CN115299946A - Self-adaptive input signal channel screening circuit introducing detection result feedback - Google Patents
Self-adaptive input signal channel screening circuit introducing detection result feedback Download PDFInfo
- Publication number
- CN115299946A CN115299946A CN202211028703.6A CN202211028703A CN115299946A CN 115299946 A CN115299946 A CN 115299946A CN 202211028703 A CN202211028703 A CN 202211028703A CN 115299946 A CN115299946 A CN 115299946A
- Authority
- CN
- China
- Prior art keywords
- channel
- unit
- signal
- screening
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 99
- 238000012216 screening Methods 0.000 title claims abstract description 81
- 238000004364 calculation method Methods 0.000 claims abstract description 51
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000012163 sequencing technique Methods 0.000 claims abstract description 3
- 230000003044 adaptive effect Effects 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 10
- 239000000872 buffer Substances 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 230000001174 ascending effect Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 6
- 230000001037 epileptic effect Effects 0.000 description 13
- 206010015037 epilepsy Diseases 0.000 description 10
- 238000001914 filtration Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000011895 specific detection Methods 0.000 description 2
- 206010010904 Convulsion Diseases 0.000 description 1
- 208000019914 Mental Fatigue Diseases 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001314 paroxysmal effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
- A61B5/165—Evaluating the state of mind, e.g. depression, anxiety
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
- A61B5/168—Evaluating attention deficit, hyperactivity
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
- A61B5/372—Analysis of electroencephalograms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
- A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
- A61B5/4094—Diagnosing or monitoring seizure diseases, e.g. epilepsy
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Veterinary Medicine (AREA)
- Medical Informatics (AREA)
- Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Psychiatry (AREA)
- Animal Behavior & Ethology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Psychology (AREA)
- Artificial Intelligence (AREA)
- Physiology (AREA)
- Neurology (AREA)
- Developmental Disabilities (AREA)
- Educational Technology (AREA)
- Neurosurgery (AREA)
- Child & Adolescent Psychology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Hospice & Palliative Care (AREA)
- Social Psychology (AREA)
- Fuzzy Systems (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a self-adaptive input signal channel screening circuit introducing detection result feedback, and belongs to the field of signal processing. The invention comprises a feature extraction unit, a similarity calculation unit and a sequencing unit; the channel screening circuit is used for carrying out channel screening on input signals and sending channel screening results to the signal detection unit, so that the signal detection unit only executes two classification detection tasks on the signals of the screened channels, meanwhile, the detection results of the signal detection unit are introduced, and the channel screening results are adaptively adjusted based on the detection results, so that a better screening effect is achieved. The invention feeds back the detection result to the channel screening part, so that the information provided by the detection result can be utilized during channel screening, the extra calculation amount is not required to be increased, the channel screening can be more depended on, the channel screening is more accurate, and the influence on the detection accuracy is minimized. Therefore, the calculation amount of the classification detection task executed by the signal detection unit is reduced, and the power consumption is reduced.
Description
Technical Field
The invention belongs to the field of signal processing, and particularly relates to a self-adaptive input signal channel screening circuit introducing detection result feedback.
Background
Signal detection is a common signal processing task, for example, epilepsy detection based on electroencephalogram signals, and electrocardiogram classification based on multi-lead electrocardiosignals all involve multiple input signal channels, in the past, all channels were directly used for detection, but not all channels contain effective information of a specific detection task, all channels are directly used to increase complexity, but detection effect cannot necessarily be improved, and the following describes epilepsy detection as an example.
Because the paroxysmal and unpredictable nature of epilepsy, the frequency of epileptic seizures varies from once a few years to several times a day, so in order not to miss epilepsy which may happen at any time, the detection must be monitored in real time for a long time, which requires hardware for realizing the detection function not only to have high detection accuracy but also to have low power consumption characteristic which can be detected for a long time. The number of channels for acquiring electroencephalogram signals can reach as many as 36, and if all channel data are directly used for epilepsy detection without distinction, two problems are caused. First, the EEG is a centralized reflection of the conduction of a large number of neuronal cells to the cortex during their playback of electrical activity, and different acquisition channels correspond to electrodes placed at different locations in the cortex. In some states, there may be redundancy in the physiological information reflected by the adjacent electrodes, or the information content of the electrodes in some areas with weak neuron activity is weak. If the channels are directly used for epilepsy detection without distinction, the detection effect may be reduced, so that the screening of effective channels is very important. Secondly, taking all channels to perform epilepsy detection inevitably increases the calculation amount of detection, which is contrary to the requirement of low power consumption, so that eliminating redundant channels as much as possible on the premise of ensuring the detection accuracy is crucial to realizing long-time monitoring with low power consumption.
As can be seen from the above, effective and intelligent channel screening is of great significance for signal detection.
Disclosure of Invention
The invention aims to: aiming at the existing problems, the self-adaptive input signal channel screening circuit introducing the detection result feedback is provided, so that partial channels are screened on the premise of ensuring the subsequent detection accuracy, the detection calculation amount is reduced, and the power consumption is reduced.
The technical scheme adopted by the invention is as follows:
an adaptive input signal channel screening circuit incorporating feedback of detection results, comprising: the device comprises a feature extraction unit, a similarity calculation unit and a sorting unit; the self-adaptive input signal channel screening circuit is used for screening channels of input signals and sending channel screening results to the signal detection unit so that the signal detection unit can only execute two classification detection tasks on the signals of the screened channels, meanwhile, the self-adaptive input signal channel screening circuit introduces the detection results of the signal detection unit and self-adaptively adjusts the channel screening results based on the detection results so as to achieve a better screening effect; the self-adaptive input signal channel screening circuit provided by the invention can be matched with various detection units, so that partial channels are screened on the premise of ensuring the subsequent detection accuracy, the calculated amount of the classification detection tasks executed by the signal detection units is reduced, and the power consumption is reduced.
The input signal of the feature extractor unit comprises a plurality of channels, the signal of one channel is input at a time, and the input signal is a positive sample signal or a negative sample signal (namely, the positive and negative of the sample signal are used for representing two categories of the binary detection task of the signal detection unit);
the characteristic extractor unit comprises a plurality of calculation paths, a switch unit and a plurality of selectors, wherein the switch unit is respectively connected with each calculation path, and the output end of each calculation path is connected with the plurality of selectors; when the number of the input positive and negative sample signals meets the requirement, triggering a similarity calculation unit to calculate the screening weight of the channel;
the similarity calculation unit reads all the positive sample time-frequency domain feature vectors and all the negative sample time-frequency domain feature vectors from the storage unit, and obtains a positive sample feature template and a negative sample feature template based on the average value of all the positive sample time-frequency domain feature vectors and all the negative sample time-frequency domain feature vectors; representing the screening weight of the current channel of the input signal based on the distance between the positive sample characteristic template and the negative sample characteristic template, caching, and sending the screening weight of all channels of the input signal to a sequencing unit after obtaining the screening weights of all channels of the input signal;
the sorting unit is used for sorting the screening weight values of all the channels in an ascending order, taking the channel identifications corresponding to the first K screening values as channel screening results and sending the channel identifications to the signal detection unit;
the signal detection unit reads signal data of a channel corresponding to a signal to be detected based on a current channel screening result, executes a binary detection task, and feeds back a detection result to the feature extractor unit, and the switch unit of the feature extractor unit controls whether to input the signal of each channel of the current signal to be detected to each path of calculation path as an input signal in sequence based on the detection result: if the type of the detection result is matched with the type requirement to be input into the calculation access, the switch unit is switched on, and if not, the switch unit is switched off; and the multiplexer of the feature extractor unit determines a storage unit of the time-frequency domain feature vector of the input signal based on the current detection result.
Further, when the signal detection unit executes the binary detection task based on the neural network, the channel screening result includes a channel number K matched with the input channel number of the input layer of the neural network.
Further, the feature extractor unit includes four paths of computation paths, which are respectively used for computing four time-frequency domain features of zero crossing, area, attenuation and line length of the input signal.
Further, the four computation paths of the feature extractor unit are respectively:
the zero-crossing point calculation path sequentially comprises two comparators, an exclusive-or gate, a multiplier and an accumulator, and is used for calculating and accumulating each sampling point of a current input signal point by point to obtain the zero-crossing point characteristic of the current sample signal (positive sample signal or negative sample signal), one comparator is used for comparing the input data of the current moment with 0, the other comparator is used for comparing the input data of the previous moment with 0, if the comparison result of the comparator is greater than 0, the output is 1, if the comparison result of the comparator is less than 0, the output result is 0, then the results output by the two comparators are subjected to exclusive-or, if the exclusive-or result is 1, the data of the current moment and the data of the previous moment are in an exclusive sign, namely, a zero-crossing point exists, and then the exclusive-or result is multiplied by a preset normalization coefficient alpha to obtain the zero-crossing point characteristic 1 Accumulating, and obtaining the zero-crossing point characteristic of the current sample signal after calculating all sampling points in one sample signal (positive sample signal or negative sample signal);
an area calculation path sequentially including an adder, a multiplier and an accumulator, adding the input data of the current time and the input data of the previous time by the adder, and multiplying the addition result by the normalization coefficient alpha by the multiplier 2 Accumulating each sampling point through an accumulator, and obtaining the area characteristic of the sample signal after all the sampling points in the sample signal are calculated;
the attenuation calculation path sequentially comprises a comparator, a multiplier and an accumulator, the comparator compares the input data of the current moment with the input data of the previous moment, and the comparison result is multiplied by a normalization coefficient alpha 3 Then accumulating, and obtaining the attenuation of the sample signal after calculating all sampling points in the sample signalSubtracting the characteristic;
the line length calculating path sequentially comprises a subtracter, an absolute value calculating unit, a multiplier and an accumulator, the subtracter subtracts input data of the current moment from input data of the previous moment, the absolute value of the subtraction result is calculated, and the subtraction result is multiplied by a normalization coefficient alpha 4 And then accumulating, and obtaining the line length characteristic of the sample signal after calculating all sampling points in the sample signal.
Furthermore, the similarity calculation unit calculates the screening weight of each channel according to the cosine distance between the positive sample feature template and the negative sample feature template.
Furthermore, the similarity calculation unit comprises a plurality of similarity sub-calculation paths, and is used for calculating the cosine distance between each dimension of the feature of the positive and negative sample feature templates of the current channel, and accumulating the cosine distance calculation result of each similarity sub-calculation path to obtain the screening weight of the current channel.
Furthermore, each similarity sub-calculation path sequentially comprises a subtracter and a multiplier; and subtracting each dimension characteristic of the positive and negative sample characteristic templates by a subtracter, and squaring by a multiplier, namely, calculating the cosine distances of different channels and multiplexing the same set of cosine distance calculation logic in a time-sharing manner.
Further, the sorting unit sorts the screening weights of the channels in a bubble sorting mode.
Further, the sorting unit includes two filtering weight caches (when cosine distances are used to calculate filtering weights of each channel, the two caches are cosine distance caches), a comparator, three multiplexers, and a demultiplexer, where a first filtering weight cache is used to store cosine distances of all channels, a second filtering weight cache is used to store channel filtering results, a second-out-of-two multiplexer selects adjacent elements from the first filtering weight cache to enter the comparator for comparison, the comparison results respectively enter two-out-of-one selectors, each of the two-out-of-two selectors selects a largest element to be sent to the second-out-of-multiple demultiplexer, and an output end of the second-out-of-multiple demultiplexer is connected to the second filtering weight cache.
When the sorting unit performs bubble sorting, adjacent elements are compared, if the first is larger than the second, the two elements are exchanged, the same work is performed on each pair of adjacent elements, and the first pair is started to the last pair at the end; the last element should be the largest number. The above steps are repeated for all elements except the last one. The above steps are repeated for fewer and fewer elements each time until no pair of numbers need to be compared.
The technical scheme provided by the invention at least has the following beneficial effects:
(1) The influence of channel screening on the detection accuracy is minimized: the detection result is fed back to the channel screening part, so that the information provided by the detection result can be utilized during channel screening, the calculated amount is not additionally increased, the channel screening can be more depended on, the channel screening is more accurate, and the influence on the detection accuracy is minimized.
(2) The channel screening operand is low, and the power consumption of the detection system is reduced: four characteristics which are very effective to channel screening are analyzed and selected, and the calculated amount for extracting the characteristics is very small; in addition, by exploring the most effective channel, the invention can achieve the same level of detection accuracy with less channel number, thus greatly reducing the calculation amount of the subsequent classifier and having more remarkable power consumption benefit for the whole system; the invention also optimizes the power consumption of the hardware realization of channel screening. This results in a relatively low power consumption of the channel screening hardware proposed by the present invention.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a diagram illustrating a hardware architecture of an adaptive channel screening circuit incorporating result feedback according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a feature extraction unit according to the present invention;
FIG. 3 is a diagram illustrating a similarity calculation unit according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a sorting unit according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The adaptive input signal channel screening circuit with the detection result feedback introduced provided by the embodiment of the invention comprises a feature extraction unit, a similarity calculation unit and a sorting unit, wherein the structural schematic diagram of the channel screening circuit is shown in fig. 1, the feature extraction unit introduces the detection result for calculating four feature values of an input signal (such as an electroencephalogram signal), then a cosine distance calculation unit calculates cosine distances of all channels by using the features obtained by the feature extraction unit, finally the cosine distance sorting unit is used for sorting the cosine distances of all channels, and after sorting is finished, the front part of channels are taken as final channel screening results.
The adaptive input signal channel screening circuit introducing the feedback of the detection result provided by the embodiment of the invention is further described in detail below by taking multichannel screening for epilepsy detection as an example.
In the invention, the channel screening circuit needs to introduce the classification detection result of the signal so as to further generate the corresponding channel screening result in a self-adaptive manner according to the detection result.
That is, the channel screening result output by the channel screening circuit is sent to the signal detection unit (to detect whether the current user is epileptic, of course, the signal detection unit related to the present invention may also be used for other two-classification detection tasks (e.g., whether depression is present, whether mental fatigue is present, whether a target object is present, etc.), and the setting may be adjusted based on the actual application requirements.
For the collected signal to be detected, because the collected signal includes a plurality of channels, if all channel data are input to the signal detection unit for detection processing, higher computational complexity is caused, so when facing to a binary detection processing task related to signal class data, an optimal channel can be determined by the channel screening circuit of the embodiment of the invention, and then the signal detection unit only needs to perform detection processing on the data of the optimal channel, so as to reduce the processing complexity of the detection task.
For the two-classification detection task, the classification detection structure is set to be a positive type and a negative type, in the initial stage, based on the sample signal (positive sample signal or negative sample signal) of the known class label, the initial channel screening result is obtained through the channel screening circuit of the invention, so that the signal detection unit reads the data of the corresponding channel based on the channel screening structure to execute the two-classification detection processing of the signal to be detected, and then based on the detection result, the signal to be detected is set to be the positive or negative sample signal, so as to realize the updating of the input sample, and further the more applicable channel screening result is obtained in a self-adaptive manner.
In this embodiment, the feature extractor unit is configured to calculate four feature values of an electroencephalogram signal, and a structure diagram of the feature extractor unit is shown in fig. 2, where fig. 2 exemplifies channel screening for M channels. Each feature is computed over n samples and normalized to make the feature more representative. The four features are respectively Zero-crossing point (Zero-crossing), area (Area), attenuation (Decay) and Line-length (Line-length), and the four features are respectively calculated by four calculation paths. The zero-crossing point calculation path comprises two comparators, an exclusive-or gate, a multiplier and an accumulator, the current input data and the input data at the previous moment are respectively compared with each other at a value of 0, if the comparison result is greater than 0, the output is 1, if the comparison result is less than 0, the output result is 0, then the results output by the two comparators are subjected to exclusive-or, and if the exclusive-or result is 1, the data at the current moment and the data at the previous moment are of an exclusive sign, namely, a zero-crossing point exists. And then, multiplying the XOR result by a normalization coefficient, accumulating, and obtaining the zero-crossing point characteristic of the sample after all points in the sample are calculated. Wherein the area calculation path includes oneThe system comprises an adder, a multiplier and an accumulator, wherein the input of the current moment is added with the input of the previous moment, the addition result is multiplied by a normalization coefficient, then accumulation is carried out, and the area characteristic of a sample is obtained after all points in the sample are calculated. The attenuation calculation path comprises a comparator, a multiplier and an accumulator, the current input data and the data of the previous moment are compared, the comparison result is multiplied by a normalization coefficient and then accumulated, and the attenuation characteristic of a sample is obtained after all points in the sample are calculated. The line length calculating path comprises a subtracter, an absolute value calculating unit, a multiplier and an accumulator, the current input data and the input data at the previous moment are subtracted, the absolute value of the subtraction result is obtained, the subtraction result is multiplied by a normalization coefficient and then accumulated, and the line length characteristic of a sample is obtained after all points in the sample are calculated. In addition, the feature extraction unit further comprises a switch unit and a multiplexer, wherein the switch unit controls to enter a specified number of non-epileptic samples first and then enter a specified number of epileptic samples (or enter a specified number of epileptic samples first and then enter a specified number of non-epileptic samples) so as to average the features of a plurality of samples of the same type. And the multiplexer registers the characteristic calculation result according to epilepsy and non-epilepsy respectively according to the current sample type. α in FIG. 2 1 、α 2 、α 3 And alpha 4 The values are all normalized coefficients, the preset values depend on the actual application scene.
The similarity calculation unit is configured to calculate the filtering weights of the channels, and has a structure shown in fig. 3, including four subtractors, four adders, and an accumulator. The screening weight of each channel is characterized by cosine distances between four characteristics of epileptic samples and non-epileptic samples, after the cosine distance calculation unit receives the characteristics of the epileptic samples and the non-epileptic samples extracted by the characteristic extractor unit, the cosine distances between the four characteristics of the epileptic samples and the four characteristics of the non-epileptic samples are respectively calculated according to the channels, namely the four characteristics of the epileptic samples and the non-epileptic samples are subtracted by a subtracter respectively, then squared by a multiplier, and then the squared results of the four characteristics are mutually accumulated to obtain the cosine distance of one channel, which is used as the screening weight of the channel, and the cosine distances of different channels are calculated and time-division multiplexed by the same set of cosine distance calculation logic.
The sorting unit is configured to sort cosine distances of the channels, and a result of the sorting is shown in fig. 4, where fig. 4 is an example of screening K channels from M channels (K < M), and includes two cosine distance buffers, a comparator, and four multiplexers. Sorting adopts a bubble sorting mode to compare adjacent elements. If the first is larger than the second, then the two are swapped. The same work is done for each pair of adjacent elements, from the first pair to the last pair at the end. At this point, the last element should be the largest number. The above steps are repeated for all elements except the last one. The above steps are repeated for fewer and fewer elements each time until no pair of numbers need to be compared. And taking the front K channels as channel screening results after the sorting is finished.
In the invention, the signal detection unit is used for carrying out two-class detection on the signals, the detection result comprises a positive class and a negative class, and the detection result can be fed back to the feature extraction unit. The invention can provide the best channel selection for the classifier and can be matched with various specific detection units, such as a common convolutional neural network. When the convolutional neural network is used as a classifier, only the optimal channels with the number equal to that of the input channels of the neural network need to be screened out, the convolutional neural network can extract the characteristics of the input sample through a plurality of convolutional layers, select and recombine the characteristics through a full connection layer, and complete classification.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
What has been described above are merely some embodiments of the present invention. It will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention.
Claims (9)
1. An adaptive input signal path screening circuit incorporating feedback of test results, comprising: the device comprises a feature extraction unit, a similarity calculation unit and a sorting unit;
the input signal of the feature extractor unit comprises a plurality of channels, the signal of one channel is input each time, the input signal is a positive sample signal or a negative sample signal, and the positive and negative of the sample signal are used for representing two categories of two classification detection tasks of the signal detection unit;
the characteristic extractor unit comprises a plurality of calculation paths, a switch unit and a plurality of selectors, wherein the switch unit is respectively connected with each calculation path, and the output end of each calculation path is connected with the plurality of selectors; when the number of the input positive and negative sample signals meets the requirement, triggering a similarity calculation unit to calculate the screening weight of the channel;
the similarity calculation unit reads all the positive sample time-frequency domain feature vectors and all the negative sample time-frequency domain feature vectors from the storage unit, and obtains a positive sample feature template and a negative sample feature template based on the average value of all the positive sample time-frequency domain feature vectors and all the negative sample time-frequency domain feature vectors; representing the screening weight of the current channel of the input signal based on the distance between the positive sample characteristic template and the negative sample characteristic template, caching, and sending the screening weight of all channels of the input signal to a sequencing unit after obtaining the screening weights of all channels of the input signal;
the sorting unit is used for sorting the screening weight values of all the channels in an ascending order, taking the channel identifications corresponding to the first K screening values as channel screening results and sending the channel identifications to the signal detection unit;
the signal detection unit reads signal data of a channel corresponding to a signal to be detected based on a current channel screening result, executes a binary detection task, and feeds back a detection result to the feature extractor unit, and the switch unit of the feature extractor unit controls whether to input the signal of each channel of the current signal to be detected to each path of calculation path as an input signal in sequence based on the detection result: if the type of the detection result is matched with the type requirement to be input into the calculation access, the switch unit is switched on, and if not, the switch unit is switched off; and the multiplexer of the feature extractor unit determines a storage unit of the time-frequency domain feature vector of the input signal based on the current detection result.
2. The adaptive input signal channel screening circuit incorporating test result feedback of claim 1, wherein the channel screening result includes a number of channels K that matches a number of input channels of an input layer of the neural network when the signal test unit performs the classification test task based on the neural network.
3. The adaptive input signal channel screening circuit with detection result feedback introduced as claimed in claim 1, wherein said feature extractor unit comprises four computation paths for computing four time-frequency domain features of zero crossing, area, attenuation and line length of the input signal, respectively.
4. The adaptive input signal channel filter circuit incorporating the feedback of detection results of claim 3, wherein the four computational paths of the feature extractor unit are respectively:
the zero-crossing point calculation path sequentially comprises two comparators, an exclusive-or gate, a multiplier and an accumulator, wherein one comparator is used for comparing the input data of the current moment with 0, and the other comparator is used for comparing the input data of the previous moment with 0Comparing the magnitude, if the comparison result of the comparator is greater than 0, the output is 1, if the comparison result of the comparator is less than 0, the output result is 0, the results output by the two comparators are subjected to XOR, and then the XOR result is multiplied by a preset normalization coefficient alpha 1 Then accumulating to obtain the zero crossing point characteristic of the current sample signal;
an area calculation path sequentially including an adder, a multiplier and an accumulator, adding the input data at the current time and the input data at the previous time, and multiplying the addition result by a normalization coefficient alpha 2 Accumulating to obtain the area characteristic of the current sample signal;
attenuation calculation path, which includes a comparator, a multiplier and an accumulator in turn, compares the input data at the current time with the input data at the previous time, and multiplies the comparison result by the normalization coefficient alpha 3 Then accumulating to obtain the attenuation characteristic of the current sample signal;
the line length calculating path comprises a subtracter, an absolute value calculating unit, a multiplier and an accumulator in sequence, the input data at the current moment is subtracted from the input data at the previous moment, the absolute value of the subtraction result is calculated, and the subtraction result is multiplied by a normalization coefficient alpha 4 And then accumulating to obtain the line length characteristic of the current sample signal.
5. The adaptive input signal channel screening circuit with introduction of test result feedback of claim 1, wherein the similarity calculation unit calculates the screening weight of each channel by a cosine distance between a positive sample feature template and a negative sample feature template.
6. The adaptive input signal channel screening circuit with introduction of detection result feedback as recited in any one of claims 1 to 5, wherein the similarity calculation unit comprises a plurality of similarity sub-calculation paths for calculating cosine distances between features of each dimension of the positive and negative sample feature templates of the current channel, and accumulates the cosine distance calculation results of each similarity sub-calculation path to obtain the screening weight of the current channel.
7. The adaptive input signal path filter circuit incorporating test result feedback of claim 6, wherein each of said similarity sub-computation paths comprises, in order, a subtractor and a multiplier; and subtracting each dimension of the features of the positive and negative sample feature templates by a subtracter, and squaring by a multiplier.
8. The adaptive input signal channel filter circuit with introduced detection result feedback of claim 1, wherein the sorting unit sorts the filter weights of the channels in a bubble sorting manner.
9. The adaptive input signal channel filter circuit with the introduction of the feedback of the detection result as claimed in claim 8, wherein the sorting unit comprises two filter weight buffers, a comparator, three multiplexers and a demultiplexer, wherein the first filter weight buffer is used to store the cosine distances of all channels, the second filter weight buffer is used to store the channel filter result, the two-out-of-two multiplexer selects the adjacent elements from the first filter weight buffer to enter the comparator for comparison, the comparison result enters the two one-out-of-two selectors respectively, each of the two one-out-of-two selectors selects one of the largest elements to be sent to the two-out-of-two demultiplexer, and the output end of the two-out-of-two demultiplexer is connected to the second filter weight buffer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211028703.6A CN115299946B (en) | 2022-08-25 | 2022-08-25 | Self-adaptive input signal channel screening circuit introducing detection result feedback |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211028703.6A CN115299946B (en) | 2022-08-25 | 2022-08-25 | Self-adaptive input signal channel screening circuit introducing detection result feedback |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115299946A true CN115299946A (en) | 2022-11-08 |
| CN115299946B CN115299946B (en) | 2024-05-28 |
Family
ID=83864156
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211028703.6A Active CN115299946B (en) | 2022-08-25 | 2022-08-25 | Self-adaptive input signal channel screening circuit introducing detection result feedback |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115299946B (en) |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040230105A1 (en) * | 2003-05-15 | 2004-11-18 | Widemed Ltd. | Adaptive prediction of changes of physiological/pathological states using processing of biomedical signals |
| US20050197590A1 (en) * | 1997-01-06 | 2005-09-08 | Ivan Osorio | System for the prediction, rapid detection, warning, prevention, or control of changes in activity states in the brain of a subject |
| US20120101401A1 (en) * | 2009-04-07 | 2012-04-26 | National University Of Ireland | Method for the real-time identification of seizures in an electroencephalogram (eeg) signal |
| US20140142958A1 (en) * | 2012-10-15 | 2014-05-22 | Digimarc Corporation | Multi-mode audio recognition and auxiliary data encoding and decoding |
| CN104080397A (en) * | 2012-01-19 | 2014-10-01 | 脑科技医疗系统股份有限公司 | Diagnostic system for detection of fluid changes |
| US20160228705A1 (en) * | 2015-02-10 | 2016-08-11 | Neuropace, Inc. | Seizure onset classification and stimulation parameter selection |
| CN109620148A (en) * | 2018-11-29 | 2019-04-16 | 西安交通大学 | A kind of epilepsy detection integrated circuit based on sparse extreme learning machine algorithm |
| CN110811609A (en) * | 2019-10-28 | 2020-02-21 | 杭州电子科技大学 | Intelligent epileptic spike detection method based on fusion of adaptive template matching and machine learning algorithm |
| CN111034041A (en) * | 2017-05-24 | 2020-04-17 | 安乐泰克有限公司 | Apparatus and method for controlling resonator |
| US20200388397A1 (en) * | 2019-06-07 | 2020-12-10 | Cornell University | Energy-efficient on-chip classifier for detecting physiological conditions |
| CN112949533A (en) * | 2021-03-15 | 2021-06-11 | 成都信息工程大学 | Motor imagery electroencephalogram identification method based on relative wavelet packet entropy brain network and improved version lasso |
| CN113936791A (en) * | 2021-10-18 | 2022-01-14 | 贵州师范大学 | Classification model training method combining active learning and transfer learning |
| CN114532993A (en) * | 2022-03-23 | 2022-05-27 | 电子科技大学 | Automatic detection method for electroencephalogram high-frequency oscillation signals of epileptic |
| CN114587381A (en) * | 2022-03-14 | 2022-06-07 | 杭州电子科技大学 | Spike detection method based on multi-channel electroencephalogram intelligent screening and weighted sample generation |
-
2022
- 2022-08-25 CN CN202211028703.6A patent/CN115299946B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050197590A1 (en) * | 1997-01-06 | 2005-09-08 | Ivan Osorio | System for the prediction, rapid detection, warning, prevention, or control of changes in activity states in the brain of a subject |
| US20040230105A1 (en) * | 2003-05-15 | 2004-11-18 | Widemed Ltd. | Adaptive prediction of changes of physiological/pathological states using processing of biomedical signals |
| US20120101401A1 (en) * | 2009-04-07 | 2012-04-26 | National University Of Ireland | Method for the real-time identification of seizures in an electroencephalogram (eeg) signal |
| CN104080397A (en) * | 2012-01-19 | 2014-10-01 | 脑科技医疗系统股份有限公司 | Diagnostic system for detection of fluid changes |
| US20140142958A1 (en) * | 2012-10-15 | 2014-05-22 | Digimarc Corporation | Multi-mode audio recognition and auxiliary data encoding and decoding |
| US20160228705A1 (en) * | 2015-02-10 | 2016-08-11 | Neuropace, Inc. | Seizure onset classification and stimulation parameter selection |
| CN111034041A (en) * | 2017-05-24 | 2020-04-17 | 安乐泰克有限公司 | Apparatus and method for controlling resonator |
| CN109620148A (en) * | 2018-11-29 | 2019-04-16 | 西安交通大学 | A kind of epilepsy detection integrated circuit based on sparse extreme learning machine algorithm |
| US20200388397A1 (en) * | 2019-06-07 | 2020-12-10 | Cornell University | Energy-efficient on-chip classifier for detecting physiological conditions |
| CN110811609A (en) * | 2019-10-28 | 2020-02-21 | 杭州电子科技大学 | Intelligent epileptic spike detection method based on fusion of adaptive template matching and machine learning algorithm |
| CN112949533A (en) * | 2021-03-15 | 2021-06-11 | 成都信息工程大学 | Motor imagery electroencephalogram identification method based on relative wavelet packet entropy brain network and improved version lasso |
| CN113936791A (en) * | 2021-10-18 | 2022-01-14 | 贵州师范大学 | Classification model training method combining active learning and transfer learning |
| CN114587381A (en) * | 2022-03-14 | 2022-06-07 | 杭州电子科技大学 | Spike detection method based on multi-channel electroencephalogram intelligent screening and weighted sample generation |
| CN114532993A (en) * | 2022-03-23 | 2022-05-27 | 电子科技大学 | Automatic detection method for electroencephalogram high-frequency oscillation signals of epileptic |
Non-Patent Citations (4)
| Title |
|---|
| HAMEED, A: "Classification of Epileptic Seizures using Artificial Neural Network with Adaptive Momentum", 《2020 INTERNATIONAL CONFERENCE ON DATA ANALYTICS FOR BUSINESS AND INDUSTRY》, 31 December 2020 (2020-12-31), pages 1 - 5 * |
| 周勇: "基于神经网络的低功耗智能癫痫检测硬件设计", 《中国优秀硕士学位论文全文数据库》, 21 March 2022 (2022-03-21), pages 1 - 50 * |
| 王明达: "自适应最近邻算法及在门禁人脸识别中的应用", 《中国优秀硕士学位论文全文数据库》, 31 December 2018 (2018-12-31), pages 1 - 62 * |
| 黄璐弘: "基于多特征融合和集成学习的癫痫脑电自动识别方法研究", 《中国优秀硕士学位论文全文数据库》, 31 December 2020 (2020-12-31), pages 1 - 58 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115299946B (en) | 2024-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021068454A1 (en) | Method for identifying energy of micro-energy device on basis of bp neural network | |
| CN111460956B (en) | Unbalanced electrocardiogram sample classification method based on data enhancement and loss weighting | |
| CN113158793B (en) | Multi-class motor imagery electroencephalogram signal identification method based on multi-feature fusion | |
| CN110796120A (en) | Time domain feature-based circuit breaker mechanical fault XGboost diagnosis method | |
| CN114224360B (en) | EEG signal processing method, equipment and storage medium based on improved EMD-ICA | |
| Bing et al. | Electrocardiogram classification using TSST-based spectrogram and ConViT | |
| Xu et al. | A unified neural network for object detection, multiple object tracking and vehicle re-identification | |
| Tehrani et al. | Frequency-based multi task learning with attention mechanism for fault detection in power systems | |
| CN114795248A (en) | Electroencephalogram signal quality evaluation method and device, electronic equipment and medium | |
| CN112562727B (en) | Audio scene classification method, device and equipment applied to audio monitoring | |
| CN115299946B (en) | Self-adaptive input signal channel screening circuit introducing detection result feedback | |
| Semmad et al. | Scalable serial hardware architecture of multilayer perceptron neural network for automatic wheezing detection | |
| CN109934248B (en) | Multi-model random generation and dynamic self-adaptive combination method for transfer learning | |
| CN112205990A (en) | Wrist angle prediction method and device based on sEMG under different loads | |
| CN116725486A (en) | Dynamic electrocardiosignal atrial fibrillation detection method and device based on double phases | |
| CN111064617A (en) | Network flow prediction method and device based on empirical mode decomposition clustering | |
| CN106709934A (en) | Frequency domain Gaussian kernel function image tracking method | |
| CN115590530A (en) | Cross-object target domain agent subdomain adaptation method, system and medium | |
| KR102290243B1 (en) | Classification method for Electroencephalography motor imagery | |
| Xue et al. | MLResNet: An Efficient Method for Automatic Modulation Classification Based on Residual Neural Network | |
| CN114366116A (en) | Parameter acquisition method based on Mask R-CNN network and electrocardiogram | |
| CN114519305A (en) | Method for predicting residual life of circuit breaker based on stage attention mechanism network model | |
| CN110448273B (en) | Low-power-consumption epilepsy prediction circuit based on support vector machine | |
| KR20230096392A (en) | Method and apparatus for classifying heartbeat, and method for learning heartbeat classification model | |
| Mohseni et al. | Heart arrhythmias classification via a sequential classifier using neural network, principal component analysis and heart rate variation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |