CN116152578B - Training method and device for noise reduction generation model, noise reduction method and medium - Google Patents
Training method and device for noise reduction generation model, noise reduction method and medium Download PDFInfo
- Publication number
- CN116152578B CN116152578B CN202310449238.1A CN202310449238A CN116152578B CN 116152578 B CN116152578 B CN 116152578B CN 202310449238 A CN202310449238 A CN 202310449238A CN 116152578 B CN116152578 B CN 116152578B
- Authority
- CN
- China
- Prior art keywords
- noise reduction
- sample
- noise
- curve
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000009467 reduction Effects 0.000 title claims abstract description 285
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000012549 training Methods 0.000 title claims abstract description 40
- 238000001228 spectrum Methods 0.000 claims abstract description 211
- 230000003595 spectral effect Effects 0.000 claims abstract description 182
- 238000003384 imaging method Methods 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims description 25
- 230000004927 fusion Effects 0.000 claims description 25
- 238000011176 pooling Methods 0.000 claims description 25
- 238000013145 classification model Methods 0.000 claims description 15
- 238000002372 labelling Methods 0.000 claims description 15
- 230000003287 optical effect Effects 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 6
- 238000011084 recovery Methods 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000000329 molecular dynamics simulation Methods 0.000 abstract description 3
- 239000000523 sample Substances 0.000 description 113
- 230000006870 function Effects 0.000 description 25
- 239000000975 dye Substances 0.000 description 15
- 230000000694 effects Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000007613 environmental effect Effects 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 210000004027 cell Anatomy 0.000 description 5
- VOFUROIFQGPCGE-UHFFFAOYSA-N nile red Chemical compound C1=CC=C2C3=NC4=CC=C(N(CC)CC)C=C4OC3=CC(=O)C2=C1 VOFUROIFQGPCGE-UHFFFAOYSA-N 0.000 description 5
- 238000005070 sampling Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 239000007850 fluorescent dye Substances 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000011088 calibration curve Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000005284 excitation Effects 0.000 description 3
- 150000002632 lipids Chemical class 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000006862 quantum yield reaction Methods 0.000 description 3
- 239000000232 Lipid Bilayer Substances 0.000 description 2
- 210000000170 cell membrane Anatomy 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000000701 chemical imaging Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000000295 emission spectrum Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000002189 fluorescence spectrum Methods 0.000 description 2
- 239000012528 membrane Substances 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 239000012472 biological sample Substances 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000799 fluorescence microscopy Methods 0.000 description 1
- 238000001506 fluorescence spectroscopy Methods 0.000 description 1
- 102000034287 fluorescent proteins Human genes 0.000 description 1
- 108091006047 fluorescent proteins Proteins 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003834 intracellular effect Effects 0.000 description 1
- 238000004020 luminiscence type Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000000386 microscopy Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 210000003463 organelle Anatomy 0.000 description 1
- 239000008188 pellet Substances 0.000 description 1
- 239000002096 quantum dot Substances 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 150000003384 small molecules Chemical class 0.000 description 1
- 230000009870 specific binding Effects 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
A training method of a spectral curve noise reduction generation type model, wherein the spectral curve noise reduction generation type model and a discrimination model form a generation countermeasure network, and the method comprises the following steps: acquiring a spectrum curve sample set; obtaining a label corresponding to a spectrum curve sample; inputting a spectral curve sample which is not noise-reduced and meets the preset condition into a spectral curve noise-reduction generation model to generate a noise-reduced spectral curve; and inputting the noise reduction sample into a judging model to obtain a judging result of the noise reduction sample, obtaining a loss function for generating an countermeasure network according to the judging result, respectively adjusting parameters of the spectral curve noise reduction generating model and the judging model according to the loss function, and obtaining the trained spectral curve noise reduction generating model. Because the acquisition is not needed for a plurality of times, the molecular dynamic imaging can be carried out, and the molecular spectrum after noise reduction can be acquired. The invention also provides a training device of the spectral curve noise reduction generation model, a spectral curve noise reduction method and a storage medium.
Description
Technical Field
The invention relates to the technical field of spectrum analysis, in particular to a training method and device of a spectrum curve noise reduction generation model, a spectrum curve noise reduction method and a storage medium.
Background
In recent years, the rapid development of microscopic imaging technology has greatly driven the advancement of life sciences technology. Among them, a fluorescence microscope observes a biological sample through a fluorescent-labeled probe, and its fundamental principle of luminescence is a process in which a fluorophore absorbs photons of excitation light of a specific wavelength and emits fluorescence longer than the excitation wavelength. Thus, by designing suitable specific binding fluorochromes, such as fluorescent proteins, organic fluorescent small molecules, quantum dots, etc., subcellular organelle-level spatial resolution and millisecond-level time resolution morphological observations can be achieved. Besides the intensity information of fluorescent molecules, the fluorescence spectrum is one of the important parameters for observing the marker molecules, and some environment-sensitive fluorescent dye molecules can provide features other than morphology, such as molecular structure, spatial arrangement of biological molecules, micro-environment around the molecules, and the like. In summary, fluorescence spectroscopy, which reveals the environmental parameters of cells in another dimension, is one of the important tools for life activity research.
The current common method of monitoring spectral distribution is a fluorescence spectrophotometer, but a fluorescence spectrophotometer is generally not used in living cell studies. In view of this problem, there are some studies currently connecting fluorescence microscopy imaging technology with multispectral imaging technology, i.e., fluorescence multispectral microscopy imaging method, which successfully achieves spectral super-resolution imaging on the nanometer scale of cell membranes.
However, noise is also present in the acquired spectroscopic signal of fluorescent molecules due to the background interference of cellular fluorescence and the influence of systematic noise, so that it is difficult to distinguish the environmental parameters and the change process reflected by the spectroscopic signal of a single fluorescent molecule. Therefore, the fluorescence multispectral microscopic imaging method generally collects the spectrum signals of fluorescent molecules for multiple times, and then performs a global averaging method to analyze the characteristics of the environment-sensitive dye molecules. The analysis can be performed after multiple acquisitions, so that the method can only statically analyze the integral property of the cells, but is difficult to be applied to dynamic life activity analysis, and a new technical scheme is also required to be provided.
Disclosure of Invention
The invention mainly solves the technical problem that the spectrum signal of a single fluorescent molecule has noise and cannot be applied to dynamic life activity analysis.
According to a first aspect, in one embodiment, there is provided a training method of a spectral curve noise reduction generating model, where the spectral curve noise reduction generating model is used to form a generating countermeasure network with a discrimination model, and the method includes:
acquiring a spectrum curve sample set, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples;
obtaining a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction;
inputting the spectrum curve sample which is not noise-reduced and meets the preset condition into the spectrum curve noise-reduction generation model to generate a noise-reduced spectrum curve;
a noise reduction sample set is obtained, wherein the noise reduction sample in the noise reduction sample set comprises the spectrum curve after noise reduction and the spectrum curve sample with noise reduction as a label;
inputting the noise reduction sample into the discrimination model to obtain a judgment result of the noise reduction sample, wherein the judgment result at least comprises non-noise reduction and noise reduction;
and obtaining a loss function of the generated countermeasure network according to the judging result, and respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
According to a second aspect, in one embodiment, there is provided a training apparatus of a spectral curve noise reduction generating model for constructing a generating countermeasure network with a discrimination model, the apparatus comprising:
the sample acquisition module is used for acquiring a spectrum curve sample set, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples;
the labeling module is used for acquiring a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction;
the sample noise reduction module is used for inputting the spectrum curve sample which is not noise reduced and meets the preset condition into the spectrum curve noise reduction generation model so as to generate a spectrum curve after noise reduction;
the judging module is used for acquiring a noise reduction sample set, wherein the noise reduction sample in the noise reduction sample set comprises the noise reduction spectrum curve and the spectrum curve sample with the noise reduced label, the noise reduction sample is input into the judging model to obtain a judging result of the noise reduction sample, and the judging result comprises non-noise reduction and noise reduction;
and the training module is used for obtaining a loss function of the generated countermeasure network according to the judging result, respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
According to a third aspect, in one embodiment, there is provided a method for noise reduction of a spectral curve, including:
acquiring an initial spectrum curve;
and inputting the initial spectral curve into a spectral curve noise reduction generating model to obtain the spectral curve after the initial spectral curve noise reduction, wherein the spectral curve noise reduction generating model is trained by the training method in the first aspect.
According to a fourth aspect, in one embodiment, there is provided a method of noise reduction of a spectral curve, comprising:
acquiring an initial spectrum curve;
inputting the initial spectral curve into a pre-trained spectral curve noise reduction generating model, wherein inputting the initial spectral curve into the pre-trained spectral curve noise reduction generating model comprises the following steps:
extracting features of the initial spectrum curve through an encoder to obtain a first feature;
the decoder is used for carrying out up-scaling and information recovery on the first characteristic so as to obtain fusion data with the same dimension as the initial spectrum curve;
and adding the fusion data with the initial spectrum curve to obtain a spectrum curve after noise reduction of the initial spectrum curve.
According to a fifth aspect, an embodiment provides a computer readable storage medium having stored thereon a program executable by a processor to implement the method according to the first, third or fourth aspects.
According to the training method of the spectral curve noise reduction generation type model, the spectral curve noise reduction generation type model and the discrimination model are formed to generate an countermeasure network and are trained, and then the spectral curve can be subjected to noise reduction based on the trained spectral curve noise reduction generation type model, so that the influence of noise on single-molecule spectral information data acquisition is reduced. Because the spectrum signal of the fluorescent molecule does not need to be acquired for multiple times, the molecular spectrum after noise reduction can be acquired while the molecular dynamic imaging is carried out, and the method can be used for dynamically monitoring the environmental parameter change of a single molecule in the intracellular activity process.
Drawings
FIG. 1 is a schematic diagram of a spectral curve noise reduction generating model according to an embodiment;
FIG. 2 is a schematic diagram of a spectral curve noise reduction generating model according to another embodiment;
FIG. 3 is a schematic diagram of a spectral curve noise reduction generating model according to another embodiment;
FIG. 4 is a schematic diagram of a classification model according to an embodiment;
FIG. 5 is a schematic diagram of a discrimination model according to an embodiment;
FIG. 6 is a flowchart of a training method of a spectral curve noise reduction generation model according to an embodiment;
FIG. 7 is a flowchart of a training method of a spectral curve noise reduction generation model according to another embodiment;
FIG. 8 is a schematic diagram of a training device for spectral curve noise reduction generation model according to an embodiment;
FIG. 9 is a flow chart of a spectral curve denoising method according to an embodiment;
fig. 10 is a spectral diagram of an embodiment after noise reduction.
Detailed Description
The invention will be described in further detail below with reference to the drawings by means of specific embodiments. Wherein like elements in different embodiments are numbered alike in association. In the following embodiments, numerous specific details are set forth in order to provide a better understanding of the present application. However, one skilled in the art will readily recognize that some of the features may be omitted, or replaced by other elements, materials, or methods in different situations. In some instances, some operations associated with the present application have not been shown or described in the specification to avoid obscuring the core portions of the present application, and may not be necessary for a person skilled in the art to describe in detail the relevant operations based on the description herein and the general knowledge of one skilled in the art.
Furthermore, the described features, operations, or characteristics of the description may be combined in any suitable manner in various embodiments. Also, various steps or acts in the method descriptions may be interchanged or modified in a manner apparent to those of ordinary skill in the art. Thus, the various orders in the description and drawings are for clarity of description of only certain embodiments, and are not meant to be required orders unless otherwise indicated.
The numbering of the components itself, e.g. "first", "second", etc., is used herein merely to distinguish between the described objects and does not have any sequential or technical meaning. The terms "coupled" and "connected," as used herein, are intended to encompass both direct and indirect coupling (coupling), unless otherwise indicated.
In the embodiment of the invention, a spectral curve noise reduction generation model and a discrimination model form a generation countermeasure network, training is performed based on a supervised learning method, and noise reduction is performed on the spectral curve based on the spectral curve noise reduction generation model of deep learning, so that the influence of noise on single-molecule spectral information data acquisition is reduced, and the method is used for realizing real-time dynamic environment parameter monitoring based on single fluorescent molecular spectral information. Because the spectrum signals of fluorescent molecules are not required to be acquired for multiple times, the molecular dynamic imaging can be carried out, and the molecular spectrum after noise reduction can be acquired at the same time.
Some embodiments provide a spectral curve noise reduction generating model for inputting an initial spectral curve, then performing noise reduction on the initial spectral curve and outputting a noise reduced spectral curve, wherein the noise reduced spectral curve can be directly used for analyzing the characteristics of dye molecules in the noise reduced spectral curve, so that the noise reduced spectral curve can be applied to dynamic life activity analysis. Referring to fig. 1, a specific description of the spectral curve noise reduction generating model is provided below.
The spectral curve noise reduction generating model employs a U-shaped network that includes an encoder and a decoder. The encoder is used for extracting features of the initial spectrum curve, the decoder is used for carrying out feature fusion on the features extracted by the encoder, fusion data are obtained, and finally the fusion data are added with the initial spectrum curve to obtain a spectrum curve after noise reduction of the initial spectrum curve.
Referring to fig. 1 again, the encoder is configured to perform three feature extraction on the input initial spectral curve, and the decoder is configured to perform three feature fusion correspondingly. Wherein the encoder performs a first feature extraction through the linear layer, the feature extraction block (FEB, feature extraction block), and the Pooling layer (Pooling) in sequence. Referring to fig. 2, the feature extraction block includes two deep learning layers (DL 1, DL 2), each of which includes a convolutional layer (conv), an active layer (leak ReLU), and a batch regularization layer (BN, batch Normalization). The encoder sequentially performs second feature extraction and third feature extraction through the feature extraction block and the pooling layer respectively, and finally obtains a first feature. The decoder sequentially performs a first feature fusion and a second feature fusion on the first feature through an up-sampling connection module (UCB, upsampling & Conv block) and a feature extraction block, respectively. Referring to fig. 3, the upsampling connection module includes an interpolation module (interpolation) and a convolution layer. And the decoder performs third feature fusion through the up-sampling connection module, the feature extraction block and the linear layer, and obtains final fusion data. In the three-time feature fusion, the decoder performs up-sampling through an up-sampling connection module to realize dimension increase, acquires features obtained by feature extraction of the encoder through Skip connection (Skip connection), fuses the features in the up-sampling process to realize information recovery, and finally outputs fused data, and meanwhile enables the fused data to be identical to the initial spectrum curve dimension. Therefore, the fusion data and the initial spectrum curve are added, so that the spectrum curve after noise reduction of the initial spectrum curve can be obtained and output. In this embodiment, adding the fusion data to the initial spectral curve may enable the spectral curve noise reduction generation model to learn the noise of the system, which includes an optical device for collecting the spectral signal.
The spectral curve noise reduction generation model and the discrimination model constitute a generation countermeasure network, and then are trained by the following training method, which will be described in detail below.
Firstly, a spectrum curve sample set is obtained, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples, for example, 6000 image sequences are collected through an optical device under various lipid environments where fluorescent molecules are located, each lipid environment is provided with about 11000 spectrum curves as spectrum curve samples after data screening, 10000 spectrum curve samples are used for training, and 1000 spectrum curve samples are used for testing. The method comprises the steps of respectively collecting Nile Red (Nile Red) single-molecule spectrum data of five lipid bilayer supporting membranes with different ratios, wherein the ratios of the lipid bilayer supporting membranes are the environment types of fluorescent molecules when spectrum curve samples are obtained. Different environments may also be coded as their environmental categories when comparing other environmentally sensitive dye molecules. In this embodiment, labels corresponding to each spectrum curve sample are also required to be obtained, and the labels are labels of classification results of the spectrum curve samples, where the classification results include non-noise reduction and noise reduction, and environment types, that is, whether the labels of the spectrum curve samples are noise reduction, and specific environment types. The labeling mode can be manual labeling or automatic labeling, for example, manual labeling.
For the spectral curve samples, the signal to noise ratio of the finally obtained spectral curve samples is also different due to the different quantum yields of the different fluorescent dyes. For example, for a fluorescent dye with higher quantum yield, the signal-to-noise ratio of the finally obtained spectrum curve sample is higher, at this time, the relevant spectrum characteristic can be recovered from the finally obtained spectrum curve sample by the spectrum curve noise reduction generation model, while for a fluorescent dye molecule with lower quantum yield, since the emitted fluorescence intensity is lower and the single-molecule fluorescence is more interfered by noise under the same excitation light power condition, the signal detected by the optical device is more affected by noise, at this time, the signal-to-noise ratio of the finally obtained spectrum curve sample is lower, and the relevant spectrum characteristic cannot be recovered from the finally obtained spectrum curve sample. Therefore, for the spectrum curve sample with the too low signal to noise ratio, the spectrum curve sample can be considered to belong to noise information, so that the spectrum curve sample with the too low signal to noise ratio needs to be eliminated to avoid influencing the network training and testing, the spectrum curve samples with different signal to noise ratios can be distinguished, and the spectrum curve sample with the signal to noise ratio which does not meet the condition does not participate in the subsequent network training and testing. In some embodiments, the signal-to-noise ratio of spectral curve samples may be first distinguished using a classification model that includes a linear layer and a residual network (ResNet). In this embodiment, since the label is a noise-reduced spectrum curve sample, the noise-reduced spectrum curve sample basically satisfies a preset signal-to-noise ratio, the label is mainly distinguished from a noise-reduced spectrum curve sample. Referring to fig. 4 specifically, any spectrum curve sample without noise reduction is input into a linear layer, the linear layer is used for carrying out dimension lifting on the spectrum curve sample, the spectrum curve sample in this embodiment is one-dimensional data, for example, the dimension lifting of 120x1 is one-dimensional data of 4096x1, then the one-dimensional data after dimension lifting is adjusted to two-dimensional data through a reshape function, for example, the adjustment of 4096x1 is 64x64x1, then the two-dimensional data is input into a residual network for feature extraction, so as to output a judging result of whether noise is generated, for example, a judging result output when a preset signal-to-noise ratio is met is a signal, and the signal is involved in subsequent network training and testing, otherwise, the output judging result is noise (noise), so that the noise can be filtered.
And then, inputting a spectral curve sample which is marked as non-noise-reduced and meets the preset signal-to-noise ratio into the spectral curve noise-reduction generating model, extracting features of the spectral curve noise-reduction generating model through an encoder, carrying out feature fusion by a decoder to obtain fusion data, and finally adding the fusion data with the input spectral curve sample to obtain a corresponding noise-reduced spectral curve. For the discrimination model, a noise reduction sample set needs to be acquired first, and noise reduction samples in the noise reduction sample set comprise noise reduction spectrum curves generated by the spectrum curve noise reduction generation model, and labels are noise reduction spectrum curve samples. And inputting the noise reduction sample into a judging model to obtain a judging result of the noise reduction sample, wherein the judging result comprises non-noise reduction and noise reduction, the non-noise reduction and the noise reduction correspond to true and false respectively, and the judging result can also comprise the environment category of the noise reduction sample. In this embodiment, since the spectral curve noise reduction generation model and the discrimination model form the generation countermeasure network, when the accuracy of the discrimination result of the discrimination model is high, the parameters of the spectral curve noise reduction generation model need to be adjusted to strengthen the same. For example, when the denoised spectral curve is input, if the judgment result is denoised (i.e. judged true), the judgment is wrong, otherwise (i.e. judged false), the judgment is correct, and when the accuracy of the judgment result is high, the difference between the denoised spectral curve and the spectral curve sample with the denoised label is possibly large, so that the judgment model is easy to obtain the correct judgment result, and at the moment, the parameters of the spectral curve denoise generating model need to be adjusted, so that the difference between the denoised spectral curve and the spectral curve sample with the denoised label is reduced, and the purpose of enhancing the spectral curve denoise generating model is achieved. When the spectral curve noise reduction generating model is enhanced, the accuracy of the judging result of the judging model is possibly lower, and at the moment, the fact that the judging model can not better distinguish the difference between the noise reduction spectral curve and the spectral curve sample with the noise reduction label is explained, so that the parameters of the judging model need to be adjusted, and the purpose of enhancing the judging model is achieved. The contrast between the spectral curve noise reduction generation model and the discrimination model can be enhanced continuously, and the difference between the noise reduction spectral curve and the spectral curve sample with the labels being noise reduced is smaller and smaller, so that the trained spectral curve noise reduction generation model is finally obtained.
In some embodiments, the discrimination model needs to determine the environment type of the noise-reduced sample in addition to determining whether the noise-reduced sample is noise-reduced, so that the discrimination model can pay more attention to the information of the environment where the dye molecule is located. In particular, since fluorescent molecules can have different emission spectra under different environmental categories, for example, researches have proved that nile red molecular dyes can have different emission spectra under different polar environments, and the ordered phase has obvious blue shift phenomenon compared with the unordered phase spectrum. Therefore, the discrimination model can judge the environment type through the influence of the spectral curve, so that the environment type of the fluorescent molecule can be obtained, and the accuracy of the spectral curve noise reduction generation model in generating the noise reduction spectral curve can be improved. Referring to fig. 5, in some embodiments, the discriminant model includes a feature extraction layer, a max pooling layer (max pooling), an average pooling layer (avg pooling), a first multi-layer perceptron, and a second multi-layer perceptron. The feature extraction layer is used for extracting features of the input noise reduction sample and obtaining second features. Specifically, the feature extraction layer comprises three layers of convolution layers, each comprising a 4x4 two-dimensional convolution layer, an activation layer and a batch regularization layer. The maximum pooling layer and the average pooling layer are used for respectively carrying out dimension reduction on the second features to respectively obtain third features and fourth features, the first multi-layer perceptron is used for inputting the third features to obtain a judgment result of environment types of the noise reduction samples, and the second multi-layer perceptron is used for inputting the fourth features to obtain a judgment result of whether the noise reduction samples are noise reduction.
And finally, acquiring and generating an overall loss function of the countermeasure network according to the judging result of the judging model, so as to respectively adjust parameters of the spectral curve noise reduction generating model and the judging model, wherein the overall loss function is specifically as follows:
L=argmin G max D [l wGAN (G,D)+λ 1 l L1 (G)+λ 2 l TV (G)+λ 3 l Aux (D)]；
wherein,,Las a function of the overall loss,Grepresenting a spectral curve noise reduction generative model,Dthe model of the discrimination is represented by,argmin G max D representing the relative sumGThe related function takes the minimum value and is related toDThe function of the correlation takes a maximum value,l wGAN representing a countermeasure networkwGANThe loss function is a function of the loss,l L1 (G) Representing a model of spectral curve noise reduction generationL1 the regularization loss is set to be equal to 1,l TV (G) Representing a model of spectral curve noise reduction generationTVThe total variation loss is calculated by the method,l Aux (D) Represents the environmental class classification loss of the discriminant model,λ 1 、λ 2 andλ 3 respectively representing the weight values. Wherein by addingTVTotal variation loss functionL1 regularization loss, so that the spectrum curve output by the spectrum curve noise reduction generation type model is smoother.
According to the embodiment, the spectral curve is noise-reduced through the deep-learning spectral curve noise-reduction generation model, so that the method can be used for realizing real-time dynamic environment parameter monitoring based on single fluorescent molecule spectral information, and is hopeful to provide a new research tool for understanding the vital activity process of cells from the molecular level. In the training process of the spectral curve noise reduction generation type model, a spectral curve sample with low signal-to-noise ratio is filtered, so that a good training effect is ensured, and meanwhile, the spectral curve noise reduction generation type model performs phase correlation between fusion data and an initial spectral curve, so that the system noise can be learned. The discrimination model not only judges the true or false of the spectrum curve sample, but also judges the environment type of the spectrum curve sample, thereby further improving the performance of the spectrum curve noise reduction generation model.
Referring to fig. 6, some embodiments provide a training method of a spectral curve noise reduction generation model, which is applied to the generation countermeasure network, and the training method specifically includes:
step 100: a spectral curve sample set is obtained, the spectral curve sample set comprising a plurality of spectral curve samples.
Step 200: and obtaining a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction.
Step 300: and inputting the spectral curve sample which is not noise-reduced and meets the preset condition into the spectral curve noise-reduction generating model to generate a noise-reduced spectral curve.
Step 400: the method comprises the steps of obtaining a noise reduction sample set, wherein noise reduction samples in the noise reduction sample set comprise a spectrum curve after noise reduction and a spectrum curve sample with noise reduction as labels, inputting the noise reduction samples into the discrimination model to obtain a judging result of the noise reduction samples, and the judging result at least comprises non-noise reduction and noise reduction.
Step 500: and obtaining a loss function of the generated countermeasure network according to the judging result, and respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
Referring to fig. 7, in some embodiments, the training method of the spectral curve noise reduction generating model further includes:
step 310: and inputting any spectrum curve sample with the label not subjected to noise reduction into a classification model to judge whether any spectrum curve sample with the label not subjected to noise reduction belongs to noise, wherein the signal to noise ratio of the noise is lower than a preset signal to noise ratio.
Step 320: if the classification model judges that the noise does not belong to the noise, the label is any spectrum curve sample which is not noise-reduced, and the preset condition is met.
In some embodiments, the classification model includes a linear layer and a residual network, and when determining whether the label is any of the spectrum curve samples without noise reduction belongs to noise, the classification model specifically includes: carrying out dimension lifting on any spectrum curve sample with the label not subjected to noise reduction through the linear layer, and adjusting the spectrum curve sample after dimension lifting into two-dimensional data through a reshape function; and inputting the two-dimensional data into a residual error network for feature extraction so as to output a judging result of whether the two-dimensional data is noise or not.
In some embodiments, the spectral curve noise reduction generation model adopts a U-shaped network, and the U-shaped network comprises an encoder and a decoder, and specifically comprises when generating a noise-reduced spectral curve; extracting features of the spectrum curve sample which is not noise-reduced and meets preset conditions by the aid of an encoder, and obtaining first features; the decoder is used for carrying out up-scaling and information recovery on the first characteristics so as to obtain fusion data which are the same as the spectrum curve sample dimension of which the label is not noise-reduced and meets preset conditions; and adding the fusion data and the spectrum curve sample which is not noise-reduced and meets the preset condition by the label to obtain the spectrum curve after noise reduction.
In some embodiments, the classification result further includes an environmental class in which the fluorescent molecule is located when the spectral curve sample is obtained. In some embodiments, the discrimination model includes a feature extraction layer, a maximum pooling layer, an average pooling layer, a first multi-layer perceptron, and a second multi-layer perceptron, and when a result of determining the noise reduction sample is obtained, the method specifically includes: extracting features of the noise reduction sample through the feature extraction layer, and obtaining second features; the second feature is subjected to dimension reduction through the maximum pooling layer and the average pooling layer respectively so as to obtain a third feature and a fourth feature respectively; and inputting the third characteristic and the fourth characteristic into a first multi-layer perceptron to obtain a judgment result of the environment category of the noise reduction sample, and inputting the third characteristic and the fourth characteristic into a second multi-layer perceptron to obtain a judgment result of whether the noise reduction sample is noise reduction.
In some embodiments, the training method of the spectral curve noise reduction generation model further comprises: according to the noise-reduced spectrum curve, obtaining total variation loss and/or regularization loss of a spectrum curve noise-reduction generation type model; and adjusting parameters of the spectral curve noise reduction generation type model according to the total variation loss and/or regularization loss.
Referring to fig. 8, some embodiments provide a training apparatus for a spectral curve noise reduction generating model, which includes a sample acquiring module 10, a labeling module 20, a sample noise reduction module 30, a discriminating module 40, and a training module 50, wherein the spectral curve noise reduction generating model is used to form a generating countermeasure network with the discriminating model, which is described in detail below.
The sample acquisition module 10 is configured to acquire a spectral curve sample set, which includes a plurality of spectral curve samples.
The labeling module 20 is configured to obtain a label corresponding to the spectrum curve sample, where the label is a label of a classification result of the spectrum curve sample, and the classification result at least includes non-noise reduction and noise reduction.
The sample noise reduction module 30 is configured to input the spectral curve sample, which is not noise-reduced and satisfies a preset condition, to the spectral curve noise reduction generating model, so as to generate a noise-reduced spectral curve.
The judging module 40 is configured to obtain a noise reduction sample set, where the noise reduction samples in the noise reduction sample set include the noise reduction spectrum curve and the spectrum curve sample with the noise reduced label, and input the noise reduction samples into the judging model to obtain a judging result of the noise reduction samples, where the judging result includes non-noise reduction and noise reduction.
The training module 50 is configured to obtain a loss function of the generated countermeasure network according to the determination result, and adjust parameters of the spectral curve noise reduction generating model and the determination model at least according to the loss function respectively until the generated countermeasure network converges, and obtain a trained spectral curve noise reduction generating model.
In some embodiments, the training device of the spectral curve noise reduction generating model further includes a filtering module, where the filtering module is configured to input any spectral curve sample that is not noise reduced by the tag into the classification model, so as to determine whether any spectral curve sample that is not noise reduced by the tag belongs to noise, a signal-to-noise ratio of the noise is lower than a preset signal-to-noise ratio, and if the classification model determines that the noise does not belong to the noise, any spectral curve sample that is not noise reduced by the tag meets the preset condition.
In some embodiments, the training module is further configured to obtain total variation loss and/or regularization loss of a noise reduction generating model of the spectrum curve according to the noise reduction spectrum curve; and adjusting parameters of the spectral curve noise reduction generation type model according to the total variation loss and/or regularization loss.
Some embodiments provide a method of noise reduction of a spectral curve, comprising the steps of:
an initial spectral curve is obtained.
And inputting the initial spectral curve into a spectral curve noise reduction generation type model to obtain the spectral curve after the initial spectral curve noise reduction, wherein the spectral curve noise reduction generation type model is trained by the training method.
Referring to fig. 9, some embodiments provide a method for noise reduction of a spectrum curve, which includes the following steps:
step 600: an initial spectral curve is obtained.
Step 610: and inputting the initial spectrum curve into a pre-trained spectrum curve noise reduction generating model.
Step 620: the step of inputting the initial spectrum curve into a pre-trained spectrum curve noise reduction generating model comprises the following steps: extracting features of the initial spectrum curve through an encoder to obtain a first feature; the decoder is used for carrying out up-scaling and information recovery on the first characteristic so as to obtain fusion data with the same dimension as the initial spectrum curve; and adding the fusion data with the initial spectrum curve to obtain a spectrum curve after noise reduction of the initial spectrum curve.
In some embodiments, the spectral curve noise reduction generating model is configured to form a generating countermeasure network with the discriminant model, the spectral curve noise reduction generating model being trained by: acquiring a spectrum curve sample set, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples; obtaining a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction; inputting the spectrum curve sample which is not noise-reduced and meets the preset condition into the spectrum curve noise-reduction generation model to generate a noise-reduced spectrum curve; a noise reduction sample set is obtained, wherein the noise reduction sample in the noise reduction sample set comprises the spectrum curve after noise reduction and the spectrum curve sample with noise reduction as a label; inputting the noise reduction sample into the discrimination model to obtain a judgment result of the noise reduction sample, wherein the judgment result at least comprises non-noise reduction and noise reduction; and obtaining a loss function of the generated countermeasure network according to the judging result, and respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
In some embodiments, the method for denoising a spectral curve further comprises: inputting any spectrum curve sample with the label not subjected to noise reduction into a classification model to judge whether any spectrum curve sample with the label not subjected to noise reduction belongs to noise, wherein the signal to noise ratio of the noise is lower than a preset signal to noise ratio; if the classification model judges that the noise does not belong to the noise, the label is any spectrum curve sample which is not noise-reduced, and the preset condition is met.
In some embodiments, the classification result further includes an environment category where the fluorescent molecule is located when the spectrum curve sample is obtained, the discrimination model includes a feature extraction layer, a maximum pooling layer, an average pooling layer, a first multi-layer perceptron and a second multi-layer perceptron, and the obtaining the determination result of the noise reduction sample includes: extracting features of the noise reduction sample through the feature extraction layer, and obtaining second features; the second feature is subjected to dimension reduction through the maximum pooling layer and the average pooling layer respectively so as to obtain a third feature and a fourth feature respectively; and inputting the third characteristic and the fourth characteristic into a first multi-layer perceptron to obtain a judgment result of the environment category of the noise reduction sample, and inputting the third characteristic and the fourth characteristic into a second multi-layer perceptron to obtain a judgment result of whether the noise reduction sample is noise reduction.
In some embodiments, when the initial spectral curve is acquired, it specifically includes: a sequence of positional images of dye molecules is acquired, the sequence of positional images being imaged by an optical device. And preprocessing the position image sequence to obtain the position information of dye molecules. Mapping the position information of dye molecules to the spectrum information of the dye molecules, and obtaining single-molecule spectrum data. And performing data cleaning on the single-molecule spectrum data to obtain the initial spectrum curve.
In this embodiment, after the optical device obtains the position image sequence of the dye molecule, the open source image processing software ImageJ is used to pre-process the position image sequence of the dye molecule, for example, the ImageJ modular plug-in unit specially designed for data processing of a single molecule positioning microscope (such as a light activated positioning microscope and a random optical reconstruction microscope) performs position detection on the dye molecule coordinates to obtain position information, and meanwhile, statistical parameters such as pixel position coordinates, fitting uncertainty, average photon number and the like are obtained. Then, in order to realize the mapping from the dye molecule position information to the spectrum information, a position-spectrum mapping matrix and a spectrum-pixel offset calibration curve need to be obtained in advance through calibration. The calibration steps are as follows: (1) The optical device is used for multispectral imaging of the fluorescent pellets, and a 611.5/10nm band-pass narrowband filter is used for acquiring the position corresponding to the 611.5nm wave band in the spectrum channel. (2) At least 6 position-spectrum matching points are selected and the position-spectrum coordinate transformation matrix thereof is calculated. (3) And selecting a spectrum of a certain calibrated fluorescent ball in a spectrum channel, acquiring positions of corresponding wavelengths of the calibrated fluorescent ball by using 6 narrowband filters with different wavebands, and calculating pixel offset distances corresponding to each waveband by taking 611.5nm wavelength as a zero point. (4) And fitting a spectrum-pixel offset calibration curve through a quadratic or cubic polynomial function. After the position-spectrum mapping matrix and the spectrum-pixel offset calibration curve are obtained, the mapping from the dye molecule position information to the spectrum information can be realized, and the spectrum value under the corresponding wavelength can be obtained. Further, according to the statistical characteristics of the spectrum curve, such as variation coefficient, average spectrum photon number, kurtosis and the like, the collected single-molecule spectrum data are cleaned, so that fluorescence emission spectrum information corresponding to each dye molecule in the image can be obtained, and an initial spectrum curve of one-dimensional data can be obtained.
The effect of the above-described noise reduction method of the spectrum profile is exemplified below.
The initial spectral curve is based on the result of super-resolution imaging of the spectral map of nile red, please refer to fig. 10, wherein the solid line avg.gan represents the spectral curve obtained in the present application, and the dashed line avg.raw represents the spectral curve obtained in the prior art. It can be seen that after the data processing is performed by the model of the embodiment of the invention, the lipid sequence state on the whole cell membrane is more uniform, the phenomenon of mutation does not occur, and the method is more in line with the actual situation of the cell. Meanwhile, the spectrum curve extracted by the traditional method is still influenced by noise, the position of the maximum emission peak is difficult to distinguish, and the emission peak at the current position can be roughly distinguished by the method of the embodiment of the invention to be about 620 nm. The method of the embodiment of the invention further verifies the superiority of the method compared with the traditional method, improves the time resolution of the traditional optical device, and provides a powerful research tool for spectrum detection on a single molecule level. After the model processing of the embodiment of the invention is carried out, the variance of the calculated maximum emission peak position of the nile red is smaller, and under the condition that the calculation result in the wavelength range of +/-8.5 nm of the central spectrum of the group trunk (true value) is specified as a correct value, the method can achieve the maximum emission peak identification accuracy of 99 percent, and compared with the traditional method, the method improves by 45 percent, and meanwhile, the method has different degrees of improvement in other different environment categories.
Some embodiments provide a computer readable storage medium having a program stored thereon that is executable by a processor to implement the above-described method of noise reduction of a spectral curve.
Those skilled in the art will appreciate that all or part of the functions of the various methods in the above embodiments may be implemented by hardware, or may be implemented by a computer program. When all or part of the functions in the above embodiments are implemented by means of a computer program, the program may be stored in a computer readable storage medium, and the storage medium may include: read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc., and the program is executed by a computer to realize the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the functions described above can be realized. In addition, when all or part of the functions in the above embodiments are implemented by means of a computer program, the program may be stored in a storage medium such as a server, another computer, a magnetic disk, an optical disk, a flash disk, or a removable hard disk, and the program in the above embodiments may be implemented by downloading or copying the program into a memory of a local device or updating a version of a system of the local device, and when the program in the memory is executed by a processor.
The foregoing description of the invention has been presented for purposes of illustration and description, and is not intended to be limiting. Several simple deductions, modifications or substitutions may also be made by a person skilled in the art to which the invention pertains, based on the idea of the invention.
Claims (13)
1. A training method of a spectral curve noise reduction generation model, wherein the spectral curve noise reduction generation model is used for forming a generation countermeasure network with a discrimination model, and the method comprises:
acquiring a spectrum curve sample set, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples;
obtaining a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction;
inputting the spectrum curve sample which is not noise-reduced and meets the preset condition into the spectrum curve noise-reduction generation model to generate a noise-reduced spectrum curve;
a noise reduction sample set is obtained, wherein the noise reduction sample in the noise reduction sample set comprises the spectrum curve after noise reduction and the spectrum curve sample with noise reduction as a label;
inputting the noise reduction sample into the discrimination model to obtain a judgment result of the noise reduction sample, wherein the judgment result at least comprises non-noise reduction and noise reduction; the classification result also comprises the environment category of fluorescent molecules when the spectrum curve sample is obtained, and the judgment result also comprises the environment category of the noise reduction sample;
And obtaining a loss function of the generated countermeasure network according to the judging result, and respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
2. The training method of a spectral curve noise reduction generation model of claim 1, further comprising:
inputting any spectrum curve sample with the label not subjected to noise reduction into a classification model to judge whether any spectrum curve sample with the label not subjected to noise reduction belongs to noise, wherein the signal to noise ratio of the noise is lower than a preset signal to noise ratio;
if the classification model judges that the noise does not belong to the noise, the label is any spectrum curve sample which is not noise-reduced, and the preset condition is met.
3. The method of training a model for generating spectral curve noise reduction according to claim 2, wherein the classification model includes a linear layer and a residual network, and the determining whether the label is any of the spectral curve samples that are not noise reduced includes:
carrying out dimension lifting on any spectrum curve sample with the label not subjected to noise reduction through the linear layer, and adjusting the spectrum curve sample after dimension lifting into two-dimensional data through a reshape function;
And inputting the two-dimensional data into a residual error network for feature extraction so as to output a judging result of whether the two-dimensional data is noise or not.
4. The training method of a spectral curve noise reduction generating model according to claim 1, wherein the spectral curve noise reduction generating model adopts a U-shaped network, the U-shaped network comprises an encoder and a decoder, and the generating of the noise reduced spectral curve comprises;
extracting features of the spectrum curve sample which is not noise-reduced and meets preset conditions by the aid of an encoder, and obtaining first features;
the decoder is used for carrying out up-scaling and information recovery on the first characteristics so as to obtain fusion data which are the same as the spectrum curve sample dimension of which the label is not noise-reduced and meets preset conditions;
and adding the fusion data and the spectrum curve sample which is not noise-reduced and meets the preset condition by the label to obtain the spectrum curve after noise reduction.
5. The training method of the spectral curve noise reduction generation model according to claim 1, wherein the discrimination model includes a feature extraction layer, a maximum pooling layer, an average pooling layer, a first multi-layer perceptron and a second multi-layer perceptron, and the obtaining the determination result of the noise reduction sample includes:
Extracting features of the noise reduction sample through the feature extraction layer, and obtaining second features;
the second feature is subjected to dimension reduction through the maximum pooling layer and the average pooling layer respectively so as to obtain a third feature and a fourth feature respectively;
and inputting the third characteristic and the fourth characteristic into a first multi-layer perceptron to obtain a judgment result of the environment category of the noise reduction sample, and inputting the third characteristic and the fourth characteristic into a second multi-layer perceptron to obtain a judgment result of whether the noise reduction sample is noise reduction.
6. The training method of a spectral curve noise reduction generator model according to any one of claims 1-5, further comprising:
according to the noise-reduced spectrum curve, obtaining total variation loss and/or regularization loss of a spectrum curve noise-reduction generation type model;
and adjusting parameters of the spectral curve noise reduction generation type model according to the total variation loss and/or regularization loss.
7. A training device for a spectral curve noise reduction generation model, wherein the spectral curve noise reduction generation model is used for forming a generation countermeasure network with a discrimination model, and the device comprises:
The sample acquisition module is used for acquiring a spectrum curve sample set, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples;
the labeling module is used for acquiring a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction;
the sample noise reduction module is used for inputting the spectrum curve sample which is not noise reduced and meets the preset condition into the spectrum curve noise reduction generation model so as to generate a spectrum curve after noise reduction;
the judging module is used for acquiring a noise reduction sample set, wherein the noise reduction sample in the noise reduction sample set comprises the noise reduction spectrum curve and the spectrum curve sample with the noise reduced label, the noise reduction sample is input into the judging model to obtain a judging result of the noise reduction sample, and the judging result comprises non-noise reduction and noise reduction; the classification result also comprises the environment category of fluorescent molecules when the spectrum curve sample is obtained, and the judgment result also comprises the environment category of the noise reduction sample;
and the training module is used for obtaining a loss function of the generated countermeasure network according to the judging result, respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
8. A method of noise reduction of a spectral curve, comprising:
acquiring an initial spectrum curve;
inputting the initial spectral curve into a spectral curve noise reduction generating model to obtain a spectral curve after the initial spectral curve noise reduction, wherein the spectral curve noise reduction generating model is trained by the training method according to any one of claims 1-6.
9. A method of noise reduction of a spectral curve, comprising:
acquiring an initial spectrum curve;
inputting the initial spectral curve into a pre-trained spectral curve noise reduction generating model, wherein inputting the initial spectral curve into the pre-trained spectral curve noise reduction generating model comprises the following steps:
extracting features of the initial spectrum curve through an encoder to obtain a first feature;
the decoder is used for carrying out up-scaling and information recovery on the first characteristic so as to obtain fusion data with the same dimension as the initial spectrum curve;
adding the fusion data with the initial spectrum curve to obtain a spectrum curve after noise reduction of the initial spectrum curve;
the spectral curve noise reduction generation model is used for forming a generation countermeasure network with the discrimination model, and is trained by the following modes:
Acquiring a spectrum curve sample set, wherein the spectrum curve sample set comprises a plurality of spectrum curve samples;
obtaining a label corresponding to the spectrum curve sample, wherein the label is used for labeling a classification result of the spectrum curve sample, and the classification result at least comprises non-noise reduction and noise reduction;
inputting the spectrum curve sample which is not noise-reduced and meets the preset condition into the spectrum curve noise-reduction generation model to generate a noise-reduced spectrum curve;
a noise reduction sample set is obtained, wherein the noise reduction sample in the noise reduction sample set comprises the spectrum curve after noise reduction and the spectrum curve sample with noise reduction as a label;
inputting the noise reduction sample into the discrimination model to obtain a judgment result of the noise reduction sample, wherein the judgment result at least comprises non-noise reduction and noise reduction; the classification result also comprises the environment category of fluorescent molecules when the spectrum curve sample is obtained, and the judgment result also comprises the environment category of the noise reduction sample;
and obtaining a loss function of the generated countermeasure network according to the judging result, and respectively adjusting parameters of the spectral curve noise reduction generation type model and the judging model at least according to the loss function until the generated countermeasure network converges, and obtaining a trained spectral curve noise reduction generation type model.
10. The method of noise reduction of a spectral curve according to claim 9, further comprising:
inputting any spectrum curve sample with the label not subjected to noise reduction into a classification model to judge whether any spectrum curve sample with the label not subjected to noise reduction belongs to noise, wherein the signal to noise ratio of the noise is lower than a preset signal to noise ratio;
if the classification model judges that the noise does not belong to the noise, the label is any spectrum curve sample which is not noise-reduced, and the preset condition is met.
11. The method for denoising a spectral curve according to claim 9, wherein the discrimination model comprises a feature extraction layer, a maximum pooling layer, an average pooling layer, a first multi-layer perceptron and a second multi-layer perceptron, and the obtaining a determination result of the denoising sample comprises:
extracting features of the noise reduction sample through the feature extraction layer, and obtaining second features;
the second feature is subjected to dimension reduction through the maximum pooling layer and the average pooling layer respectively so as to obtain a third feature and a fourth feature respectively;
and inputting the third characteristic and the fourth characteristic into a first multi-layer perceptron to obtain a judgment result of the environment category of the noise reduction sample, and inputting the third characteristic and the fourth characteristic into a second multi-layer perceptron to obtain a judgment result of whether the noise reduction sample is noise reduction.
12. The method for noise reduction of a spectral curve according to claim 8 or 9, wherein the obtaining an initial spectral curve comprises:
acquiring a position image sequence of dye molecules, wherein the position image sequence is obtained by imaging by an optical device;
preprocessing the position image sequence to obtain the position information of dye molecules;
mapping the position information of dye molecules to spectrum information of the dye molecules, and obtaining single-molecule spectrum data;
and performing data cleaning on the single-molecule spectrum data to obtain the initial spectrum curve.
13. A computer readable storage medium, characterized in that the medium has stored thereon a program executable by a processor to implement the method of any of claims 1-6 or 8-12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310449238.1A CN116152578B (en) | 2023-04-25 | 2023-04-25 | Training method and device for noise reduction generation model, noise reduction method and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310449238.1A CN116152578B (en) | 2023-04-25 | 2023-04-25 | Training method and device for noise reduction generation model, noise reduction method and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116152578A CN116152578A (en) | 2023-05-23 |
| CN116152578B true CN116152578B (en) | 2023-07-18 |
Family
ID=86352919
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310449238.1A Active CN116152578B (en) | 2023-04-25 | 2023-04-25 | Training method and device for noise reduction generation model, noise reduction method and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116152578B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163230A (en) * | 2018-06-15 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of image labeling method and device |
| CN110610156A (en) * | 2019-09-12 | 2019-12-24 | 河海大学 | Hyperspectral image classification method based on generation countermeasure network |
| CN111626317A (en) * | 2019-08-14 | 2020-09-04 | 广东省智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
| CN113011321A (en) * | 2021-03-17 | 2021-06-22 | 中南大学 | Spectral signal denoising method, system, terminal and readable storage medium based on joint dictionary |
| CN115326783A (en) * | 2022-10-13 | 2022-11-11 | 南方科技大学 | Raman spectrum preprocessing model generation method, system, terminal and storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021253316A1 (en) * | 2020-06-18 | 2021-12-23 | 深圳先进技术研究院 | Method and apparatus for training image noise reduction model, electronic device, and storage medium |
| KR102238300B1 (en) * | 2020-09-24 | 2021-04-12 | 국방과학연구소 | Method and apparatus for producing infrared spectrum |
| CN113239782B (en) * | 2021-05-11 | 2023-04-28 | 广西科学院 | Pedestrian re-recognition system and method integrating multi-scale GAN and tag learning |
-
2023
- 2023-04-25 CN CN202310449238.1A patent/CN116152578B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163230A (en) * | 2018-06-15 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of image labeling method and device |
| CN111626317A (en) * | 2019-08-14 | 2020-09-04 | 广东省智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
| CN110610156A (en) * | 2019-09-12 | 2019-12-24 | 河海大学 | Hyperspectral image classification method based on generation countermeasure network |
| CN113011321A (en) * | 2021-03-17 | 2021-06-22 | 中南大学 | Spectral signal denoising method, system, terminal and readable storage medium based on joint dictionary |
| CN115326783A (en) * | 2022-10-13 | 2022-11-11 | 南方科技大学 | Raman spectrum preprocessing model generation method, system, terminal and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116152578A (en) | 2023-05-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11644395B2 (en) | Multi-spectral imaging including at least one common stain | |
| US8639043B2 (en) | Classifying image features | |
| JP7273215B2 (en) | Automated assay evaluation and normalization for image processing | |
| EP2345995B1 (en) | Method for investigating live cells | |
| JP2022500744A (en) | Computer implementation methods, computer program products and systems for analysis of cell images | |
| Kong et al. | Analysis of hyperspectral fluorescence images for poultry skin tumor inspection | |
| EP3721406A1 (en) | Method of computing tumor spatial and inter-marker heterogeneity | |
| CN111433817A (en) | Generating a virtual stain image of an unstained sample | |
| Mahbub et al. | Statistically strong label-free quantitative identification of native fluorophores in a biological sample | |
| CN107111874A (en) | System and method for the coexpression analysis during fraction is calculated to be immunized | |
| JP2006526783A (en) | Color separation and detection of areas of interest in tissue samples | |
| CN115909006B (en) | Mammary tissue image classification method and system based on convolution transducer | |
| JP2004535569A (en) | Computer techniques for image pattern recognition of organic matter. | |
| Bayas et al. | Easy-DHPSF 2.0: open-source software for three-dimensional localization and two-color registration of single molecules with nanoscale accuracy | |
| CN116152578B (en) | Training method and device for noise reduction generation model, noise reduction method and medium | |
| Begelman et al. | Blind decomposition of transmission light microscopic hyperspectral cube using sparse representation | |
| WO2023061162A1 (en) | Systems and methods for label-free multi-histochemical virtual staining | |
| Chen et al. | Single color virtual H&E staining with In-and-Out Net | |
| Motyko et al. | Hyperspectral Technology in State of the Art Computer Vision Application | |
| Alom et al. | Deep learning-driven automatic nuclei segmentation of live-cell chromatin-sensitive partial wave spectroscopic microscopy imaging | |
| Cayuela López | Desarrollo y Aplicación de Métodos de Análisis de BioImágenes para Microscopía Óptica Avanzada | |
| CN117132814A (en) | Hyperspectral data classification method for grasslands | |
| Aymerich et al. | Fast phytoplankton classification from emission fluorescence spectra based on Self-Organizing Maps |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |