Computer-implemented method, electronic device, and non-transitory computer-readable storage medium for context-aware classification of physiological signal data
A computer-implemented method, an electronic device, and a non-transitory computer-readable storage medium for context-aware classification of physiological signal data. The method includes the following. A data string is obtained. The data string is fed into a first deep neural network to generate a first feature map. Multi-dimensional data is generated based on the data string. The multi-dimensional data is fed into a second deep neural network to generate a second feature map. At least the first feature map and the second feature map are fused into a specific feature vector. The specific feature vector is fed into a machine learning model. The machine learning model outputs an identification result corresponding to the data string in response to the specific feature vector.
1 . A computer-implemented method for context-aware classification of physiological signal data, comprising:
obtaining a first signal stream comprising one-dimensional physiological data;
processing the first signal stream using a first neural network, to generate a first feature map;
transforming the one-dimensional physiological data of the first signal stream into a higher-dimensional representation suitable for spatial processing;
processing the higher-dimensional representation using a second neural network;
projecting the first feature map to generate a reference feature vector aligned with the second feature map;
fusing the reference feature vector and the second feature map; and
classifying to generate a semantic output of a physiological condition.
2 . The method according to claim 1 , wherein the first neural network comprises a first convolutional neural network and a recurrent neural network, and the method comprises:
feeding the first signal stream into the first convolutional neural network, wherein the first convolutional neural network generates a first spatial feature vector in response to the first signal stream; and
feeding the first spatial feature vector into the recurrent neural network, wherein the recurrent neural network generates a first temporal feature vector as the first feature map in response to the first spatial feature vector.
3 . The method according to claim 1 , wherein the second neural network comprises a second convolutional neural network, and the method comprises:
feeding the multi-dimensional data into the second convolutional neural network, wherein the second convolutional neural network generates a second spatial feature vector as the second feature map in response to the multi-dimensional data.
4 . The method according to claim 1 , wherein fusing the reference feature vector and the second feature map comprises:
feeding the first feature map into a third convolutional neural network, wherein the third convolutional neural network generates a first reference feature vector in response to the first feature map;
stacking a plurality of the first reference feature vectors into a second reference feature vector based on a size of the second feature map, wherein the size of the second feature map is equal to a size of the second reference feature vector;
transforming the second reference feature vector into a third reference feature vector; and
generating the specific feature vector based on the second feature map and the third reference feature vector.
5 . The method according to claim 4 , wherein transforming the second reference feature vector into the third reference feature vector comprises:
inputting the second reference feature vector into a Sigmoid function, wherein the Sigmoid function generates the third reference feature vector in response to the second reference feature vector, and each element in the third reference feature vector is within a value range between 0 and 1.
6 . The method according to claim 4 , wherein generating the specific feature vector based on the second feature map and the third reference feature vector comprises:
applying an attention mechanism configured to weight features in the second feature map based on corresponding values in the third reference feature vector, and generating the specific feature vector based on the attention-weighted features.
7 . The method according to claim 1 , further comprising:
processing the first signal stream using a third neural network to generate a third feature map; and wherein fusing the reference feature vector and the second feature map comprises:
fusing the first feature map, the second feature map, and the third feature map into the specific feature vector.
8 . The method according to claim 7 , wherein the first neural network comprises a convolutional neural network, and the method comprises:
processing the first signal stream using the convolutional neural network to generate a first feature map.
9 . The method according to claim 7 , wherein the third neural network comprises a recurrent neural network, and the method comprises:
processing the first signal stream using the recurrent neural network to generate a third feature map.
10 . The method according to claim 7 , wherein fusing the first feature map, the second feature map, and the third feature map to generate the specific feature vector comprises:
processing the first feature map using a fourth convolutional neural network to generate a fourth reference feature vector;
stacking a plurality of the fourth reference feature vectors into a fifth reference feature vector based on a spatial dimension of the second feature map, wherein a size of the second feature map corresponds to a size of the fifth reference feature vector;
transforming the fifth reference feature vector into a sixth reference feature vector;
processing the third feature map using a fifth convolutional neural network to generate a seventh reference feature vector;
stacking a plurality of the seventh reference feature vectors into an eighth reference feature vector based on the spatial dimension of the second feature map, wherein a size of the second feature map corresponds to a size of the eighth reference feature vector;
transforming the eighth reference feature vector into a ninth reference feature vector; and
generating the specific feature vector based on the second feature map, the sixth reference feature vector, and the ninth reference feature vector.
11 . The method according to claim 10 , wherein transforming the fifth reference feature vector into the sixth reference feature vector comprises:
applying a Sigmoid function to the fifth reference feature vector, wherein the Sigmoid function outputs the sixth reference feature vector in response to the fifth reference feature vector, wherein each element in the sixth reference feature vector is between 0 and 1; and
wherein transforming the eighth reference feature vector into the ninth reference feature vector comprises:
applying the Sigmoid function to the eighth reference feature vector, wherein the Sigmoid function outputs the ninth reference feature vector in response to the eighth reference feature vector, wherein each element in the ninth reference feature vector is between 0 and 1.
12 . The method according to claim 10 , wherein generating the specific feature vector based on the second feature map, the sixth reference feature vector, and the ninth reference feature vector comprises:
applying an attention mechanism to compute the specific feature vector from the second feature map, the sixth reference feature vector, and the ninth reference feature vector.
13 . The method according to claim 1 , wherein the higher-dimensional representation comprises a waveform image generated based on the first signal stream.
14 . An electronic device for classifying physiological signal data with contextual awareness, comprising:
a memory storing executable instructions; and
a processor configured to execute the instructions to:
receive a one-dimensional physiological signal;
process the signal using a first neural network to obtain a first feature map;
transform the one-dimensional physiological data of the first signal stream into a higher-dimensional representation suitable for spatial processing;
process the higher-dimensional data using a second neural network to obtain a second feature map;
project the first feature map to generate a reference feature vector aligned with the second feature map;
fuse the reference feature vector and the second feature map; and
classify a result to output a semantic indicator of physiological condition.
15 . The electronic device according to claim 14 , wherein the first neural network comprises a first convolutional neural network and a recurrent neural network, and the processor is configured to:
feed the first signal stream into the first convolutional neural network, wherein the first convolutional neural network generates a first spatial feature vector in response to the first signal stream; and
feed the first spatial feature vector into the recurrent neural network, wherein the recurrent neural network generates a first temporal feature vector as the first feature map in response to the first spatial feature vector.
16 . The electronic device according to claim 14 , wherein the second neural network comprises a second convolutional neural network, and the processor is configured to:
feed the multi-dimensional data into the second convolutional neural network, wherein the second convolutional neural network generates a second spatial feature vector as the second feature map in response to the multi-dimensional data.
17 . The electronic device according to claim 14 , wherein the processor is configured to:
feed the first feature map into a third convolutional neural network, wherein the third convolutional neural network generates a first reference feature vector in response to the first feature map;
stack a plurality of the first reference feature vectors into a second reference feature vector based on a size of the second feature map, wherein the size of the second feature map is equal to a size of the second reference feature vector;
transform the second reference feature vector into a third reference feature vector; and
generate the specific feature vector based on the second feature map and the third reference feature vector.
18 . The electronic device according to claim 17 , wherein the processor is configured to:
input the second reference feature vector into a Sigmoid function, wherein the Sigmoid function generates the third reference feature vector in response to the second reference feature vector, and each element in the third reference feature vector is within a value range between 0 and 1.
19 . The electronic device according to claim 17 , wherein generating the specific feature vector based on the second feature map and the third reference feature vector comprises:
applying an attention mechanism configured to weight features in the second feature map based on corresponding values in the third reference feature vector, and generating the specific feature vector based on the attention-weighted features.
20 . The electronic device according to claim 14 , the processor is further configured to:
process the first signal stream using a third neural network to generate a third feature map; and wherein fusing the reference feature vector and the second feature map comprises:
fusing the first feature map, the second feature map, and the third feature map into the specific feature vector.
21 . The electronic device according to claim 20 , wherein the first neural network comprises a convolutional neural network, and the processor is configured to:
process the first signal stream using the convolutional neural network to generate a first feature map.
22 . The electronic device according to claim 20 , wherein the third neural network comprises a recurrent neural network, and the processor is configured to:
process the first signal stream using the recurrent neural network to generate a third feature map.
23 . The electronic device according to claim 20 , wherein the processor is configured to:
process the first feature map using a fourth convolutional neural network to generate a fourth reference feature vector;
stack a plurality of the fourth reference feature vectors into a fifth reference feature vector based on a spatial dimension of the second feature map, wherein a size of the second feature map corresponds to a size of the fifth reference feature vector;
transform the fifth reference feature vector into a sixth reference feature vector;
process the third feature map using a fifth convolutional neural network to generate a seventh reference feature vector;
stack a plurality of the seventh reference feature vectors into an eighth reference feature vector based on the spatial dimension of the second feature map, wherein a size of the second feature map corresponds to a size of the eighth reference feature vector;
transform the eighth reference feature vector into a ninth reference feature vector; and
generate the specific feature vector based on the second feature map, the sixth reference feature vector, and the ninth reference feature vector.
24 . The electronic device according to claim 23 , wherein the processor is configured to:
apply a Sigmoid function to the fifth reference feature vector, wherein the Sigmoid function outputs the sixth reference feature vector in response to the fifth reference feature vector, wherein each element in the sixth reference feature vector is between 0 and 1; and
wherein transforming the eighth reference feature vector into the ninth reference feature vector comprises:
applying the Sigmoid function to the eighth reference feature vector, wherein the Sigmoid function outputs the ninth reference feature vector in response to the eighth reference feature vector, wherein each element in the ninth reference feature vector is between 0 and 1.
25 . The electronic device according to claim 23 , wherein the processor is configured to:
apply an attention mechanism to compute the specific feature vector from the second feature map, the sixth reference feature vector, and the ninth reference feature vector.
26 . The electronic device according to claim 14 , wherein the higher-dimensional representation comprises a waveform image generated based on the first signal stream.
27 . A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause a computing system to perform the method of claim 1 .