Data classification method for near-infrared functional imaging based on deep learning

Abstract

The present application relates to the technical field of near-infrared brain function imaging enhancement, in particular to a data classification method for near-infrared brain function imaging based on deep learning. The method acquires brain blood oxygen data through fNIRS and obtains an activation distribution map of a task; according to the position and optical density change value of the light source detector under each task, the emission signal intensity data of the light source emitter, the overall scattering interference degree of each light source detector under each task is obtained, and then the reference interference value of each pixel point in the activation distribution map is determined, the filter weight of each pixel point in the pixel point convolution kernel is corrected to obtain the corrected filter weight, and then the optimized activation distribution map is obtained, the near-infrared brain function imaging data classification model is trained, and the data classification of near-infrared brain function imaging is realized. The present application accurately obtains the corrected filter weight, accurately filters the activation distribution map, and effectively improves the accuracy of the near-infrared brain function imaging data classification.

Description

Technical Field

[0001] This invention relates to the field of near-infrared brain functional imaging enhancement technology, specifically to a data classification method for near-infrared brain functional imaging based on deep learning. Background Technology

[0002] Functional near-infrared spectroscopy (fNIRS) utilizes the excellent scattering of 600-900 nm near-infrared light by the main components of blood to obtain changes in oxyhemoglobin and deoxyhemoglobin during brain activity. This technique has advantages such as low cost, portability, noiselessness, non-invasiveness, and minimal sensitivity to subject movements during experiments. Currently, this technique is being applied to research in various fields, including higher cognition, developmental psychology, psychology of abnormalities, and more, in natural settings. In recent years, with the development of deep learning technology, techniques such as convolutional neural networks and recurrent neural networks have been increasingly applied to fNIRS data analysis. Deep learning methods can automatically extract high-dimensional, nonlinear features from raw signals, reducing manual intervention and effectively improving classification performance. It is known that fNIRS data classification models based on deep learning technology have shown good performance in applications such as motion intention recognition, cognitive task classification, and brain-computer interface control command discrimination.

[0003] However, in practice, the process of converting fNIRS signals into activation distribution maps is affected by factors such as photon scattering effects and signal overlap between the light source and detector. This leads to phenomena such as blurred boundaries of activation regions, signal diffusion, and signal contamination from adjacent brain regions in the generated activation distribution maps. Directly filtering the activation distribution maps can easily produce false edges and cannot accurately obtain spatial resolution, resulting in the inability to accurately divide the activation regions in the activation distribution maps, which seriously affects the performance of deep learning classification based on spatial distribution features. Summary of the Invention

[0004] To address the technical problem that inaccurate filtering of activation distribution maps leads to the inability to accurately delineate activation regions, thus affecting the performance of deep learning classification based on spatial distribution features, this invention aims to provide a deep learning-based data classification method for near-infrared brain functional imaging. The specific technical solution adopted is as follows:

[0005] This invention provides a deep learning-based data classification method for near-infrared brain functional imaging, the method comprising the following steps:

[0006] Brain blood oxygenation data was collected using fNIRS technology to obtain activation distribution maps for each task. The brain blood oxygenation data included the optical density change values ​​of each light source detector at each time step and the emission signal intensity data of each light source transmitter at each time step for each task.

[0007] Based on the position distribution and optical density variation of each light source detector under each task, the distance between each light source detector and the directly connected light source transmitter, the transmitted signal strength data of the directly connected light source transmitter, the distance between each light source detector and its preset neighboring light source detectors, and the change in the optical density variation of each light source detector, the overall scattering interference level of each light source detector under each task is obtained.

[0008] Based on the overall scattering interference level, the reference interference value of each pixel in the activation distribution map is determined. According to the reference interference value of each pixel corresponding to the convolution kernel of each pixel, the filtering weight of each pixel in the convolution kernel of each pixel is corrected. The corrected filtering weight of each pixel in the convolution kernel of each pixel is obtained, and then the optimized activation distribution map of each task is obtained.

[0009] A near-infrared brain functional imaging data classification model was trained based on an optimized activation distribution map to achieve near-infrared brain functional imaging data classification.

[0010] Furthermore, the method for obtaining the overall scattering interference level is as follows:

[0011] Based on the positional distribution of each light source detector under each task and the change in optical density at each time, the first scattering interference level of each light source detector under each task is obtained.

[0012] Based on the distance between each light source detector and the directly connected light source transmitter under each task, the transmitted signal strength data of the light source transmitter directly connected to each light source detector, and the distance between each light source detector and its preset neighboring light source detectors, the second scattering interference level of each light source detector under each task is obtained.

[0013] The activation level of each task is obtained by analyzing the changes in the optical density of each light source detector under each task.

[0014] Based on the activation level of each task and the first and second scattering interference levels of each light source detector under each task, the overall scattering interference level of each light source detector under each task is obtained; among them, the activation level is negatively correlated with the overall scattering interference level, while the first and second scattering interference levels are both positively correlated with the overall scattering interference level.

[0015] Furthermore, the method for obtaining the degree of the first scattering interference is as follows:

[0016] For any task and any light source detector under that task, fit the optical density change value of the light source detector at each moment within the task phase to a curve, and obtain the derivative of each optical density change value on the curve as the change analysis value.

[0017] The moment corresponding to the largest change analysis value is taken as the activation moment of the light source detector, and the optical density change value at the activation moment of the light source detector is taken as the target optical density change value of the light source detector.

[0018] The light source detector that is closest to the center of the brain activation region corresponding to the task is used as the activation center detector for the task.

[0019] The degree of delayed response of the light source detector is obtained based on the difference in activation time and distance between the light source detector and the activation center detector, as well as the change value of the target optical density of the light source detector.

[0020] The time corresponding to the maximum optical density change value of the activated center detector is taken as the target time. Based on the difference in optical density change value and distance between the activated center detector and the light source detector at the target time, the attenuation interference degree of the light source detector is obtained.

[0021] The product of the delay response of the light source detector and the attenuation interference level is taken as the first scattering interference level of the light source detector.

[0022] Furthermore, the method for obtaining the degree of delayed response is as follows:

[0023] The result of normalizing the absolute value of the difference between the activation time of the light source detector and the activation center detector is used as the delay reference value of the light source detector.

[0024] The normalized spatial distance between the light source detector and the activation center detector is used as the reference distance for the light source detector.

[0025] The delay response degree of the light source detector is obtained based on the delay reference value, reference distance, and target optical density change value of the light source detector; among them, the delay reference value and reference distance are negatively correlated with the delay response degree, while the target optical density change value is positively correlated with the delay response degree.

[0026] Furthermore, the method for obtaining the degree of interference attenuation is as follows:

[0027] The result of normalizing the difference between the optical density change values ​​of the activated center detector and the light source detector at the target time is used as the first characteristic value of the light source detector.

[0028] The ratio of the first characteristic value of the light source detector to the reference distance is used as the degree of optical density attenuation of the light source detector;

[0029] The difference between the optical density attenuation degree and the first preset constant is taken as the attenuation interference degree of the light source detector.

[0030] Furthermore, the method for obtaining the second scattering interference level is as follows:

[0031] For any task and any light source detector under that task, the light source emitter directly connected to the light source detector shall be used as the reference emitter of the light source detector.

[0032] The spatial distance between the light source detector and each reference transmitter is taken as the first distance for each reference transmitter;

[0033] The average of the product of the first distance and the transmitted signal strength data of each reference transmitter is taken as the emission influence of the light source detector;

[0034] The nearest light source detector in each direction to which the light source detector is connected is taken as the preset neighbor light source detector of the light source detector;

[0035] The result of normalizing the spatial distance between the light source detector and each preset neighboring light source detector is used as the reference weight for each preset neighboring light source detector.

[0036] The average of the product of the reference weight and the emission influence of each preset neighborhood light source detector is taken as the detection influence of that light source detector.

[0037] The product of the emission influence and the detection influence of the light source detector is taken as the second scattering interference level of the light source detector.

[0038] Furthermore, the method for obtaining the activation level is as follows:

[0039] For any task and any light source detector under that task, the duration of the increasing interval to which the largest optical density change value in the curve corresponding to the light source detector belongs is taken as the first duration of the light source detector.

[0040] The ratio of the maximum optical density change value of the light source detector to the first time duration is used as the activation analysis value of the light source detector;

[0041] The normalized result of the mean of the activation analysis values ​​of all light source detectors in the brain activation region corresponding to the task is taken as the activation level of the task.

[0042] Furthermore, the method for obtaining the reference interference value is as follows:

[0043] For any pixel in the activation distribution map of any task, the overall scattering interference level of the light source detector of that task that is closest to that pixel is used as the reference interference value of that pixel.

[0044] Furthermore, the method for obtaining the modified filter weights is as follows:

[0045] For any pixel in the activation distribution map, the result of negatively correlating the reference interference values ​​of each pixel in the convolution kernel of that pixel is used as the first correction weight of each pixel in the convolution kernel of that pixel.

[0046] The difference between each pixel in the convolution kernel of that pixel and the reference interference value of that pixel is used as the reference interference difference of each pixel in the convolution kernel of that pixel.

[0047] The reference interference difference of each pixel in the convolution kernel of that pixel is multiplied by the preset analysis ratio value, and then added to the second preset constant. The result is used as the second correction weight of each pixel in the convolution kernel of that pixel.

[0048] The product of the first corrected weight, the second corrected weight, and the initial filter weight of each pixel in the convolution kernel of that pixel is used as the corrected filter weight of each pixel in the convolution kernel of that pixel.

[0049] Furthermore, the method for obtaining the optical density change value is as follows:

[0050] For any given light source detector and any given time, the difference between the optical density of that light source detector at that time and the baseline optical density is taken as the change in optical density of that light source detector at that time.

[0051] The present invention has the following beneficial effects:

[0052] This invention first obtains the overall scattering interference level of each light source detector under each task based on the position distribution and optical density variation value of each light source detector, the distance between each light source detector and the directly connected light source transmitter, the emission signal strength data of the directly connected light source transmitter, the distance between each light source detector and its preset neighboring light source detectors, and the change in optical density variation value of each light source detector. This accurately reflects the scattering interference experienced by each light source detector under each task, which is beneficial for subsequent accurate filtering and correction of the activation distribution map for each task, enabling accurate identification of the brain activation region corresponding to each task in the activation distribution map. Then, based on the overall scattering interference level, a reference interference value is determined for each pixel in the activation distribution map, accurately reflecting the scattering interference experienced by each pixel. Furthermore, based on... The reference interference value corresponding to each pixel's convolution kernel is used to correct the filtering weights of each pixel in the convolution kernel, thus obtaining the corrected filtering weights for each pixel. This allows for accurate filtering of the activation distribution map for each task, effectively avoiding the problem of blurred boundaries of activation regions in the activation distribution map. Consequently, the optimized activation distribution map for each task is accurately obtained, effectively improving the recognition accuracy of spatial gradients in the activation distribution map. At the same time, it enhances boundary sharpness, eliminates "edge smoothing" artifacts, and effectively improves the accuracy of feature extraction of the activation region corresponding to each task. Furthermore, based on the optimized activation distribution map, the near-infrared brain functional imaging data classification model is accurately trained, accurately achieving data classification of near-infrared brain functional imaging, and effectively improving the data classification accuracy of near-infrared brain functional imaging. Attached Figure Description

[0053] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A schematic flowchart of a data classification method for near-infrared brain functional imaging based on deep learning, provided as an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of the position structure of the light source emitter, light source detector, and channel provided in one embodiment of the present invention;

[0056] Figure 3 This is an activation distribution map corresponding to a task provided in one embodiment of the present invention;

[0057] Figure 4A flowchart illustrating a method for obtaining the overall scattering interference level according to an embodiment of the present invention;

[0058] Figure 5 This is a structural diagram of a data classification system for near-infrared brain functional imaging based on deep learning, provided in one embodiment of the present invention.

[0059] Figure 6 This is a schematic diagram of a computer device provided according to an embodiment of the present invention. Detailed Implementation

[0060] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the deep learning-based data classification method for near-infrared brain functional imaging proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0062] The following description, in conjunction with the accompanying drawings, details the specific scheme of the data classification method for near-infrared brain functional imaging based on deep learning provided by this invention.

[0063] Example 1:

[0064] This invention proposes a deep learning-based data classification method for near-infrared brain functional imaging. Please refer to [link / reference]. Figure 1 The diagram illustrates a schematic flowchart of a deep learning-based data classification method for near-infrared brain functional imaging, according to an embodiment of the present invention. The method includes the following steps:

[0065] Step S1: Collect brain blood oxygenation data using fNIRS technology to obtain the activation distribution map for each task; wherein, the brain blood oxygenation data includes the optical density change value of each light source detector at each time step and the transmission signal intensity data of each light source transmitter at each time step under each task.

[0066] Specifically, the principle of functional near-infrared spectroscopy (fNIRS) for brain functional imaging is similar to that of functional magnetic resonance imaging (fMRI), namely, that neural activity in the brain leads to local hemodynamic changes. fNIRS mainly utilizes the difference in the absorption rate of near-infrared light with wavelengths of 600-900 nm between oxyhemoglobin and deoxyhemoglobin in brain tissue to directly detect hemodynamic activity in the cerebral cortex in real time. By observing these hemodynamic changes, the neural activity of the brain can be inferred. Both fNIRS and fMRI are well-known techniques and will not be elaborated further.

[0067] Near-infrared spectroscopy (fNIRS) is a non-invasive brain imaging method that primarily utilizes the propagation and absorption characteristics of near-infrared light in the scalp, skull, and cerebral cortex to monitor changes in intracranial blood oxygen dynamics parameters in real time, indirectly reflecting the activity state of brain functional areas. It is known that different tasks correspond to different brain functional states, meaning different tasks activate different brain regions. For example, the language naming task corresponds to Broca's area in the left inferior frontal gyrus, while the speech auditory comprehension task corresponds to Wernicke's areas in the bilateral superior temporal gyruses. To analyze the information of each task in its corresponding brain activation region, this embodiment uses a single user as an example for analysis; all subsequent mentions of "user" will refer to this user. The user wears an fNIRS optical headband. Following international 10–20 system or brain function localization requirements, light source emitters and detectors are arranged to form several channels. The channel distance is generally 2.5–3.5 cm, but the implementer can set the channel distance according to the actual situation; no limit is imposed here. This ensures that the near-infrared light from the light source emitter penetrates to the surface of the cerebral cortex. Simultaneously, the channel number and scalp coordinates, as well as the scalp coordinates of each light source emitter and detector, are recorded. Figure 2 This is a schematic diagram showing the location and structure of the light source emitter, light source detector, and channel. Figure 2 The blue sphere represents the light source emitter, the yellow sphere represents the light source detector, the purple edge represents the channel, and the data on the purple edge is the channel encoding. In this embodiment, the sampling frequency of the fNIRS optical head-mounted device is set to 10Hz, and the wavelengths are 760nm and 850nm. Implementers can set the sampling frequency and wavelength of the fNIRS optical head-mounted device according to the actual situation, and no limitation is made here.

[0068] During each task, the fNIRS optical head-mounted device records the near-infrared light intensity received by each channel in real time, generating raw light signal intensity data. This raw light signal intensity data undergoes preprocessing such as baseline correction, filtering, motion artifact detection, and removal to ensure signal quality. Based on the modified Beer-Lambert law, the changes in oxyhemoglobin and deoxyhemoglobin are calculated, and the activation intensity t-value for each channel is calculated by comparing the task and resting states. According to the scalp coordinates corresponding to each channel, the t-values ​​are mapped to scalp spatial coordinates to generate a scattered activation intensity matrix. Distance-weighted interpolation is used to reconstruct a continuous activation map, initially preserving edges to generate a continuous two-dimensional pseudo-color layer. Using a standard cerebral cortex model to register the measurement point coordinates, the two-dimensional pseudo-color layer is superimposed on the three-dimensional surface model of the brain to generate an activation distribution map, such as... Figure 3 The diagram shown is the activation distribution map for a given task. The preprocessing of the original optical signal intensity data, the correction of the Beer-Lambert theorem, distance-weighted interpolation, and the generation of the activation distribution map are all well-known techniques and will not be elaborated upon further.

[0069] Therefore, this embodiment uses fNIRS technology to collect cerebral blood oxygenation data and obtain an activation distribution map for each task. The cerebral blood oxygenation data includes the optical density change value of each light source detector at each time step under each task and the emission signal intensity data of each light source transmitter at each time step. The method for obtaining the optical density change value is as follows: for any light source detector and any time step, the difference between the optical density of that light source detector at that time step and the baseline optical density (known) is taken as the optical density change value of that light source detector at that time step, accurately reflecting the rate at which light is absorbed by hemoglobin in tissues. Optical density and baseline optical density are both well-known and will not be described further.

[0070] It should be noted that one light source detector can correspond to multiple light source emitters, meaning it can directly connect to multiple light source emitters. However, the multiple light source emitters corresponding to one light source detector will definitely emit light signals at different time periods, and these time periods will not overlap but will be continuous. For example, suppose a light source detector corresponds to two light source emitters, namely light source emitter A and light source emitter B. The light signal emission time period of light source emitter A is 0-10ms, then the light signal emission time period of light source emitter B is 10-20ms. Furthermore, the emitted signal intensity data of each light source emitter remains unchanged within the same task.

[0071] Step S2: Based on the position distribution and optical density change value of each light source detector under each task, the distance between each light source detector and the directly connected light source transmitter, the transmission signal strength data of the directly connected light source transmitter, the distance between each light source detector and its preset neighboring light source detectors, and the change of optical density change value of each light source detector, obtain the overall scattering interference degree of each light source detector under each task.

[0072] Specifically, in reality, near-infrared light experiences significant scattering in the scalp, skull, and brain tissue. The propagation path of the emitted light deviates after entering the brain tissue, resulting in the actual coverage area of ​​the light signal received by the light source detector being much larger than the ideal target area. Furthermore, in multi-channel deployments, light source detectors in adjacent channels may receive residual scattered light from multiple light source emitters, causing spatial crossover of channel signals, especially in densely populated areas where signal interference is severe. These factors lead to unclear boundaries between brain activation regions, reducing the accuracy of zoning in the activation distribution map and consequently affecting the data classification model's ability to distinguish differences in brain activation regions across different tasks.

[0073] Ideally, the near-infrared light emitted by the light source emitter should penetrate the skin, skull, or brain tissue within a preset optical path (channel), ultimately acquiring an absorbed and reduced-scattering light signal at the corresponding light source detector. However, due to the strong scattering characteristics of biological tissue, emitted photons undergo multiple random scatterings within the brain tissue. The actual distribution of photons reaching the light source detector extends beyond the target area, forming a diffusion tail. This results in the signal recorded by the light source detector simultaneously containing signals from the corresponding channel's activation area, residual signals from neighboring channels, and multiple scattering components from distant background photons. This diffusion leads to a high signal intensity at the center and a gradual decrease at the edges within the brain activation area. Simultaneously, the gradient between adjacent brain activation areas is small, i.e., the boundary bandwidth is large. This directly causes numerous slow transition zones at the edges of brain activation areas in the activation distribution map, resulting in boundary blurring. To ensure more accurate information in the brain activation areas corresponding to each task, and to accurately filter the activation distribution map for each task to eliminate the edge blurring problem, this embodiment needs to analyze the scattering interference experienced by each light source detector under each task.

[0074] Under normal circumstances, during a task, the optical density change values ​​of the light source detectors within the brain activation region corresponding to that task should increase synchronously, while the optical density change values ​​of other light source detectors should remain almost unchanged or show a slight delay. However, when severe scattering and diffusion occur, the optical density change values ​​of other light source detectors closer to the brain activation region corresponding to the task are more likely to increase almost synchronously, eliminating the time delay and thus reducing spatial discrimination. Furthermore, under normal circumstances, the peak value of the optical density change value of the light source detectors within the brain activation region corresponding to the task is relatively high, and the peak value should decrease more significantly with increasing distance from the brain activation region. When scattering and diffusion interference exists, the peak value of the optical density change value of the light source detectors decreases slowly with increasing distance from the brain activation region, widening the spatial transition band. Therefore, this embodiment can preliminarily determine the scattering interference situation of each light source detector under each task by analyzing the positional distribution of each light source detector and its optical density change value at each time point.

[0075] On the other hand, it is known that the greater the distance between the light source emitter and the light source detector, the more tissue layers photons traverse, resulting in more scattering paths and a stronger scattering effect on the light source detector. Simultaneously, scattering interference is also related to the emission signal strength of the light source emitter; the stronger the emission signal power, the greater the total number of residual scattered photons along a single path, leading to stronger scattering interference. Furthermore, in densely deployed multi-channel areas, due to the overlap of the sensitive volumes of the emission-detection paths, some scattered photons from the active regions of adjacent channel light source emitters may cross the channel's sensitive region and enter the side light source detector. This superposition effect of scattered light signals is particularly significant when adjacent active regions are close to or contain high-intensity active regions. Therefore, this embodiment can determine the direct emission scattering impact on each light source detector under each task by analyzing the distance between each light source detector and the directly connected light source emitter, as well as the emission signal strength data of the light source emitter directly connected to each light source detector. Then, based on the distance between each light source detector and its preset neighboring light source detectors, and the direct emission impact on its preset neighboring light source detectors, the indirect emission scattering impact on each light source detector under each task can be further determined.

[0076] Furthermore, considering that different tasks activate different brain regions, and that the scalp, skull thickness, blood flow distribution, and local tissue scattering coefficients vary across these regions, the photon scattering path length, directionality, and multiple scattering probability differ. Therefore, when evaluating the scattering interference of each light source detector, analysis needs to be performed for each task. It is known that high-intensity cognitive tasks (such as working memory and verbal naming) show significant changes in blood oxygenation and short scattering trails, while low-intensity tasks (such as simple motor skills and visual recognition) show small changes in blood flow and long scattering trails, making it easier for the light source detector to receive residual light from distant areas. In high-intensity activation tasks, the optical density change value of the activated region increases relatively quickly and has a high peak value, resulting in relatively less scattering interference during detection. Conversely, for low-intensity or broad-spectrum tasks, the optical density change value of the light source detector increases slowly and has a low peak value, resulting in relatively greater scattering interference. Therefore, this embodiment can determine the activation level of each task by analyzing the changes in the optical density change value of each light source detector under each task, thus more accurately obtaining the scattering interference of each light source detector under each task.

[0077] Therefore, this embodiment obtains the overall scattering interference level of each light source detector under each task based on the position distribution and optical density variation value of each light source detector, the distance between each light source detector and the directly connected light source transmitter, the transmission signal strength data of the directly connected light source transmitter, the distance between each light source detector and its preset neighboring light source detectors, and the variation value of the optical density of each light source detector. The greater the overall scattering interference level, the greater the scattering interference level experienced by the corresponding light source detector under the corresponding task.

[0078] Preferably, in one feasible embodiment, the method for obtaining the overall scattering interference level is described in [reference needed]. Figure 4 The document presents a flowchart of a method for obtaining the overall scattering interference level provided in this embodiment. The method includes the following steps:

[0079] Step S201: Based on the position distribution of each light source detector under each task and the optical density change value at each time, obtain the first scattering interference degree of each light source detector under each task.

[0080] The greater the degree of the first scattering interference, the greater the scattering interference experienced by the corresponding light source detector under the corresponding task.

[0081] In one possible implementation of this embodiment, the method for obtaining the first scattering interference level is as follows: For any task and any light source detector under that task, the optical density change value of the light source detector at each moment within the task phase is fitted into a curve according to the time sequence, and the derivative of each optical density change value on the curve is used as the change analysis value; wherein, the fitting curve and the method for obtaining the derivative are well-known techniques and will not be described in detail. The larger the change analysis value, the greater the change in the corresponding optical density change value, which indirectly indicates that the change in the optical signal in the channel corresponding to the light source detector is stronger at the corresponding moment, and the more likely the corresponding moment is the activation moment of the light source detector. Therefore, in this embodiment, the moment corresponding to the largest change analysis value is taken as the activation moment of the light source detector. It should be noted that if there are multiple largest change analysis values, the moment corresponding to the first occurrence of the largest change analysis value is taken as the activation moment of the light source detector. At the same time, the optical density change value at the activation moment of the light source detector is taken as the target optical density change value of the light source detector. The larger the target optical density change value, the stronger the signal strength corresponding to the light source detector, which indirectly indicates that the light source detector is more likely to have scattering interference;

[0082] It is known that the center of the brain activation region corresponding to this task is the location with the highest activation intensity and the least scattering interference. To quantify scattering interference and optimize spatial resolution, this embodiment uses the light source detector closest to the center of the brain activation region corresponding to this task as the activation center detector, i.e., the default location of the brain activation region. The closer the distance between the light source detector and the activation center detector, the more consistent the activation times, and the larger the target optical density change value of the light source detector, the shorter the delay in the response to the optical density change value. Furthermore, a larger optical density change value indirectly indicates that the light source detector possesses more scattering interference characteristics. Therefore, this embodiment obtains the delay response degree of the light source detector based on the difference in activation times and distance between the light source detector and the activation center detector, as well as the target optical density change value of the light source detector. A larger delay response degree indicates a shorter and more significant response delay to the optical density change value, indirectly reflecting that the light source detector is more likely to have heat dissipation interference.

[0083] The method for obtaining the degree of delayed response is as follows: the absolute value of the difference between the activation times of the light source detector and the activation center detector is normalized and used as the delay reference value of the light source detector; the spatial distance between the light source detector and the activation center detector is normalized and used as the reference distance of the light source detector. The smaller the delay reference value and the reference distance, the more timely the response of the light source detector, indirectly reflecting the greater the degree of scattering interference received by the light source detector. In this embodiment, the absolute value of the difference between the activation times is normalized using the norm normalization function, and the spatial distance between the light source detector and the activation center detector is normalized using the sigmoid function. It is known that the larger the target optical density change value of the light source detector, the more timely the response of the light source detector can be indicated from the side. Therefore, this embodiment obtains the degree of delayed response of the light source detector based on the delay reference value, the reference distance, and the target optical density change value; wherein the delay reference value and the reference distance are negatively correlated with the degree of delayed response, and the target optical density change value is positively correlated with the degree of delayed response. Specifically, in this embodiment, the product of the delay reference value and the reference distance of the light source detector and the sum of the product and the third preset constant are used as the first result. Then, the ratio of the target optical density change value of the light source detector to the first result is used as the delay response degree of the light source detector. The third preset constant is a non-negative number. In this embodiment, the third preset constant is set to 1 to avoid a denominator of 0. The implementer can set the value of the third preset constant according to the actual situation; no limitation is imposed here.

[0084] Furthermore, considering that under ideal conditions, i.e., when there is no scattering, the optical density change value should decrease linearly with distance, in order to more accurately show the change of optical density change value with distance, this embodiment takes the time corresponding to the maximum optical density change value of the activated center detector as the target time. Then, based on the difference in optical density change value between the activated center detector and the light source detector at the target time and the distance, the attenuation interference degree of the light source detector is obtained. The greater the attenuation interference degree, the more significant the spatial diffusion characteristics of the light signal at the location of the light source detector, and the greater the scattering interference experienced by the light source detector.

[0085] The method for obtaining the degree of interference attenuation is as follows: the difference between the optical density change value of the activated center detector and the light source detector at the target time is normalized and used as the first characteristic value of the light source detector. In this embodiment, the difference between the optical density change value of the activated center detector and the light source detector at the target time is normalized using the sigmoid function. Then, the ratio of the first characteristic value of the light source detector to the reference distance is used as the degree of optical density attenuation of the light source detector. It should be noted that, in order to avoid the reference distance being 0, this embodiment uses the sum of the reference distance and the fourth preset constant as the denominator. The fourth preset constant is a non-negative number. In this embodiment, the fourth preset constant is set to 0.1 to avoid the denominator being 0. The implementer can set the size of the fourth preset constant according to the actual situation, which is not limited here. When the degree of optical density attenuation is closer to 1, it indicates that the light source detector is less likely to have scattering interference. Therefore, in this embodiment, the first preset constant is set to 1, and the absolute value of the difference between the degree of optical density attenuation and the first preset constant is used as the degree of interference attenuation of the light source detector.

[0086] It is known that a greater degree of delayed response and a greater degree of attenuation interference both indicate a greater degree of scattering interference experienced by the light source detector. In order to accurately characterize the scattering interference experienced by the light source detector as reflected by the change in optical density, this embodiment uses the product of the delay response and the attenuation interference of the light source detector as the first degree of scattering interference of the light source detector.

[0087] At this point, the initial scattering interference level of each light source detector under each task is obtained.

[0088] Step S202: Based on the distance between each light source detector and the directly connected light source transmitter under each task, the transmission signal strength data of the light source transmitter directly connected to each light source detector, and the distance between each light source detector and its preset neighboring light source detectors, obtain the second scattering interference level of each light source detector under each task.

[0089] The greater the degree of the second scattering interference, the greater the scattering interference experienced by the corresponding light source detector under the corresponding task.

[0090] In one possible implementation of this embodiment, the method for obtaining the second scattering interference level is as follows: For any task and any light source detector under that task, the light source transmitter directly connected to the light source detector is taken as the reference transmitter of the light source detector; the spatial distance between the light source detector and each reference transmitter is taken as the first distance of each reference transmitter; the larger the first distance, the greater the scattering effect on the light source detector; at the same time, the greater the transmitted signal strength data of the reference transmitter, the greater the scattering effect on the light source detector. Therefore, in this embodiment, the average value of the product of the first distance of each reference transmitter and the transmitted signal strength data is taken as the emission influence level of the light source detector; the greater the emission influence level, the more obvious the scattering interference of the received optical signal caused by the direct influence of the reference transmitter on the light source detector;

[0091] Considering that the light source detector may also be indirectly affected by scattered signals from other surrounding channels, this embodiment designates the nearest light source detector in each direction connected to the light source detector as a preset neighboring light source detector. The greater the emission influence of a preset neighboring light source detector, the stronger the signal scattering in the channel corresponding to the preset neighboring light source detector, which indirectly indicates that the received signal of the light source detector is more likely to be affected. At the same time, the greater the spatial distance between the preset neighboring light source detector and the light source detector, the more scattering processes the scattering influence signal corresponding to the preset neighboring light source detector undergoes, and the stronger the scattering interference generated. Therefore, in this embodiment, the result of normalizing the spatial distance between the light source detector and each preset neighboring light source detector is used as the reference weight of each preset neighboring light source detector. Specifically, this embodiment normalizes the spatial distance between the light source detector and each preset neighboring light source detector using a normalization function. To accurately characterize the indirect scattering influence on the light source detector, the average of the product of the reference weight of each preset neighboring light source detector and the emission influence level is used as the detection influence level of the light source detector. The greater the detection influence level, the greater the scattering interference of the received light signal caused by the indirect influence of other light source emitters on the light source detector.

[0092] In order to accurately characterize the scattering interference of the light signal received by the light source detector, this embodiment uses the product of the emission influence degree and the detection influence degree of the light source detector as the second scattering interference degree of the light source detector.

[0093] At this point, the second scattering interference level of each light source detector under each task is obtained.

[0094] Step S203: Obtain the activation level of each task based on the change in optical density of each light source detector under each task.

[0095] The greater the activation level, the smaller the scattering interference of the corresponding task, which indirectly indicates that the scattering interference experienced by each light source detector under the task is smaller.

[0096] In one possible implementation of this embodiment, the activation level is obtained as follows: For any task and any light source detector under that task, the duration of the increasing interval to which the largest optical density change value in the curve corresponding to the light source detector belongs is taken as the first duration of the light source detector. It should be noted that if there are multiple largest optical density change values, the duration of the increasing interval to which the first largest optical density change value belongs is taken as the first duration of the light source detector. It should be noted that the method for obtaining the increasing interval to which the largest optical density change value belongs is existing content and will not be elaborated further. The ratio of the largest optical density change value of the light source detector to the first duration is taken as the activation analysis value of the light source detector; the larger the activation analysis value, the greater the activation level of the task. To accurately characterize the activation level of the task, the mean of the activation analysis values ​​of all light source detectors in the brain activation region corresponding to the task is normalized, and this result is taken as the activation level of the task. In this embodiment, the mean of the above activation analysis values ​​is normalized using the norm normalization function.

[0097] At this point, the activation level of each task is obtained.

[0098] Step S204: Based on the activation level of each task and the first and second scattering interference levels of each light source detector under each task, obtain the overall scattering interference level of each light source detector under each task; wherein, the activation level is negatively correlated with the overall scattering interference level, and the first and second scattering interference levels are both positively correlated with the overall scattering interference level.

[0099] It is known that the higher the activation level of a task, the lower the scattering interference experienced by each light source detector under that task; conversely, the higher both the first and second scattering interference levels of a light source detector under that task, the greater the scattering interference experienced by that light source detector. Therefore, this embodiment obtains the overall scattering interference level of each light source detector under each task based on the activation level of each task and the first and second scattering interference levels of each light source detector under each task; wherein, the activation level is negatively correlated with the overall scattering interference level, while the first and second scattering interference levels are both positively correlated with the overall scattering interference level.

[0100] The specific process for obtaining the overall scattering interference level is as follows: First, the product of the first and second scattering interference levels of each light source detector under each task is normalized and used as the scattering interference coefficient of each light source detector under each task. Then, the product of the scattering interference coefficient of each light source detector under each task and the negative correlation result of the activation level of each task is used as the overall scattering interference level of each light source detector under each task. In this embodiment, the product of the first and second scattering interference levels is normalized using the norm normalization function; the difference between the constant 1 and the activation level of each task is used as the negative correlation result of the activation level of each task.

[0101] This allows us to obtain the overall scattering interference level of each light source detector under each task, which is beneficial for subsequent accurate filtering of the activation distribution map of each task, enabling more accurate identification of the corresponding region of each task in the activation distribution map.

[0102] Step S3: Determine the reference interference value of each pixel in the activation distribution map based on the overall scattering interference level. According to the reference interference value of each pixel corresponding to the convolution kernel of each pixel, correct the filtering weight of each pixel in the convolution kernel of each pixel, obtain the corrected filtering weight of each pixel in the convolution kernel of each pixel, and then obtain the optimized activation distribution map of each task.

[0103] Specifically, in order to accurately filter the pixels in the activation distribution map of each task and enable accurate identification of the activation region corresponding to each task, this embodiment first determines the reference interference value of each pixel in the activation distribution map based on the overall scattering interference level. The larger the reference interference value, the greater the degree of scattering interference the corresponding pixel is subjected to, and the corresponding filtering weight in the convolution kernel should be reduced to effectively suppress scattering trails and reduce the impact of scattering interference. In addition, in order to avoid over-filtering of real edge pixels, which would blur the real boundaries and affect the accurate identification of the region corresponding to each task, when the reference interference value of the neighboring pixels in the convolution kernel of a certain pixel is inconsistent with that pixel, it indicates that the region corresponding to the convolution kernel of that pixel may be the boundary of the real functional area. In this case, the corresponding filtering weight in the convolution kernel should be retained to avoid excessive blurring of the edges during smoothing convolution. Therefore, this embodiment corrects the filtering weights of each pixel in the convolution kernel of each pixel based on the reference interference value corresponding to each pixel in the convolution kernel of each pixel, and obtains the corrected filtering weights of each pixel in the convolution kernel of each pixel. This allows for accurate filtering of the activation distribution map of each task, and real-time accurate acquisition of the optimized activation distribution map of each task. This effectively improves the spatial gradient recognition capability in the activation distribution map, enhances boundary sharpness, and reduces the degree of false activation diffusion, boundary blurring, and signal accumulation caused by multiple scattering wakes, thereby eliminating "edge smoothing" artifacts and further improving the accuracy of the classification model in extracting regional features (area, shape, distribution center). It should be noted that the size of the convolution kernel is set in this embodiment, but the implementer can set the size of the convolution kernel according to the actual situation, and it is not limited here. Each convolution kernel has known filtering weights corresponding to each pixel.

[0104] Preferably, in one feasible manner of this embodiment, the method for obtaining the reference interference value is as follows: for any pixel in the activation distribution map of any task, the overall scattering interference degree of the light source detector of the task closest to the pixel is taken as the reference interference value of the pixel.

[0105] At this point, the reference interference value for each pixel in the activation distribution map of each task is obtained.

[0106] Preferably, in one feasible embodiment of this invention, the method for obtaining the corrected filtering weights is as follows: for any pixel in the activation distribution map, the result of negatively correlating the reference interference values ​​of each pixel in the convolution kernel of that pixel is used as the first corrected weight of each pixel in the convolution kernel of that pixel; the smaller the first corrected weight, the greater the scattering interference received by the corresponding pixel, and the greater the degree of filtering required; in this embodiment, the difference between the constant 1 and the reference interference value of each pixel is used as the result of negatively correlating the reference interference value of each pixel.

[0107] To avoid over-filtering of real edge pixels, the absolute value of the difference between each pixel in the convolution kernel and the reference interference value of that pixel is obtained as the reference interference difference of each pixel in the convolution kernel. The larger the reference interference difference, the more likely the corresponding pixel is to be a real edge pixel, and the filtering weight of the corresponding pixel cannot be too small. Therefore, the reference interference difference of each pixel in the convolution kernel is multiplied by a preset analysis ratio value, and then added to a second preset constant. The result is used as the second correction weight of each pixel in the convolution kernel. In this embodiment, the preset analysis ratio value is set to 0.5 and the second preset constant value is set to 1. Implementers can set the size of the preset analysis ratio value and the second preset constant value according to the actual situation, which is not limited here.

[0108] In order to superimpose edge-sensitive adjustment on the basis of scattering interference correction, and finally form a dual adaptive convolution kernel with both scattering suppression and edge enhancement functions, this embodiment uses the product of the first correction weight, the second correction weight and the initial filtering weight of each pixel in the convolution kernel of the pixel as the correction filtering weight of each pixel in the convolution kernel of the pixel.

[0109] At this point, the corrected filtering weights corresponding to each pixel in the convolution kernel of each pixel in the activation distribution map of each task are obtained.

[0110] Step S4: Train the near-infrared brain functional imaging data classification model based on the optimized activation distribution map to achieve near-infrared brain functional imaging data classification.

[0111] Specifically, after obtaining the optimized activation distribution maps, they are classified using deep learning methods: First, optimized activation distribution maps for different tasks are obtained and manually labeled. The labeled optimized activation distribution maps are then divided into training and test sets in a 7:3 ratio, serving as input to the deep learning model. Then, based on a deep learning model structure using a Convolutional Neural Network (CNN), the optimized activation distribution maps for different tasks are automatically classified. In this embodiment, the cross-entropy loss function is used to measure the difference between the predicted class probability distribution and the true label. The Adam optimizer is selected in conjunction with an adaptive learning rate adjustment mechanism to improve the convergence speed and stability of the model in complex nonlinear feature spaces. Various hyperparameters are set, such as a learning rate of 0.001, a batch size of 32 optimized activation distribution maps, and 100 training epochs. Implementers can set various hyperparameters according to actual conditions; no limitations are imposed here. Convolutional Neural Networks (CNNs) are well-known technologies and will not be elaborated further.

[0112] The optimized activation distribution map from the training set is sequentially input into the deep learning model. Forward propagation is performed to calculate the predicted class, and the loss between the predicted value and the true label is calculated. The model parameters are corrected through the backpropagation algorithm. The process is iterated repeatedly until the loss function value converges or the early termination condition is met. Then, the model is tested on the test set to complete the training of the near-infrared brain functional imaging data classification (i.e., classification of different task categories) model, thereby enabling accurate classification of near-infrared brain functional imaging data.

[0113] In summary, this embodiment acquires brain blood oxygenation data using fNIRS to obtain an activation distribution map for each task. Based on the position and optical density changes of the light source detectors and the emission signal intensity data of the light source emitters under each task, the overall scattering interference level of each light source detector under each task is obtained. This allows for the determination of the reference interference value for each pixel in the activation distribution map. The filtering weights of each pixel in the convolution kernel are then corrected to obtain corrected filtering weights, thereby obtaining an optimized activation distribution map. This optimized map is then used to train a near-infrared brain functional imaging data classification model, achieving data classification for near-infrared brain functional imaging. This invention, by accurately obtaining corrected filtering weights, enables accurate filtering of the activation distribution map, effectively improving the accuracy of near-infrared brain functional imaging data classification.

[0114] Example 2:

[0115] This invention also proposes a deep learning-based data classification system for near-infrared brain functional imaging. Please refer to [link to relevant documentation]. Figure 5 The diagram illustrates a data classification system structure for near-infrared brain functional imaging based on deep learning, provided by an embodiment of the present invention. The system includes: an activation distribution map acquisition module 10, an overall scattering interference degree acquisition module 20, an optimized activation distribution map acquisition module 30, and a classification module 40.

[0116] The activation distribution map acquisition module 10 is used to collect brain blood oxygenation data through fNIRS technology and obtain the activation distribution map for each task. The brain blood oxygenation data includes the optical density change value of each light source detector at each time step and the transmission signal intensity data of each light source transmitter at each time step under each task.

[0117] The overall scattering interference level acquisition module 20 is used to acquire the overall scattering interference level of each light source detector under each task based on the position distribution and optical density change value of each light source detector under each task, the distance between each light source detector and the directly connected light source transmitter, the transmission signal strength data of the directly connected light source transmitter, the distance between each light source detector and its preset neighboring light source detectors, and the change of optical density change value of each light source detector.

[0118] The optimized activation distribution map acquisition module 30 is used to determine the reference interference value of each pixel in the activation distribution map based on the overall scattering interference level. Based on the reference interference value of each pixel corresponding to the convolution kernel of each pixel, the filtering weight of each pixel in the convolution kernel of each pixel is corrected to obtain the corrected filtering weight of each pixel in the convolution kernel of each pixel, and then the optimized activation distribution map of each task is obtained.

[0119] Classification module 40 is used to train a classification model for near-infrared brain functional imaging data based on an optimized activation distribution map, thereby achieving data classification for near-infrared brain functional imaging.

[0120] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the deep learning-based data classification system for near-infrared brain functional imaging and the deep learning-based data classification method for near-infrared brain functional imaging provided in the above embodiments belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0121] Example 3:

[0122] This invention also proposes a data classification device for near-infrared brain functional imaging based on deep learning. The device includes a memory and a processor. The memory stores executable program code, and the processor calls and executes the executable program code to perform a data classification method for near-infrared brain functional imaging based on deep learning provided in the embodiments of this application. Specifically, the device may be a chip, component, or module. The chip may include a connected processor and memory; the memory stores instructions, and when the processor calls and executes the instructions, the chip can perform the data classification method for near-infrared brain functional imaging based on deep learning provided in the above embodiments.

[0123] In addition, this embodiment also protects a computer device; please refer to [link to relevant documentation]. Figure 6 The computer device includes a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402. When the processor 402 executes the computer program 403, the computer device is able to execute any of the deep learning-based data classification methods for near-infrared brain functional imaging described above.

[0124] Example 4:

[0125] The present invention also provides a computer-readable storage medium storing computer program code, which, when executed on a computer, causes the computer to perform the aforementioned method steps to implement the data classification method for near-infrared brain functional imaging based on deep learning provided in the above embodiments.

[0126] Example 5:

[0127] The present invention also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to implement the data classification method for near-infrared brain functional imaging based on deep learning provided in the above embodiments.

[0128] In this embodiment, the device, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0129] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0130] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A data classification method for near-infrared brain functional imaging based on deep learning, characterized in that, The method includes the following steps: Brain blood oxygenation data was collected using fNIRS technology to obtain activation distribution maps for each task. The brain blood oxygenation data included the optical density change values ​​of each light source detector at each time step and the emission signal intensity data of each light source transmitter at each time step for each task. Based on the position distribution and optical density variation of each light source detector under each task, the distance between each light source detector and the directly connected light source transmitter, the transmitted signal strength data of the directly connected light source transmitter, the distance between each light source detector and its preset neighboring light source detectors, and the change in the optical density variation of each light source detector, the overall scattering interference level of each light source detector under each task is obtained. Based on the overall scattering interference level, the reference interference value of each pixel in the activation distribution map is determined. According to the reference interference value of each pixel corresponding to the convolution kernel of each pixel, the filtering weight of each pixel in the convolution kernel of each pixel is corrected. The corrected filtering weight of each pixel in the convolution kernel of each pixel is obtained, and then the optimized activation distribution map of each task is obtained. A near-infrared brain functional imaging data classification model was trained based on an optimized activation distribution map to achieve near-infrared brain functional imaging data classification.

2. The data classification method for near-infrared brain functional imaging based on deep learning as described in claim 1, characterized in that, The method for obtaining the overall scattering interference level is as follows: Based on the positional distribution of each light source detector under each task and the change in optical density at each time, the first scattering interference level of each light source detector under each task is obtained. Based on the distance between each light source detector and the directly connected light source transmitter under each task, the transmitted signal strength data of the light source transmitter directly connected to each light source detector, and the distance between each light source detector and its preset neighboring light source detectors, the second scattering interference level of each light source detector under each task is obtained. The activation level of each task is obtained by analyzing the changes in the optical density of each light source detector under each task. Based on the activation level of each task and the first and second scattering interference levels of each light source detector under each task, the overall scattering interference level of each light source detector under each task is obtained; among them, the activation level is negatively correlated with the overall scattering interference level, while the first and second scattering interference levels are both positively correlated with the overall scattering interference level.

3. The data classification method for near-infrared brain functional imaging based on deep learning as described in claim 2, characterized in that, The method for obtaining the first level of scattering interference is as follows: For any task and any light source detector under that task, fit the optical density change value of the light source detector at each moment within the task phase to a curve, and obtain the derivative of each optical density change value on the curve as the change analysis value. The moment corresponding to the largest change analysis value is taken as the activation moment of the light source detector, and the optical density change value at the activation moment of the light source detector is taken as the target optical density change value of the light source detector. The light source detector that is closest to the center of the brain activation region corresponding to the task is used as the activation center detector for the task. The degree of delayed response of the light source detector is obtained based on the difference in activation time and distance between the light source detector and the activation center detector, as well as the change value of the target optical density of the light source detector. The time corresponding to the maximum optical density change value of the activated center detector is taken as the target time. Based on the difference in optical density change value and distance between the activated center detector and the light source detector at the target time, the attenuation interference degree of the light source detector is obtained. The product of the delay response of the light source detector and the attenuation interference level is taken as the first scattering interference level of the light source detector.

4. The data classification method for near-infrared brain functional imaging based on deep learning as described in claim 3, characterized in that, The method for obtaining the degree of delayed response is as follows: The result of normalizing the absolute value of the difference between the activation time of the light source detector and the activation center detector is used as the delay reference value of the light source detector. The normalized spatial distance between the light source detector and the activation center detector is used as the reference distance for the light source detector. The delay response degree of the light source detector is obtained based on the delay reference value, reference distance, and target optical density change value of the light source detector; among them, the delay reference value and reference distance are negatively correlated with the delay response degree, while the target optical density change value is positively correlated with the delay response degree.

5. A data classification method for near-infrared brain functional imaging based on deep learning as described in claim 4, characterized in that, The method for obtaining the degree of interference attenuation is as follows: The result of normalizing the difference between the optical density change value of the activated center detector and the light source detector at the target time is used as the first characteristic value of the light source detector. The ratio of the first characteristic value of the light source detector to the reference distance is used as the degree of optical density attenuation of the light source detector; The difference between the optical density attenuation degree and the first preset constant is taken as the attenuation interference degree of the light source detector.

6. The data classification method for near-infrared brain functional imaging based on deep learning as described in claim 2, characterized in that, The method for obtaining the second level of scattering interference is as follows: For any task and any light source detector under that task, the light source emitter directly connected to the light source detector shall be used as the reference emitter of the light source detector. The spatial distance between the light source detector and each reference transmitter is taken as the first distance for each reference transmitter; The average of the product of the first distance and the transmitted signal strength data of each reference transmitter is taken as the emission influence of the light source detector. The nearest light source detector in each direction to which the light source detector is connected is taken as the preset neighbor light source detector of the light source detector; The result of normalizing the spatial distance between the light source detector and each preset neighboring light source detector is used as the reference weight for each preset neighboring light source detector. The average of the product of the reference weight and the emission influence of each preset neighborhood light source detector is taken as the detection influence of that light source detector. The product of the emission influence and the detection influence of the light source detector is taken as the second scattering interference level of the light source detector.

7. A data classification method for near-infrared brain functional imaging based on deep learning as described in claim 3, characterized in that, The method for obtaining the activation level is as follows: For any task and any light source detector under that task, the duration of the increasing interval to which the largest optical density change value in the curve corresponding to the light source detector belongs is taken as the first duration of the light source detector. The ratio of the maximum optical density change value of the light source detector to the first time duration is used as the activation analysis value of the light source detector; The normalized result of the mean of the activation analysis values ​​of all light source detectors in the brain activation region corresponding to the task is taken as the activation level of the task.

8. A data classification method for near-infrared brain functional imaging based on deep learning as described in claim 1, characterized in that, The method for obtaining the reference interference value is as follows: For any pixel in the activation distribution map of any task, the overall scattering interference level of the light source detector of that task that is closest to that pixel is used as the reference interference value of that pixel.

9. A data classification method for near-infrared brain functional imaging based on deep learning as described in claim 1, characterized in that, The method for obtaining the corrected filter weights is as follows: For any pixel in the activation distribution map, the result of negatively correlating the reference interference values ​​of each pixel in the convolution kernel of that pixel is used as the first correction weight of each pixel in the convolution kernel of that pixel. The difference between each pixel in the convolution kernel of that pixel and the reference interference value of that pixel is used as the reference interference difference of each pixel in the convolution kernel of that pixel. The reference interference difference of each pixel in the convolution kernel of that pixel is multiplied by the preset analysis ratio value, and then added to the second preset constant. The result is used as the second correction weight of each pixel in the convolution kernel of that pixel. The product of the first corrected weight, the second corrected weight, and the initial filter weight of each pixel in the convolution kernel of that pixel is used as the corrected filter weight of each pixel in the convolution kernel of that pixel.

10. A data classification method for near-infrared brain functional imaging based on deep learning as described in claim 1, characterized in that, The method for obtaining the optical density change value is as follows: For any given light source detector and any given time, the difference between the optical density of that light source detector at that time and the baseline optical density is taken as the change in optical density of that light source detector at that time.

Images