Learning method, inference model, noise reduction method, noise reduction program, and noise reduction system

Abstract

Even if sufficient training data cannot be obtained, the system will properly remove noise from the data. The learning method is a method for generating an inference model used to remove noise from data, and includes a learning acquisition step (S01) for acquiring training data containing noise, a noise removal step (S02) for generating training data after noise removal from the training data using an inference model in the process of training, a noise addition step (S03) for generating training data with noise added by adding a predetermined noise to the training data after noise removal, and a training step (S04) for training the machine learning of the inference model using combinations of training data and training data with noise added, each as data containing noise.

Description

Technical Field

[0001] The present invention relates to a learning method for generating an inference model used for removing noise from data, the generated inference model, and a noise removal method, a noise removal program, and a noise removal system for removing noise from data using the generated inference model.

Background Art

[0002] Conventionally, a technique for generating an inference model for removing image noise by machine learning has been proposed. One such technique is Noise2Noise (see, for example, Non-Patent Document 1). According to Noise2Noise, an inference model can be generated without using an image that does not contain noise.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Noise2Noise requires multiple images with identical signal components but different noise levels for use in machine learning. However, depending on the target object, acquiring such images may require special processing such as alignment. Alternatively, there are many targets from which such images cannot be obtained. Furthermore, it can be difficult to acquire a sufficient number of images for machine learning. With conventional methods, if a sufficient number of images suitable for machine learning cannot be acquired, it is not possible to generate an appropriate inference model.

[0005] One embodiment of the present invention has been made in view of the above, and aims to provide a learning method, inference model, noise reduction method, noise reduction program, and noise reduction system that can appropriately remove noise from data even when sufficient training data cannot be obtained. [Means for solving the problem]

[0006] To achieve the above objective, a learning method according to one embodiment of the present invention is a learning method for generating an inference model used to remove noise from data having multiple values, and includes: a learning acquisition step of acquiring learning data having multiple values ​​and containing noise; a noise removal step of generating noise-removed learning data from the learning data acquired in the learning acquisition step using an inference model in the process of training; a noise addition step of adding a predetermined noise to the noise-removed learning data generated in the noise removal step to generate noise-added learning data; and a training step of training the machine learning of the inference model using the combination of the learning data acquired in the learning acquisition step and the noise-added learning data generated in the noise addition step as data containing noise, wherein after the inference model is trained in the training step, the noise removal step generates noise-removed learning data using the trained inference model as an inference model in the process of training, and the noise removal step, the noise addition step and the training step are repeated.

[0007] In a learning method according to one embodiment of the present invention, noisy learning data, which is used to train an inference model together with the learning data, is generated from the learning data. The noisy learning data is obtained by removing the noise contained in the learning data using an inference model during training, and then adding a predetermined noise. Therefore, the noise contained in the learning data and the noise contained in the noisy learning data are independent of each other, and the combination of the learning data and the noisy learning data is appropriate for training the inference model. As a result, the learning method according to one embodiment of the present invention can generate an inference model that can appropriately remove noise.

[0008] Furthermore, by using an inference model in the process of training to generate noise-added training data, and adapting it to the training stage each time noise-added training data is generated, new noise-added training data can be obtained according to the training stage. In other words, new combinations of training data and noise-added training data can be obtained from the same training data according to the training stage. Therefore, the learning method according to one embodiment of the present invention does not require a large amount of training data to train the inference model. Thus, according to the learning method according to one embodiment of the present invention, noise can be appropriately removed from the data even when sufficient training data cannot be obtained.

[0009] In the training step, each time the inference model is trained, the denoising step may be used to generate denoised training data using the trained inference model as an in-progress inference model. With this configuration, a new combination of training data and denoised training data can be obtained with each training iteration using the same training data. As a result, an even more appropriate inference model can be generated.

[0010] In one of the iterations, during the training step, the machine learning of the inference model may be trained using the training data acquired in the training data acquisition step, instead of the noisy training data generated in the noise addition step. This configuration allows for a more appropriate range of values ​​for the data output from the generated inference model, and enables more effective noise removal from the data.

[0011] In the noise addition step, noise that is amplified from the noise expected to be removed may be added as a pre-set noise to generate data for noise addition training. With this configuration, noise can be removed from the data more appropriately.

[0012] In the noise addition step, pre-set noise may be added according to the iteration to generate noise-added training data. With this configuration, noise-added training data with various types of noise added according to the iteration can be used to train the inference model. As a result, noise can be removed from the data more appropriately.

[0013] In the noise addition step, noise based on the same type of noise as the noise expected to be removed may be added as a pre-set noise to generate noise-added training data. With this configuration, the noise-added training data, to which the noise expected to be removed has been added, can be used to train the inference model. As a result, if the noise is the type expected to be removed, the noise can be removed from the data more appropriately.

[0014] The inference model generated by the learning method according to one embodiment of the present invention is itself an invention with a novel configuration. That is, the inference model according to one embodiment of the present invention is an inference model that causes a computer to function by taking data having multiple values ​​as input, performing calculations according to the input, and outputting information, and is generated by the learning method described above.

[0015] To achieve the above objective, a noise reduction method according to one embodiment of the present invention is a noise reduction method that removes noise from data having multiple values ​​using an inference model generated by the above learning method, and includes a removal acquisition step of acquiring data to be denoised that has multiple values ​​and is subject to noise reduction, and a noise reduction step of generating denoised data using an inference model from the data to be denoised acquired in the removal acquisition step.

[0016] In the noise reduction method according to one embodiment of the present invention, the above-described inference model is used to perform noise reduction. Therefore, according to the noise reduction method according to one embodiment of the present invention, noise can be appropriately removed from the data.

[0017] The inference model is for removing speckle noise from optical interference images, and in the removal acquisition step, the optical interference image may be acquired as the data to be denoised. With this configuration, speckle noise can be appropriately removed from optical interference images.

[0018] One embodiment of this invention can be described as a noise reduction method as described above, but it can also be described as an invention of a noise reduction program and a noise reduction system as follows. These are substantially the same invention, differing only in category, and produce similar functions and effects.

[0019] In other words, a noise reduction program according to one embodiment of the present invention is a noise reduction program that causes a computer to operate as a noise reduction system that removes noise from data having multiple values ​​using an inference model generated by the learning method described above, wherein the computer functions as a removal acquisition means for acquiring noise reduction target data which has multiple values ​​and is subject to noise reduction, and a noise reduction means for generating noise-reduced data from the noise reduction target data acquired by the removal acquisition means using an inference model.

[0020] A noise removal system according to an embodiment of the present invention is a noise removal system that removes noise from data having a plurality of values using an inference model generated by the above learning method, and includes a removal acquisition unit that acquires noise removal target data having a plurality of values and being a target for noise removal, and a noise removal unit that generates data after noise removal using the inference model from the noise removal target data acquired by the removal acquisition unit.

Advantages of the Invention

[0021] According to an embodiment of the present invention, even when sufficient learning data cannot be acquired, noise can be appropriately removed from the data.

Brief Description of the Drawings

[0022] [Figure 1] It is a diagram showing the configuration of a learning system and a noise removal system according to an embodiment of the present invention. [Figure 2] It is a diagram showing an outline of a learning method according to an embodiment. [Figure 3] It is a graph showing data used to explain the reason why the method according to the embodiment holds. [Figure 4] It is a graph showing an example of noise removal according to input data input to an inference model when an inference model is generated by Noise2Noise (N2N) and noise is removed. [Figure 5] It is a graph showing an example of noise removal according to teacher data for comparison with the output from an inference model when an inference model is generated by N2N and noise is removed. [Figure 6] It is a graph showing an example of noise removal according to noise included in input data when an inference model is generated by N2N and noise is removed. [Figure 7] It is a graph showing an example of noise removal according to noise included in input data when an inference model is generated by N2N and noise is removed. [Figure 8]This graph shows an example of noise reduction based on the noise contained in the input data when an inference model is generated using N2N and noise reduction is performed. [Figure 9] This flowchart shows a learning method, which is a process performed in a learning system according to an embodiment of the present invention. [Figure 10] This flowchart shows a noise reduction method, which is a process performed by the noise reduction system according to an embodiment of the present invention. [Figure 11] This is an example of data (images) for each iteration (training cycle) of machine learning used to generate an inference model. [Figure 12] This is an example of noise reduction performed by the embodiment. [Figure 13] This is an example of noise reduction performed by the embodiment. [Figure 14] This is an example of noise reduction performed by the embodiment. [Figure 15] This figure shows the configuration of a learning program and a noise reduction program according to an embodiment of the present invention, along with a recording medium. [Modes for carrying out the invention]

[0023] Hereinafter, embodiments of the learning method, inference model, noise reduction method, noise reduction program, and noise reduction system according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and redundant explanations are omitted.

[0024] Figure 1(a) shows a learning system 10 that performs the learning method according to this embodiment. Figure 1(b) shows a noise reduction system 20 that performs the noise reduction method according to this embodiment. The learning system 10 is a system (device) that generates an inference model (learning model) used to remove noise from data having multiple values. The noise reduction system 20 is a system (device) that removes noise from data having multiple values ​​using the inference model generated by the learning system 10.

[0025] In this embodiment, the data having multiple values ​​to be denoised is, for example, an image. The image to be denoised may be, for example, an optical interference image (data). An optical interference image is, for example, an image obtained by optical coherence tomography (OCT) or optical diffraction tomography (ODT). The image to be denoised may be a fluorescence image obtained by imaging with a digital CMOS (Complementary Metal Oxide Semiconductor) camera. The image to be denoised may be an image used for emission analysis (for example, an emission image obtained by imaging with an InGaAs camera).

[0026] The images to be denoised are, for example, images of substrates manufactured on a production line, and the denoised images may be used to check for defects in the substrates. The images to be denoised are also images of cells, and the denoised images may be used to analyze the cells. Furthermore, any image containing noise can be used to denoise, other than those mentioned above.

[0027] Furthermore, data containing multiple values ​​to be removed from noise does not necessarily have to be an image; for example, it may be spectral data or time-series data. Spectral data is, for example, intensity data against wavenumber (wavelength, frequency). Time-series data is, for example, intensity data for each time period, and specifically, it may be output data (output signals) from detectors such as photomultiplier tubes or point sensors, or measurement data of biological information such as hemoglobin concentration.

[0028] The noise contained in the data may be uncorrelated or correlated between multiple values ​​that make up the data (for example, between pixels that make up an image). Uncorrelated noise is, for example, noise that arises randomly from the noise distribution. For example, if the data is acquired from a point sensor, the next noise value cannot be predicted from only one point in the time series. Or, if the data is acquired from a line or 2D sensor, the noise value of adjacent pixels cannot be predicted. Such noise may be, for example, thermal noise, shot noise, or pixel defects or misalignment.

[0029] Correlated noise is noise that is not randomly generated but is related to other noises. For example, a portion of the noise that is extracted may be correlated with noise in other areas. For instance, if the data is acquired from a point sensor, the continuity of the noise's time series can predict the next noise value. Alternatively, if the data is acquired from a line or 2D sensor, the noise value of adjacent pixels can be predicted. Such noises include, for example, periodic noise, speckle noise, or unwanted signal components.

[0030] The noise contained in the data may include both uncorrelated noise and correlated noise as described above. Furthermore, the noise contained in the data may be of a type other than those described above. Also, if the noise's brightness varies depending on the object (i.e., the distribution differs), it can be subjected to noise reduction if scaling can be used to create a consistent distribution of the noise.

[0031] The learning system 10 and the noise reduction system 20 are comprised of a conventional computer including hardware such as a processor, memory, and communication modules. The processor is, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0032] Furthermore, the learning system 10 and the noise reduction system 20 may include imaging devices such as cameras for acquiring images used in processing. Alternatively, the learning system 10 and the noise reduction system 20 may be included in imaging devices such as cameras for acquiring images used in processing. In Figure 1, the learning system 10 and the noise reduction system 20 are shown as separate systems (devices), but they may be implemented by the same system (device).

[0033] The computers comprising the learning system 10 and the noise reduction system 20 may be a computer system including multiple computers. Furthermore, the computers may be configured using cloud computing or edge computing. The functions of the learning system 10 and the noise reduction system 20, as described later, are performed by these components operating through programs or the like.

[0034] Next, the functions of the learning system 10 and noise reduction system 20 according to this embodiment will be described. As shown in Figure 1, the learning system 10 is configured to include a learning acquisition unit 11, a noise reduction unit 12, a noise addition unit 13, and a training unit 14.

[0035] The learning system 10 generates an inference model through machine learning training. The inference model is composed of, for example, a neural network. The neural network may be multi-layered. That is, the inference model may be generated by deep learning. The neural network may also be a convolutional neural network (CNN). The form of the inference model may be the same as conventional inference models generated by N2N, for example, a 6-layer Conv2D+ReLU.

[0036] An inference model is used to remove noise from data that has multiple values. For example, an inference model takes data to be denoised as input and outputs denoised data. In this case, the inference model has neurons in its input layer to receive the data to be denoised. For example, the information input to the inference model is the pixel value (luminance value) of each pixel in an image, which is the data to be denoised. In this case, the input layer has neurons equal to the number of pixels in the image, and each neuron is input with the pixel value of the corresponding pixel.

[0037] Furthermore, the inference model includes neurons in the output layer to output denoised data. For example, the information output from the inference model is the pixel value of each pixel in the image, which is the denoised data. In this case, the output layer has as many neurons as there are pixels in the image, and the pixel value of the corresponding pixel is output from each neuron.

[0038] Alternatively, the inference model may take data to be denoised as input and output data containing the noise present in that data. In this case, the denoised data can be obtained by removing the output noise from the data to be denoised. Furthermore, the inference model may be any other type of model that removes noise from the data to be denoised.

[0039] An inference model is designed to allow a computer to take data with multiple values ​​as input, perform calculations based on the input, and output information. Inference models are intended to be used as program modules, which are part of artificial intelligence software. For example, an inference model is used in a computer equipped with a processor and memory, where the computer's processor operates according to instructions from the model stored in memory. For instance, the computer's processor operates according to these instructions, inputting information into the model, performing calculations appropriate to the model, and outputting results from the model. Specifically, the computer's processor operates according to these instructions, inputting information into the input layer of a neural network, performing calculations based on parameters such as learning weighting coefficients in the neural network, and outputting results from the output layer of the neural network. Note that an inference model may be composed of something other than a neural network.

[0040] Figure 2 shows an overview of the learning method for generating the inference model according to this embodiment. In this embodiment, noisy data is used to generate the inference model, and it is not necessary to use noise-free data. The learning method in this embodiment, described below, is called Noise2Prior, in contrast to the conventional Noise2Noise method.

[0041] The learning data acquisition unit 11 is a learning data acquisition means that acquires learning data having multiple values ​​and containing noise. The learning data acquisition unit 11 acquires learning data as follows, for example.

[0042] The training data is in the same format as the data to be denoised. For example, if the data to be denoised is an image, the training data is an image of the same size as that image. The training acquisition unit 11 acquires an image that contains noise and is larger in size than the training data as the training source image (source data). An image containing noise is, for example, an image obtained by imaging a pre-prepared object with an imaging device while the captured image is in a state where noise is present. The source image is acquired, for example, by receiving it from the imaging device that captured the source image, or by accepting a source image input operation from the user to the training system 10.

[0043] The training data acquisition unit 11 extracts and acquires images from the original image that are smaller than the original image and of a predetermined size as training data. The training data acquisition unit 11 may, for example, acquire multiple training data at randomly different positions from a single original image. Alternatively, as shown in Figure 2, the training data acquisition unit 11 may, for example, acquire training data 30 (images) from multiple original images (original image pool).

[0044] The learning data acquisition unit 11 may acquire learning data by methods other than cropping from the original image as described above. The number of learning data acquired may be one or multiple. The learning data acquisition unit 11 outputs the acquired learning data to the noise reduction unit 12 and the training unit 14.

[0045] The noise reduction unit 12 is a noise reduction means that generates noise-reduced training data from training data acquired by the training data acquisition unit 11 using an inference model in the process of training. After the inference model has been trained by the training unit 14, the noise reduction unit 12 generates noise-reduced training data using the trained inference model as an inference model in the process of training. The noise reduction unit 12 may generate noise-reduced training data using the trained inference model as an inference model in the process of training each time the inference model is trained by the training unit 14.

[0046] The noise reduction unit 12 generates training data after noise reduction, for example, as follows. The generation of the inference model is performed by repeatedly processing by the noise reduction unit 12 (noise reduction step), processing by the noise addition unit 13 (noise addition step), and processing by the training unit 14 (training step).

[0047] As described later, the training unit 14 trains the machine learning of the inference model. The training unit 14 outputs the inference model after machine learning training, i.e., the inference model in the middle of training, to the denoising unit 12 during the above iteration. The denoising unit 12 receives the inference model from the training unit 14 and stores it. The output of the inference model from the training unit 14 to the denoising unit 12 may be performed, for example, by sharing the inference model or copying the inference model. The output of the inference model from the training unit 14 to the denoising unit 12 may be performed each time the inference model is trained. Alternatively, the output may be performed not each time the inference model is trained, but at a predetermined timing in the above iteration.

[0048] The noise reduction unit 12 receives training data from the training data acquisition unit 11. The noise reduction unit 12 generates denoised training data from the input training data using an inference model that is in the process of being trained. For example, if the inference model outputs a denoised image, as shown in Figure 2, the noise reduction unit 12 inputs the training data 30 (image) into the inference model and obtains the output of denoised training data 31 (image) from the inference model.

[0049] For example, if the inference model outputs noise, the noise reduction unit 12 removes the noise obtained using the inference model from the training data to obtain denoised training data. Alternatively, the noise reduction unit 12 may generate denoised training data using methods other than those described above, as long as they utilize an inference model.

[0050] While the inference model in the training phase may not have sufficient noise reduction capabilities, it can still be used for noise reduction. Therefore, the denoised training data generated by the noise reduction unit 12 will have less noise than the original training data. The noise reduction unit 12 outputs the generated denoised training data to the noise addition unit 13.

[0051] The noise addition unit 13 is a noise addition means that adds pre-set noise to the training data after noise removal generated by the noise removal unit 12 to generate noise-added training data. The noise addition unit 13 may generate noise-added training data by adding noise that is an amplified version of the noise expected to be removed as pre-set noise. The noise addition unit 13 may generate noise-added training data by adding pre-set noise according to the repetition. The noise addition unit 13 may generate noise-added training data by adding noise based on the same type of noise as the noise expected to be removed as pre-set noise.

[0052] The noise addition unit 13 generates noise-added training data, for example, as follows. The noise-added training data generated by the noise addition unit 13 is used for training the machine learning of the inference model by the training unit 14. The noise addition unit 13 receives the denoised training data from the denoising unit 12. The noise addition unit 13 adds (combines) pre-set noise to the input denoised training data to generate noise-added training data.

[0053] For example, the noise addition unit 13 stores noise data in advance to create noise-added learning data. For example, the noise addition unit 13 acquires an image containing only noise as the original noise image (original data). An image containing only noise is, for example, an image obtained by imaging for noise generation. Imaging for noise generation is imaging by an imaging device in a state where noise is generated in the captured image and there are no objects to be captured. Alternatively, an image containing only noise may be mechanically generated by simulation or the like. The acquisition of the original noise image is performed, for example, by receiving it from the imaging device that captured the original image, or by accepting the user's input operation of the original image to the learning system 10.

[0054] If imaging to generate noise data is performed in the same way as imaging to generate data to be denoised, except for the presence or absence of objects being imaged, then the generated noise data will be of the same type as the noise intended to be removed. This is because if similar imaging is performed, the data is expected to contain the same type of noise. The same applies even if the noise data is generated by means other than imaging. Furthermore, even when noise data is generated by simulation, by performing a simulation that simulates the generation of data to be denoised, the noise data can be of the same type as the noise intended to be removed. Thus, the noise data used for noise-added learning may be of the same type as the noise intended to be removed. However, the noise data used for noise-added learning does not necessarily have to be of the same type as the noise intended to be removed.

[0055] The noise addition unit 13 extracts and acquires an image from the original image that is the same size as the training data, as noise data to be used to generate noise-added training data. The noise addition unit 13 may, for example, acquire noise data at random positions from a single original image each time it is acquired. Alternatively, as shown in Figure 2, the noise addition unit 13 may acquire noise data 32 from multiple original images (noise pool to be removed), for example.

[0056] The noise addition unit 13 generates noise-added training data by adding the pixel value of the denoised training data 31 and the pixel value of the noise data 32 for each pixel at the same position.

[0057] The noise addition unit 13 may generate noise-added training data by adding amplified noise, obtained by amplifying the noise shown in the acquired data, to the training data after noise removal. For example, as shown in Figure 2, the noise addition unit 13 generates noise-added training data 33 (image) by multiplying the noise data 32 (image) by an adjustment variable and adding it to the training data 31 (image) after noise removal.

[0058] The adjustment variable is a value pre-set and stored in the noise addition unit 13 for amplifying noise. The adjustment variable is a value greater than 1 (specifically, 2, 3, or 4). For example, the noise addition unit 13 multiplies the pixel value of each pixel in the noise data 32 (image) by the adjustment variable. The noise addition unit 13 then adds the pixel values ​​of the noise data 32 (image) multiplied by the adjustment variable to the pixel values ​​of the training data 31 (image) after noise removal to generate noise-added training data 33 (image).

[0059] The noise addition unit 13 may generate noise-added training data by adding noise according to the repetition. For example, the noise addition unit 13 may generate noise-added training data by obtaining different noise data from the noise pool to be removed each time noise-added training data is generated.

[0060] Furthermore, the noise addition unit 13 may generate noise-added training data other than that described above by adding a predetermined noise to the training data after noise removal generated by the noise removal unit 12, so as long as it generates noise-added training data that can be used to train the machine learning of the inference model described later. The noise addition unit 13 outputs the generated noise-added training data to the training unit 14.

[0061] The training unit 14 is a training means that trains the machine learning of an inference model using a combination of training data acquired by the training data acquisition unit 11 and noise-added training data generated by the noise-adding unit 13, each of which is noise-containing data. In any of the above iterations, the training unit 14 may train the machine learning of the inference model using training data acquired by the training data acquisition unit 11 instead of noise-added training data generated by the noise-adding unit 13.

[0062] The training unit 14 performs machine learning training of an inference model, for example, as follows: The training unit 14 receives training data from the training data acquisition unit 11. The training unit 14 receives noise-added training data based on the training data from the noise-adding unit 13. The training unit 14 trains the inference model using a combination of this training data and noise-added training data.

[0063] The training data input from the training data acquisition unit 11 and the noise-added training data input from the noise-adding unit 13 are identical data (for example, the same image) with different noises added to each other. The training unit 14 uses these combinations as training data to train the inference model using machine learning. For example, the training unit 14 trains the inference model using N2N as shown in Non-Patent Literature 1.

[0064] In this case, for example, as shown in Figure 2, the training unit 14 inputs the noise-added training data 33 (image) into the inference model and obtains noise-removed data 34 (image) from the inference model at that point in time. The training unit 14 compares the noise-removed data 34 with the training data 30 (image) and updates the parameters of the inference model by backpropagation based on the loss based on the comparison. In the above example, the training is performed by using the noise-added training data 33 as input to the inference model and comparing the training data 30 with the output of the inference model. However, the training may also be performed by using the training data 30 as input to the inference model and comparing the noise-added training data 33 with the output of the inference model.

[0065] The training unit 14 trains the inference model, i.e., updates the parameters of the inference model, and then outputs the trained inference model to the denoising unit 12. The denoising unit 12 receives the inference model from the training unit 14 and uses the inference model to generate denoised training data as described above. The training data used to generate the denoised training data may be data that has already been used to train the inference model, or it may be data that has not yet been used to train the inference model. The denoised training data generated by the denoising unit 12 is used in the same way as described above. The training unit 14 uses the combination of training data and noisy training data to train the machine learning model. This process is repeated, and the training of the inference model progresses.

[0066] In any of the above iterations, the training unit 14 may train the inference model using training data instead of noise-added training data. For example, as shown in Figure 2, the training unit 14 selects either training data 30 (images) or noise-added training data 33 (images) as the data to be used for training, and trains the inference model using the selected data. If noise-added training data 33 is selected, the training unit 14 trains the inference model as described above.

[0067] If training data 30 is selected, the training unit 14 will train the inference model using only one training data 30. In this case as well, the training unit 14 can train the inference model in the same way as when using noisy training data 33 (for example, N2N in the above example).

[0068] For example, the training unit 14 inputs the training data 30 into the inference model and obtains data with noise removed by the inference model at that point in time. The training unit 14 compares the noise-removed data with the training data 30 and updates the parameters of the inference model by backpropagation based on the loss calculated from the comparison.

[0069] The noise-added training data described above is the training data from which noise has been removed by an inference model during training. Because the noise removal function of the inference model during training is not sufficient, the noise-added training data may have had a significant amount of non-noise components removed as well. Therefore, if an inference model is trained using this noise-added training data, the inference model may not be able to output appropriate data. For example, the range of data values ​​output by the inference model may deviate significantly from the range of the noise-removed data.

[0070] Although the training data contains noise, the range of data values ​​does not deviate significantly from that of data with proper noise removal. Therefore, by training the inference model using only the training data as described above, it is possible to prevent the range of data values ​​output by the inference model from deviating significantly from the normal range.

[0071] The choice between using noisy training data or using training data instead of noisy training data may be made probabilistically, for example. Alternatively, training data may be used instead of noisy training data for a predetermined number of iterations, and noisy training data may be used in all other cases. Training an inference model using training data instead of noisy training data is, as described above, to bring the output values ​​of the inference model within a specified range. Therefore, the number of such training iterations may be less than the number of training iterations using noisy training data for noise reduction, provided that this objective is achieved. For example, the number of such training iterations may be about 10% of the total number of iterations.

[0072] The training of the inference model described above is repeated, similar to conventional machine learning training, for example, until a predetermined number of repetitions or until the generation of the inference model converges based on predetermined conditions. Once the training of the inference model is completed, the training unit 14 outputs the generated inference model to the noise reduction system 20. This completes the functions of the learning system 10.

[0073] As shown in Figure 1, the noise reduction system 20 is configured to include a noise acquisition unit 21 and a noise reduction unit 22.

[0074] The noise removal acquisition unit 21 is a noise removal acquisition means that acquires noise removal target data which has multiple values ​​and is subject to noise removal. The noise removal acquisition unit 21 acquires noise removal target data as follows, for example.

[0075] The noise reduction acquisition unit 21 acquires noise reduction target data by receiving it from the imaging device that captured the image which is the noise reduction target data, or by accepting the user's image input operation to the noise reduction system 20. If the size of the noise reduction target data input to the inference model is smaller than the original image obtained by imaging by the imaging device, similar to the training data described above, the noise reduction acquisition unit 21 may acquire the original image, divide the original image into sizes corresponding to the noise reduction target data input to the inference model, and use the divided images as the noise reduction target data.

[0076] Furthermore, the noise reduction acquisition unit 21 may acquire an optical interference image as data to be denoised. The noise reduction acquisition unit 21 may also acquire data to be denoised by methods other than those described above. The noise reduction acquisition unit 21 outputs the acquired data to be denoised to the noise reduction unit 22.

[0077] The noise reduction unit 22 is a noise reduction means that generates noise-reduced data using an inference model from the noise reduction target data acquired by the noise reduction acquisition unit 21. The noise reduction unit 22 takes the inference model generated by the learning system 10 as input and stores it, and uses it for noise reduction.

[0078] The noise reduction unit 22 receives the data to be denoised from the noise reduction acquisition unit 21. The noise reduction unit 22 inputs the received data to be denoised into the inference model to generate denoised data. For example, if the inference model outputs denoised data, the noise reduction unit 22 acquires the output from the inference model as denoised data. Alternatively, if the inference model outputs noise, the noise reduction unit 22 removes the noise obtained using the inference model from the data to be denoised to obtain denoised data. The noise reduction unit 22 may also generate denoised data by methods other than those described above, as long as they utilize an inference model.

[0079] If the data to be denoised is an optical interference image, the inference model may be one for removing speckle noise from the optical interference image. The optical interference image may contain speckle noise as the noise to be removed; in this case, the denoising unit 22 removes the speckle noise from the optical interference image using the inference model.

[0080] The noise reduction unit 22 outputs the generated noise-reduced data. The output of the noise-reduced data can be done in the same way as conventional methods, depending on the purpose of using the data. The above describes the functions of the noise reduction system 20.

[0081] Next, we will explain why the method according to this embodiment is valid. As mentioned above, the noise contained in the data includes noise that is not correlated with each other among the multiple values ​​that make up the data, i.e., noise that is independent, and noise that is correlated with each other among the multiple values, i.e., noise that is not independent.

[0082] Figure 3(a) shows an example of data containing independent noise. Figure 3(b) shows an example of data containing non-independent noise. In the graph in Figure 3, the horizontal axis represents the data step (order, position), and the vertical axis represents the data value. The data shown in Figure 3 consists of 100,000 values. That is, the signal length is 100,000. As a High-Low condition, the data takes either a value of 0.625 or 0.375 with a probability of 1 / 2 every 64 steps, and a noise value is also assigned to each step.

[0083] Figure 3(a) shows the signal with Gaussian noise (σ=0.1) added as noise. Figure 3(b) shows the signal with two different phase sine waves added as noise. In the graph in Figure 3, the solid line represents the signal with noise, and the dashed line represents the signal without noise.

[0084] Figure 4 shows an example of generating an inference model using N2N based on independent noise-containing data and then removing the noise. Figure 4(a) shows a graph of the input data (signal + noise data) input to the inference model when training machine learning using N2N.

[0085] Figure 4(b) shows a graph of an example where noise reduction was performed using the same data as the input data (referred to here as training data) for comparison with the output from the inference model when training N2N machine learning. In the graph in Figure 4(b), the dashed line shows the signal portion (signal without noise) of the data before noise reduction, and the solid line shows the data after noise reduction (inference result) (the same applies to the following graphs for inference results). Figure 4(c) shows a graph of an example where noise reduction was performed using different data as training data for comparison with the output from the inference model when training N2N machine learning. In this case, although the conditions for adding noise are the same for the input data and the training data, the noise values ​​contained in the data are different from each other.

[0086] In the example in Figure 4(b), the generated inference model is merely an identity map, and the noise is not properly removed. In the example in Figure 4(c), the noise is properly removed. Thus, with N2N machine learning, if the noise contained in the input data and the training data are different from each other, proper noise removal becomes possible.

[0087] Figure 4(d) shows a graph of an example where noise reduction was performed when training a machine learning model using N2N, using data obtained by adding the training data used in Figure 4(b) and the training data used in Figure 4(c) as training data to compare with the output from the inference model. The training data was added at each data step. Figure 4(e) shows a graph of an example where noise reduction was performed when training a machine learning model using N2N, using data obtained by averaging the training data used in Figure 4(b) and the training data used in Figure 4(c) as training data to compare with the output from the inference model. The training data was averaged at each data step.

[0088] In the examples in Figures 4(d) and 4(e), the noise is not properly removed in either case. This indicates that even if the noise contained in the input data and the training data is different, the accuracy of noise reduction decreases if the same noise components are present. To improve the accuracy of noise reduction, it is necessary to make the noise contained in the training data completely different from the noise contained in the input data.

[0089] Figure 5 shows an example of generating an inference model using N2N and removing noise based on independent noise-containing data. Figure 5(a) shows a graph of the training data (signal + noise data) compared with the output from the inference model when training machine learning with N2N.

[0090] Figure 5(b) shows a graph of an example where noise reduction was performed using the same input data as the training data when training a machine learning model with N2N. Figure 5(c) shows a graph of an example where noise reduction was performed using different input data than the training data when training a machine learning model with N2N. In this case, although the conditions for adding noise are the same for the input data and the training data, the noise values ​​contained in the data are different from those of the training data.

[0091] In the example in Figure 5(b), the generated inference model is merely an identity map, and the noise is not properly removed. In the example in Figure 5(c), the noise is properly removed. Thus, with N2N machine learning, if the noise contained in the input data and the training data are different from each other, proper noise removal becomes possible.

[0092] Figure 5(d) shows a graph of an example where noise reduction was performed when training a machine learning model using N2N, using data obtained by adding the input data used in Figure 5(b) and the input data used in Figure 5(c) as input data for the inference model. The input data was added at each data step. Figure 5(e) shows a graph of an example where noise reduction was performed when training a machine learning model using N2N, using data obtained by averaging the input data used in Figure 5(b) and the input data used in Figure 5(c) as input data for the inference model. The training data was averaged at each data step.

[0093] In the examples in Figures 5(d) and 5(e), the noise is not properly removed in either case. This indicates that even if the noise contained in the input data and the training data are different, the accuracy of noise reduction decreases if the same noise components are present. To improve the accuracy of noise reduction, it is necessary to make the noise contained in the training data completely different from the noise contained in the input data.

[0094] In this embodiment, the data used to train the inference model consists of training data and noise-added training data. As described above, the noise-added training data is obtained by denoising the training data and adding other noise. Therefore, the training data and the noise-added training data are designed so that they do not contain common noise and contain different noises from each other. Accordingly, this embodiment enables appropriate noise removal.

[0095] Next, we will explain the learning effect due to differences in noise contained in the input data fed into the inference model when training machine learning using N2N. Figure 6 shows an example in which an inference model was generated using N2N based on data containing independent noise, and then noise was removed. Figure 6(a) shows a graph of the data (signal + noise data) used as input data for the inference model when training machine learning using N2N.

[0096] Here, we show an example of generating an inference model and removing noise using data with the noise multiplied by a constant (data with the noise multiplied by a constant) as input data, as shown in Figure 6(a). Figure 6(b) shows a graph of an example where noise was removed by generating an inference model using input data with the noise multiplied by 1. Figure 6(c) shows a graph of an example where noise was removed by generating an inference model using input data with the noise multiplied by 2. Figure 6(d) shows a graph of an example where noise was removed by generating an inference model using input data with the noise multiplied by 3. Figure 6(e) shows a graph of an example where noise was removed by generating an inference model using input data with the noise multiplied by 4.

[0097] As shown in Figures 6(b) to (e), the accuracy of noise reduction (noise reduction effect) is higher when the noise is amplified (to 2 to 4 times) compared to when the noise is 1 times. In this way, the accuracy of noise reduction can be increased by multiplying the noise contained in the input data by a constant. The effect of noise reduction varies depending on the magnitude of the noise being added, and there is an appropriate value. If the noise being added is too large, the contrast of the signal will decrease.

[0098] In the example shown in Figure 6, input data and training data were prepared separately, and the noise in the input data was amplified. However, Figure 7 shows an example in which, as in this embodiment, noise reduction is performed on the training data using an inference model in the process of training, and data with pre-prepared noise multiplied by a constant is used as input data. Figures 7(a) to 7(d) are graphs when input data and training data are prepared separately, and the noise in the input data is multiplied by 1, 2, 3, and 4, respectively (corresponding to Figures 6(b) to 7(e)). Figures 7(e) to 7(h) are graphs when noise reduction is performed on the training data using an inference model in the process of training, and data with pre-prepared noise multiplied by 1, 2, 3, and 4 is used as input data (the multipliers correspond vertically).

[0099] The above corresponds to generating noise-added learning data by amplifying noise using adjustment variables in this embodiment. This configuration allows for higher accuracy in noise reduction.

[0100] Figure 8 shows an example of generating an inference model using N2N based on data containing non-independent noise, and then removing the noise. This example, like Figures 6 and 7, is one in which the noise in the input data is multiplied by a constant. Figure 8(a) shows a graph of the data (signal + noise data) used as input data for the inference model when training the machine learning model using N2N.

[0101] Figures 8(b) to 8(d) are graphs showing the results when input data and training data are prepared, and the noise in the input data is multiplied by 1, 2, and 3, respectively. Figures 8(e) to 8(g) are graphs showing the results when noise reduction is performed on the training data using an inference model in the process, and the data with pre-prepared noise multiplied by 1, 2, and 3 is used as input data (the multipliers correspond vertically).

[0102] Even when the data contains noise that lacks independence, the accuracy of noise reduction can be improved by amplifying the noise contained in both the input data and the training data by an appropriate multiplier. This is why the method according to this embodiment is valid.

[0103] Next, the processes performed by the learning system 10 and noise reduction system 20 according to this embodiment (the operation methods performed by the learning system 10 and noise reduction system 20) will be explained using the flowcharts in Figures 9 and 10.

[0104] First, the learning method, which is the process performed by the learning system 10 according to this embodiment, will be explained using the flowchart in Figure 9. In this process, the learning acquisition unit 11 acquires learning data that has multiple values ​​and contains noise (S01, learning acquisition step). Next, the noise removal unit 12 uses an inference model in the process of training to generate learning data after noise removal from the learning data (S02, noise removal step). Next, the noise addition unit 13 adds a predetermined noise to the learning data after noise removal to generate noisy learning data (S03, noise addition step). Next, the training unit 14 uses the combination of the learning data and the noisy learning data as data containing noise to train the machine learning of the inference model (S04, training step).

[0105] Next, the training unit 14 determines whether or not to terminate the training of the inference model (S05). If it is determined not to terminate the training of the inference model (NO in S05), the processes of generating denoised training data (S02), generating noisy training data (S03), and training the machine learning of the inference model (S04) are repeated. In this case, the inference model trained by the training unit 14 is used to generate the denoised training data. After the machine learning training (S04), the decision of whether or not to terminate the training of the inference model (S05) is made again.

[0106] In the decision to terminate the training of the inference model (S05), if it is decided to terminate the training of the inference model (YES in S05), the inference model generated by the training is output from the learning system 10 to the noise reduction system 20 (S06). The noise reduction system 20 stores the inference model and uses it in the following noise reduction process. The above is the learning method which is the process executed by the learning system 10 according to this embodiment.

[0107] Next, the noise reduction method, which is the process performed by the noise reduction system 20 according to this embodiment, will be explained using the flowchart in Figure 10. In this process, the noise reduction acquisition unit 21 acquires noise reduction target data which has multiple values ​​and is the target of noise reduction (S11, noise reduction acquisition step). Subsequently, the noise reduction unit 22 generates noise reduction data from the noise reduction target data using an inference model (S12, noise reduction step). The noise reduction data is output from the noise reduction unit 22 to a predetermined output destination (S13). The above is the noise reduction method, which is the process performed by the noise reduction system 20 according to this embodiment.

[0108] In this embodiment, noisy training data is generated from the training data, which is used together with the training data to train the inference model. The noisy training data is obtained by removing the noise contained in the training data using the inference model during training, and then adding a predetermined amount of noise. Therefore, the noise contained in the training data and the noise contained in the noisy training data are independent of each other, and the combination of training data and noisy training data is appropriate for training the inference model as described above. As a result, in this embodiment, it is possible to generate an inference model that can appropriately remove noise.

[0109] Furthermore, by updating the inference model used to generate noise-added training data according to the training stage each time noise-added training data is generated, new noise-added training data can be obtained according to the training stage. In other words, new combinations of training data and noise-added training data can be obtained from the same training data according to the training stage. Thus, in this embodiment, sequential update learning can be performed in which the data used for training is updated sequentially according to the training of the inference model.

[0110] Therefore, in this embodiment, a large amount of training data is not required to train the inference model. For example, it does not require multiple images with the same signal components but different noise levels, i.e., teacher pairs, as in N2N. Thus, according to this embodiment, even when sufficient training data cannot be obtained, noise can be appropriately removed from the data.

[0111] Since signal and noise components are mixed, in training conventional inference models such as N2N, the only way to increase the number of different mixes (variations) is to increase the original dataset prepared in advance. For example, if periodic noise is mixed with a signal, and a combination with a different phase is required, new data must be prepared. Moreover, whether the required phase is obtained is a matter of chance. In this embodiment, as described above, new combinations of training data and noise-added training data can be obtained according to the training stage, so even with less data prepared in advance for machine learning training than in conventional methods, Semi-Supervised Learning, which is the training of an appropriate inference model, can be achieved.

[0112] In this embodiment, an inference model can be generated even when sufficient training data cannot be obtained; for example, an inference model can be generated from a single image. Therefore, noise reduction becomes possible even for images of objects that cannot be fixed (living organisms), images of objects that deform (2D images using line sensors and point sensors), or images acquired in the past (images that were not consciously prepared), which was previously impossible. Furthermore, according to this embodiment, an appropriate inference model can be generated with fewer training iterations compared to conventional methods.

[0113] Furthermore, as in this embodiment, in the denoising step, the denoising model may be used as the inference model in progress each time the inference model is trained in the training step to generate denoising training data. With this configuration, a new combination of training data and noisy training data can be obtained for the same training data with each iteration of training. As a result, an inference model can be generated more appropriately. However, the inference model used to generate the denoising training data does not necessarily have to be the one used each time the inference model is trained in the training step; it may be the one used after the inference model has been trained in the training step.

[0114] Furthermore, as in this embodiment, in one of the iterations, the machine learning training of the inference model may be performed using the training data acquired in the training data acquisition step instead of the noise-added training data generated in the noise-adding step during the training step. This configuration makes it possible to make the range of values ​​of the data output from the generated inference model more appropriate and to remove noise from the data more effectively. However, if the range of values ​​of the data output from the inference model becomes appropriate even when the inference model is trained using only a combination of training data and noise-added training data, it is not necessarily required to use training data instead of noise-added training data in the training step.

[0115] Furthermore, as in this embodiment, in the noise addition step, noise that is amplified from the noise expected to be removed may be added as a pre-set noise to generate noise addition learning data. For example, as described above, a tuning variable may be multiplied to the pre-prepared noise. With this configuration, as explained with reference to Figures 6 to 8, noise can be removed from the data more appropriately.

[0116] Furthermore, as in this embodiment, in the noise addition step, noise may be added according to a predetermined set noise for each iteration to generate noise-added training data. For example, as described above, different noise may be obtained from the noise pool to be removed for each iteration and added. With this configuration, noise-added training data with various noises added according to the iteration can be used to train the inference model. As a result, noise can be removed from the data more appropriately.

[0117] Furthermore, as in this embodiment, in the noise addition step, noise based on the same type of noise as the noise expected to be removed may be added as a pre-set noise to generate noise-added training data. With this configuration, noise-added training data with the expected noise to be removed can be used to train the inference model. As a result, if the noise is the type expected to be removed, the noise can be removed from the data more appropriately.

[0118] As in this embodiment, the inference model generated in this embodiment may be used to remove noise from the data. As described above, the inference model generated in this embodiment can appropriately remove noise, and with this configuration, noise can be appropriately removed from the data.

[0119] The inference model is for removing speckle noise from optical interference images, and in the removal acquisition step, the optical interference image may be acquired as the data to be denoised. With this configuration, speckle noise can be appropriately removed from optical interference images.

[0120] Next, an example of this embodiment will be shown. Figure 11 shows an example of data (images) for each number of iterations (training iterations) of machine learning training when generating an inference model. Figure 11 shows images of the training data, the training data after denoising, the training data with noise added, and the denoising result for the training data with noise added. The training data after denoising, the training data with noise added, and the denoising result for the training data with noise added were generated from the training data using the inference model for each iteration of machine learning training. Figure 11 shows examples for the 1st, 2nd, 4th, 20th, and 1000th iterations of machine learning training.

[0121] The denoised training data is generated by the denoising unit 12 using an inference model from the training data. The noisy training data is the denoised training data to which noise has been added by the noise adding unit 13. The denoising result for the noisy training data is generated during training by the training unit 14 using an inference model from the noisy training data.

[0122] Figure 12 shows an example of noise reduction performed by this embodiment. Figure 12 shows the data to be denoised (original image) and examples of noise reduction results using various methods on the data to be denoised. The image shown in Figure 12 is a fluorescence image obtained by imaging with a digital CMOS camera. The noise reduction results are those of this embodiment (proposed method), those using Noise2Noise (comparative example), and those using Noise2Void (comparative example). The upper image in Figure 12 is the entire image, and the lower image is a magnified portion of the entire image with an R1 value.

[0123] In the example shown in Figure 12, 32 images of size 192 x 192 were extracted from a 2304 x 2304 image and used as training data to generate the inference model according to this embodiment. The number of iterations for training the machine learning model according to this embodiment was 1500. Note that the images used for training the machine learning model are different from the images shown in Figure 12. In the Noise2Noise (comparative example), two images with different noise content were prepared and machine learning was trained. In the Noise2Void (comparative example), machine learning was trained using a single image.

[0124] Figure 12 shows the standard deviation (SD) of pixel values ​​in a portion R2 of the upper image. This portion R2 is the area where the object being imaged is not captured. Therefore, a smaller standard deviation of pixel values ​​in portion R2 indicates less noise.

[0125] Figures 13 and 14 show examples of noise reduction performed by this embodiment. Figures 13 and 14 show the data to be denoised (original image) and examples of noise reduction results using various methods on the data to be denoised. Figures 13 and 14 are emission images obtained by imaging with an InGaAs camera. Imaging with the InGaAs camera was performed with a 10-second exposure and a 20x objective lens. The bias was 3.0V and 8.79mA. The examples shown in Figures 13 and 14 are examples of background noise reduction. The noise reduction results are obtained by averaging four images (comparative example), Noise2Noise (N2N learning) (comparative example), and by this embodiment (proposed method).

[0126] The four images on the left in Figures 13 and 14 represent the entire image. Figure 13 shows the mean and standard deviation (SD) of pixel values ​​in a portion R3 of the left image. This portion R3 is a non-emitting area. Therefore, a smaller standard deviation of pixel values ​​in portion R3 indicates less noise. The four images on the right in Figure 14 are magnified views of the portion indicated by the arrow in the left image of Figure 14. The noise in the images shown in this example is pixel defects (poor brightness levels).

[0127] In this example, five images were acquired, and one of them was used as the source image. The average image of the four images is the average of the pixel values ​​at each pixel position of the four images other than the source image. The four images other than the source image were used to train the machine learning of the inference model in this embodiment and in Noise2Noise. Both the inference model in this embodiment and in Noise2Noise is a 6-layer Conv2D+ReLU.

[0128] As shown in the examples in Figures 12 to 14, noise reduction is effectively performed according to this embodiment. Furthermore, the accuracy of noise reduction by this embodiment is equal to or better than that of other methods.

[0129] Next, a learning program and a noise reduction program for executing the processing performed by the series of learning systems 10 and noise reduction systems 20 described above will be explained. As shown in Figure 15(a), the learning program 100 is stored in a program storage area 111 formed on a computer-readable recording medium 110 that is inserted into and accessed by a computer, or is provided by a computer. The recording medium 110 may be a non-temporary recording medium.

[0130] The learning program 100 comprises a learning acquisition module 101, a noise reduction module 102, a noise addition module 103, and a training module 104. The functions realized by executing the learning acquisition module 101, the noise reduction module 102, the noise addition module 103, and the training module 104 are the same as the functions of the learning acquisition unit 11, the noise reduction unit 12, the noise addition unit 13, and the training unit 14 of the learning system 10 described above.

[0131] As shown in Figure 15(b), the noise reduction program 200 is stored in a program storage area 211 formed on a computer-readable recording medium 210 that is inserted into and accessed by the computer, or is provided by the computer. The recording medium 210 may be a non-temporary recording medium. If the learning program 100 and the noise reduction program 200 are executed on the same computer, the recording medium 210 may be the same as the recording medium 110.

[0132] The noise reduction program 200 comprises a noise reduction acquisition module 201 and a noise reduction module 202. The functions realized by running the noise reduction acquisition module 201 and the noise reduction module 202 are the same as the functions of the noise reduction acquisition unit 21 and the noise reduction unit 22 of the noise reduction system 20 described above.

[0133] Furthermore, the learning program 100 and the noise reduction program 200 may be configured such that part or all of them are transmitted via a transmission medium such as a communication line, received by other equipment, and recorded (including installation). Also, each module of the learning program 100 and the noise reduction program 200 may be installed on any of multiple computers, not just one. In that case, the series of processes described above will be performed by a computer system consisting of these multiple computers.

[0134] The learning method, inference model, denoising method, denoising program, and denoising system of this disclosure have the following configurations. [1] A learning method for generating an inference model used to remove noise from data having multiple values, A training data acquisition step involves acquiring training data that has multiple values ​​and contains noise, A noise reduction step is performed to generate noise-reduced training data from the training data acquired in the aforementioned training acquisition step, using an inference model in the process of training. A noise addition step is performed to generate noise-added training data by adding a predetermined noise to the training data after noise removal generated in the noise removal step, The training step includes training the machine learning of an inference model using a combination of the training data acquired in the training acquisition step and the noise-added training data generated in the noise addition step, each of which is used as data containing noise. After the inference model is trained in the training step, in the denoising step, the trained inference model is used as an in-progress inference model to generate denoised training data. A learning method comprising repeating the noise reduction step, the noise addition step, and the training step. [2] The learning method according to [1], wherein each time the inference model is trained in the training step, the denoising step generates denoised training data using the trained inference model as an inference model in progress. [3] The learning method according to [1] or [2], wherein in one of the iterations, in the training step, the machine learning of the inference model is trained using the training data acquired in the training data acquisition step instead of the noise-added training data generated in the noise-adding step. [4] A learning method according to any one of [1] to [3], wherein in the noise addition step, noise amplified from the noise to be removed is added as a pre-set noise to generate noise addition learning data. [5] A learning method according to any one of [1] to [4], wherein in the noise addition step, a preset noise is added according to the repetition to generate noise-added learning data. [6] A learning method according to any one of [1] to [5], wherein in the noise addition step, noise based on the same type of noise as the noise to be removed is added as a pre-set noise to generate noise addition learning data. [7] An inference model for causing a computer to take data having multiple values ​​as input and perform operations according to the input to output information, An inference model generated by one of the learning methods described in [1] to [6]. A noise reduction method for removing noise from data having multiple values, using an inference model generated by any of the learning methods described in [8] [1] to [6], A removal acquisition step to acquire noise removal target data that has multiple values ​​and is subject to noise removal, A noise reduction step in which noise reduction data is generated from the noise reduction target data acquired in the aforementioned removal acquisition step using the inference model, A noise reduction method that includes [details omitted]. [9] The aforementioned inference model is for removing speckle noise from optical interference images, The noise reduction method described in [8], wherein in the removal acquisition step, an optical interference image is acquired as data to be denoised.

[10] A noise reduction program that causes a computer to operate as a noise reduction system that removes noise from data having multiple values ​​using an inference model generated by any of the learning methods described in [1] to [6], The aforementioned computer, A removal acquisition means for acquiring noise removal target data that has multiple values ​​and is subject to noise removal, A noise reduction means that generates noise-reduced data from the noise-reduced data acquired by the noise reduction acquisition means using the inference model, A noise reduction program that functions as such.

[11] A noise reduction system that removes noise from data having multiple values ​​using an inference model generated by any of the learning methods described in [1] to [6], A removal acquisition means for acquiring noise removal target data that has multiple values ​​and is subject to noise removal, A noise reduction means that generates noise-reduced data from the noise-reduced data acquired by the noise reduction acquisition means using the inference model, A noise reduction system equipped with the following features. [Explanation of Symbols]

[0135] 10...Learning system, 11...Learning acquisition unit, 12...Noise reduction unit, 13...Noise addition unit, 14...Training unit, 20...Noise reduction system, 21...Removal acquisition unit, 22...Noise reduction unit, 100...Learning program, 101...Learning acquisition module, 102...Noise reduction module, 103...Noise addition module, 104...Training module, 110...Recording medium, 111...Program storage area, 200...Noise reduction program, 201...Removal acquisition module, 202...Noise reduction module, 210...Recording medium, 211...Program storage area.

Claims

1. A learning method for generating an inference model used to remove noise from data having multiple values, A training data acquisition step involves acquiring training data that has multiple values ​​and contains noise, A noise reduction step is performed to generate noise-reduced training data from the training data acquired in the aforementioned training acquisition step, using an inference model in the process of training. A noise addition step is performed to generate noise-added training data by adding a predetermined noise to the training data after noise removal generated in the noise removal step, The training step includes training the machine learning of an inference model using a combination of the training data acquired in the training acquisition step and the noise-added training data generated in the noise-adding step, each of which is used as data containing noise. After the inference model is trained in the training step, in the denoising step, the trained inference model is used as an in-progress inference model to generate denoised training data. A learning method comprising repeating the noise reduction step, the noise addition step, and the training step.

2. The learning method according to claim 1, wherein each time the inference model in the training step is trained, the denoising step generates denoised training data using the trained inference model as an inference model in training.

3. The learning method according to claim 1 or 2, wherein in one of the iterations, in the training step, the machine learning of the inference model is trained using the training data acquired in the training data acquisition step instead of the noise-added training data generated in the noise-adding step.

4. The learning method according to claim 1 or 2, wherein in the noise addition step, noise amplified from noise assumed to be to be removed is added as a pre-set noise to generate noise addition learning data.

5. The learning method according to claim 1 or 2, wherein in the noise addition step, a preset noise is added according to the repetitions to generate noise-added learning data.

6. The learning method according to claim 1 or 2, wherein in the noise addition step, noise based on the same type of noise as the noise to be removed is added as a pre-set noise to generate noise-added learning data.

7. An inference model for causing a computer to function by taking data with multiple values ​​as input, performing calculations according to the input, and outputting information, An inference model generated by the learning method described in claim 1 or 2.

8. A noise reduction method for removing noise from data having multiple values ​​using an inference model generated by the learning method described in claim 1 or 2, A removal acquisition step to acquire noise removal target data that has multiple values ​​and is subject to noise removal, A noise reduction step in which noise reduction data is generated from the noise reduction target data acquired in the aforementioned removal acquisition step using the inference model, A noise reduction method that includes [details omitted].

9. The aforementioned inference model is for removing speckle noise from optical interference images. The noise reduction method according to claim 8, wherein in the removal acquisition step, an optical interference image is acquired as data to be removed from noise.

10. A noise reduction program that causes a computer to operate as a noise reduction system that removes noise from data having multiple values ​​using an inference model generated by the learning method described in claim 1 or 2, The aforementioned computer, A removal acquisition means for acquiring noise removal target data that has multiple values ​​and is subject to noise removal, A noise reduction means that generates noise-reduced data from the noise-reduced data acquired by the noise reduction acquisition means using the inference model, A noise reduction program that functions as such.

11. A noise reduction system that removes noise from data having multiple values ​​using an inference model generated by the learning method described in claim 1 or 2, A removal acquisition means for acquiring noise removal target data that has multiple values ​​and is subject to noise removal, A noise reduction means that generates noise-reduced data from the noise-reduced data acquired by the noise reduction acquisition means using the inference model, A noise reduction system equipped with the following features.

Images