Self-training-based biomedical signal-oriented data amplification method and system

Abstract

The invention belongs to the technical field of data processing. The invention provides a data amplification method and system for biomedical signals based on self-training, and the method comprises the steps: obtaining a plurality of amplification samples corresponding to a real sample according to any preprocessed real sample and a pre-trained diffusion model; calculating the similarity between the real sample and each amplified sample, selecting a unique amplified sample most similar to the real sample according to the similarity, and calculating amplified samples corresponding to all the real samples; and according to the amplified samples and a downstream model preliminarily trained based on the real samples, calculating pseudo labels of the amplified samples, and according to comparison of the pseudo labels and the real labels, taking the unique amplified samples with consistent labels as final amplified samples. According to the method, the problems of insufficient biomedical signal data volume, insufficient diversity, data and label noise and the like are solved, and the performance of a downstream model is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data amplification method and system for biomedical signals based on self-training. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] Biomedical signals refer to signals spontaneously generated by human physiological processes, such as electrocardiogram (ECG), electroencephalogram (EEG), electromyogram (EMG), etc. They play a vital role in disease diagnosis, monitoring and treatment. With the advent of the big data era, biomedical signal data are gradually stored electronically, and artificial intelligence technology, especially deep neural networks, has become an important tool for processing these signals. Deep learning models, with their complex function layer design, can effectively process and analyze complex biomedical signals and have achieved remarkable results in signal classification and diagnosis tasks. However, the performance of deep neural networks usually depends on a large amount of labeled data, while biomedical signal data is not only scarce, but also has strict requirements for privacy protection, which undoubtedly poses a huge challenge to related research.

[0004] In order to overcome the problem of data scarcity, various data enhancement methods have been proposed to improve the performance of downstream tasks by expanding the data set. Most of the early biomedical signal enhancement methods borrowed techniques from the field of image processing and used basic operations such as simple flipping, segmentation, and recombination to generate new data samples. However, the samples generated by these methods are limited in diversity and cannot effectively improve the diversity of data, which easily leads to overfitting of downstream models. To solve this problem, in recent years, generative models such as generative adversarial networks (GANs) and variational autoencoders (VAEs) have been introduced into the study of biomedical signal enhancement. These generative models can generate more realistic signal samples, but they still face some challenges. For example, due to the limitations of training data and model design, the generated data often lacks sufficient diversity. In addition, models such as VAE and GAN may introduce structural distortions in the generation process, affecting the quality of enhanced data.

[0005] At present, diffusion model, as an emerging data enhancement method, has attracted widespread attention due to its ability to increase data diversity. In the field of biomedical signals, diffusion model has shown its potential in generating diverse signal samples. However, on the one hand, diffusion model may also face the risk of generating false or over-enhanced samples when enhancing biomedical signals. Especially in the signal processing process, over-emphasizing certain features of data distribution may introduce additional noise, thereby affecting the training and testing performance of downstream models; on the other hand, the accuracy of data labels has also become an important issue. The generation process may not perfectly retain the label information of the original sample, and the labels of the enhanced data may be affected by noise or inconsistency, further exacerbating the noise problem in the data. Summary of the invention

[0006] In order to address the deficiencies in the prior art, the present invention provides a data amplification method and system for biomedical signals based on self-training, which solves the problems of insufficient biomedical signal data volume, insufficient diversity, and data and label noise, significantly improves the performance of downstream models (such as classification models), assists doctors in completing more accurate classification tasks, and lays the foundation for the integration of deep learning and the medical and health fields.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a data amplification method for biomedical signals based on self-training.

[0008] A data augmentation method for biomedical signals based on self-training, comprising the following processes: Preprocessing a plurality of real biomedical sample signals to obtain a plurality of preprocessed real samples; According to any of the real samples and the pre-trained diffusion model, a plurality of amplified samples corresponding to the real sample are obtained; Calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; According to the unique amplified sample and the downstream model preliminarily trained based on the real sample, a pseudo label of the amplified sample is calculated, and according to the comparison between the pseudo label and the real label, the unique amplified sample with the consistent label is taken as the final amplified sample.

[0009] As a further limitation of the first aspect of the present invention, preprocessing the plurality of real biomedical sample signals comprises: A bandpass filter is used to remove high-frequency noise and low-frequency interference in the real biomedical sample signal, and the mid-frequency band signal is retained, and the retained mid-frequency band signal is normalized.

[0010] As a further limitation of the first aspect of the present invention, in the pre-trained diffusion model, a Transformer model with an attention mechanism is used to learn the inverse distribution to predict the amount of noise added at each time step.

[0011] As a further limitation of the first aspect of the present invention, calculating the similarity between the real sample and each of the amplified samples, and selecting a unique amplified sample that is most similar to the real sample according to the similarity, comprises: Calculate the cosine similarity between the two-dimensional embedding of the real sample and the two-dimensional embedding of each augmented sample respectively; Sorting according to the cosine similarity values, if any cosine similarity value is greater than a set threshold, selecting the amplified sample with the highest cosine similarity value as the only amplified sample of the true sample; If the values ​​of the cosine similarity are all less than or equal to the set threshold, multiple amplified samples are regenerated for the real sample, and the most similar unique amplified sample is reselected until the selection is completed.

[0012] As a further limitation of the first aspect of the present invention, a real training set consisting of each of the real samples is used to train a downstream model to obtain a preliminarily trained downstream model; based on the preliminarily trained downstream model and the unique amplified sample, a pseudo-label of the amplified sample is obtained, and the unique amplified sample whose pseudo-label is the same as the real label of the real sample is retained; all the retained unique amplified samples and all the real samples continue to be used as training data for the downstream model to train the downstream model until the training accuracy requirement of the downstream model is met.

[0013] In a second aspect, the present invention provides a data amplification system for biomedical signals based on self-training.

[0014] A data augmentation system for biomedical signals based on self-training, comprising: The data preprocessing unit is configured to: preprocess a plurality of real biomedical sample signals to obtain a plurality of preprocessed real samples; The data amplification unit is configured to: obtain a plurality of amplified samples corresponding to any one of the real samples and the pre-trained diffusion model; The sample selection unit is configured to: calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; The self-training amplification unit is configured to: calculate the pseudo-label of the amplified sample according to the unique amplified sample and the downstream model preliminarily trained based on the real sample, and according to the comparison between the pseudo-label and the real label, take the unique amplified sample with the consistent label as the final amplified sample.

[0015] As a further limitation of the second aspect of the present invention, in the self-training amplification unit, a real training set composed of each of the real samples is used to train the downstream model to obtain a preliminarily trained downstream model; based on the preliminarily trained downstream model and the unique amplified sample, a pseudo-label of the amplified sample is obtained, and the only amplified sample whose pseudo-label is the same as the real label of the real sample is retained; all the retained unique amplified samples and all the real samples continue to be used as training data for the downstream model to train the downstream model until the training accuracy requirement of the downstream model is met.

[0016] In a third aspect, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium; a processor adapted to execute a computer program; A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the self-training-based data amplification method for biomedical signals as described in the first aspect of the present invention is implemented.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program is suitable for being loaded by a processor and executing the self-training-based data amplification method for biomedical signals as described in the first aspect of the present invention.

[0018] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the self-training-based data amplification method for biomedical signals as described in the first aspect of the present invention.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention innovatively proposes a data amplification strategy for biomedical signals based on self-training. Through data preprocessing, data generation, data selection and self-training, it solves the problems of insufficient biomedical signal data volume, insufficient diversity, and data and label noise. It significantly improves the performance of downstream models (such as classification models), assists doctors in completing more accurate classification tasks, and lays the foundation for the integration of deep learning and the medical and health fields.

[0020] 2. The present invention removes inherent noise and interference in biomedical signals through data preprocessing, ensures the accuracy of subsequent processing, and uses bandpass filters and normalization operations to optimize the quality of original signals and improve processing accuracy.

[0021] 3. The present invention solves the problems of insufficient biomedical signal data volume and lack of diversity through the data generation stage, and adopts a diffusion model for data enhancement. By generating new signal samples, it not only effectively expands the data set, but also ensures the diversity of enhanced data and enhances the generalization ability of the model.

[0022] 4. In order to address the problems of false enhancement and data noise that may be introduced by the generative model, the present invention selects the most similar generated sample for each real data sample and ensures the diversity of samples, thereby reducing the interference of invalid data on model training and optimizing the quality of the data set.

[0023] 5. The present invention solves the label noise problem caused by inaccurate generated data labels through a self-training mechanism. It can continuously screen out samples with high-confidence labels and add them to the training of downstream models. In the iterative process, it not only optimizes the label selection but also improves the performance of downstream models.

[0024] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 A schematic diagram of a process flow of a data amplification method for biomedical signals based on self-training provided in Example 1 of the present invention; Figure 2 A schematic diagram of the process flow of the self-training phase provided in Embodiment 1 of the present invention; Figure 3 A schematic diagram of a data amplification system for biomedical signals based on self-training provided in Example 2 of the present invention; Figure 4 A schematic diagram of a computer device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0029] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0030] Embodiment 1: This implementation proposes a data amplification method for biomedical signals based on self-training, designs a self-training enhancement framework, and effectively alleviates the sparsity of biomedical signal data and the need to pay attention to privacy protection through a series of precisely designed processing stages, improves data diversity, reduces data noise caused by diffusion models, and optimizes label accuracy through self-training mechanisms. Specifically, in the data preprocessing stage, bandpass filters and normalization operations are used to eliminate noise interference; in the data generation stage, the diffusion model is used to increase the amount of data and ensure diversity; in the data selection stage, false enhancement is avoided and data noise is reduced by selecting the most similar generated samples for each real sample; in the self-training stage, label noise is reduced by selecting samples with high confidence labels, thereby continuously improving the performance of downstream models. The present invention effectively improves the accuracy and robustness of biomedical signal processing and promotes the application of deep learning technology in the field of medical and health care.

[0031] like Figure 1 As shown, the real biomedical signal samples are band-pass filtered and normalized to obtain a preprocessed data set. The diffusion model is used to generate data according to a 1:N ratio to initially generate a data set. The cosine similarity is calculated for each real sample and each N generated sample to determine whether there is a generated sample with a similarity greater than a threshold λ. If so, the generated sample with the highest similarity is selected to obtain the generated data. If not, the generated data is regenerated and the cosine similarity is calculated again for each real sample and each N generated sample.

[0032] The data augmentation method, more specifically, includes the following process: S1: Data preprocessing stage. It aims to alleviate the noise and interference inherent in biomedical signals to improve the accuracy of subsequent processing and ensure the quality of the original signal by removing irrelevant noise and abnormal data.

[0033] S2: Data generation stage. In order to solve the problem of insufficient data volume and lack of diversity of biomedical signal data, this stage uses a diffusion model for data enhancement. By generating new signal samples, it not only effectively expands the data volume, but also ensures the diversity of enhanced data.

[0034] S3: Data selection stage: This stage aims to alleviate the problems of false enhancement and data noise that may be caused by the generative model. It selects the most similar generated sample (i.e., the unique augmented sample) for each real data sample while maintaining the diversity of the data, thereby reducing the interference of invalid data on model training.

[0035] S4: Self-training stage. This stage solves the label noise problem that may be caused by generating data samples. Through the self-training mechanism, the framework can continuously screen out samples with accurate labels and put them into the training process of downstream models. In the process of cyclic iteration, it not only optimizes the label selection, but also improves the classification performance of downstream models.

[0036] In step S1 of this implementation method, specifically, it includes: in the process of collecting biomedical signals, various noises and interferences are often inevitably introduced. These non-target signals may significantly affect the accuracy of subsequent signal generation and classification tasks. In order to ensure the reliability of the experimental results, the signal was carefully preprocessed in the initial stage. First, a bandpass filter was used to effectively remove high-frequency noise and low-frequency interference in the ECG signal, while retaining the intermediate frequency band where the main signal components are located; then, the filtered signal was normalized to limit the numerical range of each sample to the interval [0,1], thereby providing a unified and clear data basis for subsequent analysis.

[0037] In step S2 of this implementation, specifically, it includes: dividing the processed biomedical signal data into a training set, a validation set, and a test set in proportion, and using the training set to train the generative model, while the validation set and the test set are only used to test the performance of the downstream model. Compared with other generative models, the data enhanced by the diffusion model can show higher diversity, so the diffusion model is selected to generate more diverse data, and the generation stage can be divided into two processes, namely the training process and the sampling process.

[0038] The first is the training process, which can be divided into the forward process and the reverse process. The forward process means adding noise to the real data sample until the original data becomes pure noise. The reverse process means continuous learning to gradually remove noise from the noise and restore the original data.

[0039] Specifically, suppose there is a biomedical signal sample ,pass time steps to gradually add noise, As the sample set, we continuously obtain , the size of the noise added at each step is determined by the hyperparameter Decide, specifically, as follows: (1); and They are Moment and The sample signal at each moment can be found in the forward process Just and time, so it can also be regarded as a Markov process. After derivation, Can be directly and The calculation results are: (2); in, and , Over time With the increase of t As the value of increases, the original data gradually becomes pure noise.

[0040] Then the reverse reasoning process is performed to restore the real data sample. Considering that if the noise addition process can be reversed from By sampling in, the real sample data can be restored from pure noise, but it cannot be simply deduced from the formula , so a neural network model is needed to learn such an inverse distribution : (3); )(4).

[0041] Considering the long sequence characteristics of biomedical signals, the Transformer architecture with attention mechanism is used as the neural network model of the diffusion model. Mean and variance in distributions , to restore the real data distribution, considering that for DDPM, under a fixed variance of The best result is obtained when , for which the variance is also fixed, and the mean is calculated as follows: (5); Among them, the neural network model can be used to predict the amount of noise added at each time step , the above formula can be rewritten as: )(6); The diffusion model can then be trained using the following loss function, namely: (7).

[0042] in, represents the noise parameter, and are the corresponding noise amounts respectively.

[0043] The sampling process mainly consists of a reverse process, which starts from the original noise and continuously removes the noise, thereby continuously sampling and generating many new samples. Specifically, starting from random noise, the trained neural network model is used to continuously predict the amount of noise added in the previous time step, and then continuously moving forward to obtain , for each real data sample, a diffusion model is used to generate The corresponding generated data samples.

[0044] In step S3 of this implementation, if Figure 2 As shown, specifically, it includes: considering that the diffusion model is prone to produce false enhancement and cause data noise, if all are input into the downstream model, it will be greatly affected by noise. Therefore, after obtaining the generated data, a data selection stage is added, and cosine similarity is used for selection. The specific process is as follows: First, using the trained diffusion model, for each real sample ,generate Data samples ; Secondly, flatten the three-dimensional time series data into two-dimensional time series data (flattened in the channel dimension) to prepare for calculating cosine similarity; then calculate the two-dimensional embedding of the real sample and Two-dimensional embedding of generated samples The cosine similarity between them is then used to sort the generated samples according to these cosine similarity values. If any of the generated samples has a similarity value exceeding the threshold , then select the sample with the highest similarity as the generated sample corresponding to the real sample. If no sample meets the conditions, regenerate the real sample samples, recalculate, compare and select the appropriate sample.

[0045] The specific implementation formula of the above process is as follows: (8); (9); (10); in, Indicates the index of the selected generated sample, and its value range is from 1 to . The above data selection process includes executing the above three formulas. If no samples consistent with the trend are found in each round of generation, the process is restarted and the above formula is re-executed until a suitable sample is selected. Usually, a trained diffusion model does not require multiple iterations to generate the required data samples. Through this data selection stage, the generated data that is most similar to the real data trend can be identified, thereby ensuring the quality of the generated samples and alleviating the data noise problem caused by false enhancement. Through this step, the same number of generated samples as the real data can be obtained to form a generated data set.

[0046] In step S4 of this implementation, specifically, it includes: after the data selection stage is completed, each real data sample will be matched with a closest generated sample, and this process is not simply repeated, ensuring the diversity of generated samples. In order to solve the label noise problem that may be caused by the generated data samples, that is, the labels of the generated data may be inaccurate, the framework continuously selects samples with higher label accuracy through the self-training mechanism and puts them into the training of downstream models. In this cycle, both label optimization and classification model performance are improved. The specific cycle process is as follows: First, use the real training data set to train the downstream model to obtain a preliminarily trained downstream model, which already has a certain classification capability. Then, use the trained downstream model to predict the labels of the generated data to obtain pseudo labels. These pseudo labels are compared with the original labels of the generated data. It is believed that the pseudo labels and original labels each have an accuracy of 50%. According to this criterion, the generated samples with consistent labels are retained, thereby effectively reducing label noise. Next, the retained generated data is mixed with the real data and continued to be used to train the downstream model, thereby further improving the classification performance of the model. This process not only enhances the downstream model's ability to process real data, but also improves the accuracy of the generated sample label prediction in the next cycle. The above steps are continuously iterated until the performance of the downstream model is no longer significantly improved. Through the continuous optimization process, a downstream model (such as a classification model) with continuously improved performance is finally obtained.

[0047] The above is the entire process for a patient sample instance. In practical applications, multiple real data samples are required to generate multiple generated data, and then carry out the subsequent selection process.

[0048] Embodiment 2: like Figure 3 As shown, this implementation provides a data amplification system for biomedical signals based on self-training, including: The data preprocessing unit is configured to: preprocess the plurality of real biomedical sample signals to obtain a plurality of preprocessed real samples; the specific process is as described in step S1 in embodiment 1, which will not be described again here; The data amplification unit is configured to: obtain a plurality of amplified samples corresponding to any one of the real samples and the pre-trained diffusion model; the specific process is as described in step S2 in embodiment 1, which will not be repeated here; The sample selection unit is configured to: calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; the specific process is described in step S3 of embodiment 1, which will not be repeated here; The self-training amplification unit is configured to: calculate the pseudo-label of the amplified sample based on the unique amplified sample and the downstream model preliminarily trained based on the real sample, and based on the comparison between the pseudo-label and the real label, take the unique amplified sample with the consistent label as the final amplified sample; the specific process is described in step S4 in Example 1, which will not be repeated here.

[0049] It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0050] According to another embodiment of the present application, the system described in this embodiment can be constructed, and the method of Example 1 of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0051] Embodiment 3: like Figure 4 As shown, this implementation provides an electronic device, which includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. The processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 may be connected via a bus or other means.

[0052] Among them, the communication interface 1002 is used to receive and send data, the computer-readable storage medium 1003 can be stored in the memory of the electronic device, the computer-readable storage medium 1003 is used to store a computer program, the computer program includes program instructions, and the processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003.

[0053] The processor 1001 (or CPU (Central Processing Unit)) is the computing core and control core of the electronic device, which is suitable for implementing one or more instructions, and specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions.

[0054] The processor 1001 is configured to execute the following process: Preprocessing multiple real biomedical sample signals to obtain multiple preprocessed real samples; the specific process is described in step S1 in embodiment 1, which will not be repeated here; According to any of the real samples and the pre-trained diffusion model, a plurality of amplified samples corresponding to the real sample are obtained; the specific process is as described in step S2 in embodiment 1, which will not be described again here; Calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; the specific process is described in step S3 of embodiment 1, which will not be repeated here; According to the unique amplified sample and the downstream model preliminarily trained based on the real sample, the pseudo-label of the amplified sample is calculated. According to the comparison between the pseudo-label and the real label, the unique amplified sample with the consistent label is taken as the final amplified sample. The specific process is described in step S4 in Example 1, which will not be repeated here.

[0055] Embodiment 4: This implementation provides a computer-readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The computer-readable storage medium provides a storage space that stores the processing system of the electronic device.

[0056] In addition, the storage space also stores one or more instructions suitable for being loaded and executed by the processor, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here may be a high-speed RAM memory, or a non-volatile memory, such as at least one disk memory; optionally, it may also be at least one computer-readable storage medium located away from the aforementioned processor.

[0057] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the following process: Preprocessing multiple real biomedical sample signals to obtain multiple preprocessed real samples; the specific process is described in step S1 in embodiment 1, which will not be repeated here; According to any of the real samples and the pre-trained diffusion model, a plurality of amplified samples corresponding to the real sample are obtained; the specific process is as described in step S2 in embodiment 1, which will not be described again here; Calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; the specific process is described in step S3 of embodiment 1, which will not be repeated here; According to the unique amplified sample and the downstream model preliminarily trained based on the real sample, the pseudo-label of the amplified sample is calculated. According to the comparison between the pseudo-label and the real label, the unique amplified sample with the consistent label is taken as the final amplified sample. The specific process is described in step S4 in Example 1, which will not be repeated here.

[0058] Embodiment 5: The present implementation provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs the following process: Preprocessing multiple real biomedical sample signals to obtain multiple preprocessed real samples; the specific process is described in step S1 in embodiment 1, which will not be repeated here; According to any of the real samples and the pre-trained diffusion model, a plurality of amplified samples corresponding to the real sample are obtained; the specific process is as described in step S2 in embodiment 1, which will not be described again here; Calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; the specific process is described in step S3 of embodiment 1, which will not be repeated here; According to the unique amplified sample and the downstream model preliminarily trained based on the real sample, the pseudo-label of the amplified sample is calculated. According to the comparison between the pseudo-label and the real label, the unique amplified sample with the consistent label is taken as the final amplified sample. The specific process is described in step S4 in Example 1, which will not be repeated here.

[0059] A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0060] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server, a data center, etc. that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data amplification method for biomedical signals based on self-training, characterized in that: The process includes: Preprocessing a plurality of real biomedical sample signals to obtain a plurality of preprocessed real samples; According to any of the real samples and the pre-trained diffusion model, a plurality of amplified samples corresponding to the real sample are obtained; Calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; According to the unique amplified sample and the downstream model preliminarily trained based on the real sample, a pseudo label of the amplified sample is calculated, and according to the comparison between the pseudo label and the real label, the unique amplified sample with the consistent label is taken as the final amplified sample.

2. The method for data amplification based on self-training for biomedical signals according to claim 1, characterized in that: Preprocess multiple real biomedical sample signals, including: A bandpass filter is used to remove high-frequency noise and low-frequency interference in the real biomedical sample signal, and the mid-frequency band signal is retained, and the retained mid-frequency band signal is normalized.

3. The data amplification method for biomedical signals based on self-training according to claim 1, characterized in that: In the pre-trained diffusion model, a Transformer model with an attention mechanism is used to learn the inverse distribution to predict the amount of noise added at each time step.

4. The method for data amplification based on self-training for biomedical signals according to claim 1, characterized in that: Calculating the similarity between the real sample and each of the amplified samples, and selecting a unique amplified sample that is most similar to the real sample according to the similarity, including: Calculate the cosine similarity between the two-dimensional embedding of the real sample and the two-dimensional embedding of each augmented sample respectively; Sorting according to the cosine similarity values, if any cosine similarity value is greater than a set threshold, selecting the amplified sample with the highest cosine similarity value as the only amplified sample of the true sample; If the values ​​of the cosine similarity are all less than or equal to the set threshold, multiple amplified samples are regenerated for the real sample, and the most similar unique amplified sample is reselected until the selection is completed.

5. The method for data amplification based on self-training for biomedical signals according to claim 1, characterized in that: The downstream model is trained using a real training set consisting of the real samples to obtain a preliminarily trained downstream model; a pseudo-label of the amplified sample is obtained based on the preliminarily trained downstream model and the unique amplified sample, and the unique amplified sample whose pseudo-label is the same as the real label of the real sample is retained; all the retained unique amplified samples and all the real samples continue to be used as training data for the downstream model to train the downstream model until the training accuracy requirement of the downstream model is met.

6. A data amplification system for biomedical signals based on self-training, characterized in that: include: The data preprocessing unit is configured to: preprocess a plurality of real biomedical sample signals to obtain a plurality of preprocessed real samples; The data amplification unit is configured to: obtain a plurality of amplified samples corresponding to any one of the real samples and the pre-trained diffusion model; The sample selection unit is configured to: calculate the similarity between the real sample and each of the amplified samples, select a unique amplified sample that is most similar to the real sample according to the similarity, and calculate the unique amplified samples corresponding to all the real samples; The self-training amplification unit is configured to: calculate the pseudo-label of the amplified sample according to the unique amplified sample and the downstream model preliminarily trained based on the real sample, and according to the comparison between the pseudo-label and the real label, take the unique amplified sample with the consistent label as the final amplified sample.

7. The data amplification system for biomedical signals based on self-training according to claim 6, characterized in that: In the self-training amplification unit, a real training set consisting of each of the real samples is used to train the downstream model to obtain a preliminarily trained downstream model; a pseudo-label of the amplified sample is obtained based on the preliminarily trained downstream model and the unique amplified sample, and the unique amplified sample whose pseudo-label is the same as the real label of the real sample is retained; all the retained unique amplified samples and all the real samples continue to be used as training data for the downstream model to train the downstream model until the training accuracy requirement of the downstream model is met.

8. A computer device, characterized in that: include: a processor and a computer readable storage medium; a processor adapted to execute a computer program; A computer-readable storage medium having a computer program stored therein, wherein the computer program, when executed by the processor, implements the self-training-based data amplification method for biomedical signals as described in any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is suitable for being loaded by a processor and executing the self-training-based data amplification method for biomedical signals as described in any one of claims 1 to 5.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method for data amplification based on self-training and oriented to biomedical signals is implemented as claimed in any one of claims 1 to 5.

Images