microscopic measurement data is clearer and more accurate. By targeting different noise types, the advantages of various denoising algorithms can be fully utilized to improve denoising effects, thereby effectively improving image quality.

[0203] Figure 8 This is a schematic diagram of the structure of a microscopic measurement data processing system provided in this application. Figure 8 As shown, the microscopic measurement data processing system 800 includes a microscopic measurement device 801, a microscopic data processing platform 802, and a microscopic measurement database 803. Among them,

[0204] The microscopic measuring device 801 is used to: acquire microscopic measurement data and upload the microscopic measurement data to the microscopic data processing platform.

[0205] Specifically, the microscopic measurement device 801 is a device used to acquire microscopic measurement data. Its main function is to observe and measure tiny objects through a microscope and collect the measurement results as data. The collected data is processed and uploaded to a microscopic data processing platform for further analysis and research. The microscopic measurement device 801 typically includes components such as a lens, light source, and image sensor, and can be customized and configured according to the needs of specific application scenarios. In fields such as biology, materials science, and electronic engineering, the microscopic measurement device 801 is widely used for the observation and analysis of microstructures, providing reliable data support for scientific research and engineering development.

[0206] The 802 microscopic data processing platform is used for:

[0207] The microscopic measurement data were normalized using an adaptive normalization method to obtain the first microscopic measurement data.

[0208] The noise classification model in the pre-trained noise processing large model is used to classify the noise type and noise quantity in the first microscopic measurement data, so as to obtain at least one noise type and the noise quantity corresponding to each noise type.

[0209] Based on at least one noise type and the number of noises corresponding to each noise type, the noise removal model in the large noise processing model is used to remove noise from the first microscopic measurement data to obtain the denoised second microscopic measurement data.

[0210] Specifically, the 802 microscopic data processing platform is a dedicated platform for processing and optimizing microscopic measurement data. It normalizes the data using an adaptive normalization method to ensure uniform scale and range, helping to eliminate data discrepancies and making subsequent processing and analysis more accurate and reliable, resulting in the first set of microscopic measurement data. Based on this, a noise classification model from a pre-trained noise processing large-scale model is used to classify the noise types and quantities in the first set of microscopic measurement data. This model identifies at least one noise type and calculates the quantity of noise corresponding to each type. This helps to further understand the noise situation in the data and provides a basis for subsequent noise removal. Finally, based on at least one noise type and the quantity of noise corresponding to each type, a noise removal model from the noise processing large-scale model is used to remove noise from the first set of microscopic measurement data. This processing yields the noise-removed second set of microscopic measurement data, thus improving the data quality and accuracy.

[0211] In summary, the 802 microscopic data processing platform uses various processing methods and models to normalize, classify, and remove noise from microscopic measurement data, thereby providing more reliable and accurate second microscopic measurement data and providing strong support for subsequent analysis and applications.

[0212] The microscopic measurement database 803 is used to store the second microscopic measurement data.

[0213] Specifically, the microscopic measurement database 803 is a database specifically designed to store second-stage microscopic measurement data. Its main function is to store and manage the second-stage microscopic measurement data processed by the microscopic data processing platform 802 for subsequent access and use. The microscopic measurement database 803 typically employs a distributed database design, enabling distributed storage and backup of data across multiple nodes, thereby improving data reliability and availability. Furthermore, this database supports various query and retrieval methods, allowing users to easily retrieve and obtain data according to their needs.

[0214] The microscopic measurement data processing system provided in this application includes a microscopic measurement device, a microscopic data processing platform, and a microscopic measurement database. Through the coordinated operation of the microscopic measurement device, the data processing platform, and the data storage database, comprehensive processing and management of microscopic measurement data can be achieved, thereby improving data quality and accuracy.

[0215] Figure 9 This is a schematic diagram of a microscopic measurement data processing device provided in this application. Figure 9 As shown, the microscopic measurement data processing device 900 includes:

[0216] The acquisition module 901 is used to acquire microscopic measurement data by the microscopic measuring device and upload the microscopic measurement data to the microscopic data processing platform;

[0217] Processing module 902 is used by the microscopic data processing platform to normalize the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data;

[0218] The processing module 902 is also used by the microscopic data processing platform to classify the noise type and noise quantity in the first microscopic measurement data using a noise classification model in a pre-trained noise processing large model, so as to obtain at least one noise type and the noise quantity corresponding to each noise type.

[0219] The processing module 902 is also used by the microscopic data processing platform to remove noise from the first microscopic measurement data by using the noise removal model in the noise processing large model according to at least one noise type and the noise quantity corresponding to each noise type, so as to obtain the noise-removed second microscopic measurement data.

[0220] Storage module 903 is used to store the second microscopic measurement data into the microscopic measurement database.

[0221] Optionally, the processing module 902 is also used for:

[0222] Determine the denoising processing order for at least one noise type based on the number of noises corresponding to each noise type;

[0223] According to the denoising process order, each noise type in the first microscopic measurement data is denoised using a variety of denoising algorithms corresponding to the pre-set noise type. The best result is selected from the results of different denoising algorithms as the input of the denoising algorithm for the next noise type. This process continues until all noise of at least one noise type in the first microscopic measurement data is denoised, resulting in the second microscopic measurement data.

[0224] Optionally, the processing module 902 is also used for:

[0225] The peak signal-to-noise ratio and structural similarity of the first microscopic measurement data with the results of each denoising algorithm were calculated respectively.

[0226] Based on peak signal-to-noise ratio and structural similarity, the processing result with the best effect is selected from the results corresponding to different denoising algorithms.

[0227] Optionally, the noise classification model is a neural network model that is pre-trained on an initial model built based on a convolutional neural network model and used to classify the data corresponding to the microscopic image.

[0228] The initial model includes a first microscopic data classification sub-model, a second microscopic data classification sub-model, a third microscopic data classification sub-model, and an output unit. The output unit is used to calculate the average value of the classification results of the first microscopic data classification sub-model, the second microscopic data classification sub-model, and the third microscopic data classification sub-model and then output it.

[0229] The first microscopic data classification sub-model includes one input layer, three output layers, and 20 hidden layers. The 20 hidden layers include 10 convolutional layers, 6 pooling layers, 3 fully connected layers, and one activation layer.

[0230] The second microscopic data classification sub-model includes one input layer, three output layers, and 30 hidden layers. The 30 hidden layers include 17 convolutional layers, 7 pooling layers, 5 fully connected layers, and one activation layer.

[0231] The third microscopic data classification sub-model includes one input layer, three output layers, and 25 hidden layers. The 25 hidden layers include 12 convolutional layers, eight pooling layers, four fully connected layers, and one activation layer.

[0232] Optionally, the number of pixels in the microscopic image corresponding to the microscopic measurement data is M*N*3, where M represents the number of rows, N represents the number of columns, and 3 represents the number of channels;

[0233] Correspondingly, the kernel size of the convolutional layer in the first microscopic data classification sub-model is a*a*3, where the first a represents the length of the kernel, the second a represents the width of the kernel, 3 represents the number of channels, and a takes any integer value between 1 and n, where n is the minimum value of M and N.

[0234] Optionally, in the second microscopic data classification sub-model, the size of the convolutional kernel of the convolutional layer is b*1*3, where b represents the length of the convolutional kernel, the width of the convolutional kernel is 1, the number of channels of the convolutional kernel is 3, and the value of b is any integer between 1 and M.

[0235] Optionally, in the third microscopic data classification sub-model, the size of the convolutional kernel of the convolutional layer is 1*c*3, the length of the convolutional kernel is 1, c represents the width of the convolutional kernel, the number of channels of the convolutional kernel is 3, and the value of c is any integer between 1 and N.

[0236] Optionally, in the first microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output;

[0237] In the second microscopic data classification sub-model, the size of the convolutional kernel in the convolutional layer decreases sequentially from input to output.

[0238] In the third microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output.

[0239] Optionally, at least one noise type includes at least one of the following:

[0240] Device noise, thermal noise, electronic noise, scattering noise, fluorescent dye noise, frosted glass noise, and analog-to-digital conversion noise.

[0241] This embodiment provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the microscopic measurement data processing method described in the above embodiment.

[0242] This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements the microscopic measurement data processing method provided in any of the above embodiments.

[0243] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0244] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for processing microscopic measurement data, characterized in that, An application is made in a microscopic measurement data processing system, the microscopic measurement data processing system comprising a microscopic measurement device, a microscopic data processing platform, and a microscopic measurement database, the method comprising: The microscopic measuring device acquires microscopic measurement data and uploads the microscopic measurement data to the microscopic data processing platform; The microscopic data processing platform normalizes the microscopic measurement data using an adaptive normalization method to obtain the first microscopic measurement data. The microscopic data processing platform uses a noise classification model in a pre-trained noise processing large model to classify the noise type and noise quantity in the first microscopic measurement data, and obtain at least one noise type and the noise quantity corresponding to each noise type. The microscopic data processing platform, based on the at least one noise type and the quantity of noise corresponding to each noise type, uses a noise removal model in the large noise processing model to remove noise from the first microscopic measurement data, obtaining denoised second microscopic measurement data, including: The denoising processing order for the at least one noise type is determined based on the number of noises corresponding to each noise type. According to the denoising processing order, each noise type in the first microscopic measurement data is denoised using a variety of denoising algorithms corresponding to the pre-set noise type. The best result is selected from the results of different denoising algorithms as the input of the denoising algorithm corresponding to the next noise type. This process continues until all noise of at least one noise type in the first microscopic measurement data is denoised, and the second microscopic measurement data is obtained. The step of selecting the optimal processing result from the results corresponding to different denoising algorithms includes: Calculate the peak signal-to-noise ratio and structural similarity of the first microscopic measurement data with the results corresponding to each denoising algorithm; Based on peak signal-to-noise ratio and structural similarity, the processing result with the best effect is selected from the results corresponding to different denoising algorithms; The second microscopic measurement data is stored in the microscopic measurement database.

2. The method according to any one of claims 1, characterized in that, The noise classification model is a neural network model used to classify data corresponding to microscopic images, which is obtained by pre-training an initial model constructed based on a convolutional neural network model. The initial model includes a first microscopic data classification sub-model, a second microscopic data classification sub-model, a third microscopic data classification sub-model, and an output unit. The output unit is used to calculate the average value of the classification results of the first microscopic data classification sub-model, the second microscopic data classification sub-model, and the third microscopic data classification sub-model and then output it. The first microscopic data classification sub-model includes one input layer, three output layers, and 20 hidden layers. The 20 hidden layers include 10 convolutional layers, 6 pooling layers, 3 fully connected layers, and one activation layer. The second microscopic data classification sub-model includes one input layer, three output layers, and 30 hidden layers. The 30 hidden layers include 17 convolutional layers, 7 pooling layers, 5 fully connected layers, and one activation layer. The third microscopic data classification sub-model includes one input layer, three output layers, and 25 hidden layers. The 25 hidden layers include 12 convolutional layers, eight pooling layers, four fully connected layers, and one activation layer.

3. The method according to claim 2, characterized in that, The number of pixels in the microscopic image corresponding to the microscopic measurement data is M*N*3, where M represents the number of row pixels, N represents the number of column pixels, and 3 represents the number of channels. Accordingly, the kernel size of the convolutional layer in the first microscopic data classification sub-model is a*a*3, where the first a represents the length of the kernel, the second a represents the width of the kernel, the number of channels of the kernel is 3, and the value of a is any integer between 1 and n, where n is the minimum value of M and N.

4. The method according to claim 3, characterized in that, In the second microscopic data classification sub-model, the size of the convolutional kernel of the convolutional layer is b*1*3, where b represents the length of the convolutional kernel, the width of the convolutional kernel is 1, the number of channels of the convolutional kernel is 3, and the value of b is any integer between 1 and M.

5. The method according to claim 3, characterized in that, In the third microscopic data classification sub-model, the size of the convolutional kernel of the convolutional layer is 1*c*3, the length of the convolutional kernel is 1, c represents the width of the convolutional kernel, the number of channels of the convolutional kernel is 3, and the value of c is any integer between 1 and N.

6. The method according to claim 2, characterized in that, In the first microscopic data classification sub-model, the size of the convolutional kernel in the convolutional layer decreases sequentially from input to output; In the second microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output. In the third microscopic data classification sub-model, the size of the convolutional kernels in the convolutional layers decreases sequentially from input to output.

7. The method according to claim 1, characterized in that, The at least one noise type includes at least one of the following: Device noise, thermal noise, electronic noise, scattering noise, fluorescent dye noise, frosted glass noise, and analog-to-digital conversion noise.

8. A system applied to the microscopic measurement data processing method of claim 1, characterized in that, include: Microscopic measurement apparatus, microscopic data processing platform, and microscopic measurement database, among which, The microscopic measuring device is used to: acquire microscopic measurement data and upload the microscopic measurement data to the microscopic data processing platform; The microscopic data processing platform is used for: The microscopic measurement data are normalized using an adaptive normalization method to obtain the first microscopic measurement data. The noise classification model in the pre-trained noise processing large model is used to classify the noise type and noise quantity in the first microscopic measurement data, so as to obtain at least one noise type and the noise quantity corresponding to each noise type. Based on the at least one noise type and the number of noises corresponding to each noise type, the noise removal model in the large noise processing model is used to remove the noise in the first microscopic measurement data to obtain the denoised second microscopic measurement data. The microscopic measurement database is used to store the second microscopic measurement data.

Images