Cacosphyxia discrimination method and system based on fusion attention mechanism

Abstract

The invention provides a cacosphyxia discrimination method and system based on a fusion attention mechanism. The method comprises the steps: acquiring historical pulse wave data, and carrying out preprocessing on the historical pulse wave data so as to obtain a pulse wave image meeting a preset rule; according to the pulse wave image, establishing an attention module, and conducting feature extraction in a spatial dimension and a channel dimension through the attention module so as to obtain spatial feature data and dimension feature data respectively; performing adaptive feature learning and training on a preset CNN neural network model according to the spatial feature data and the dimension feature data to obtain a cacosphyxia discrimination neural network model; and collecting current pulse wave data, inputting the current pulse wave data serving as input signals input into the cacosphyxia judgment neural network model, and allowing the cacosphyxia judgment neural network model to output a detection result of the current pulse wave data so as to realize abnormality discrimination of the current pulse wave. According to the invention, the accuracy of pulse wave image discrimination is improved.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a pulse abnormality discrimination method and system based on a fusion attention mechanism. Background technique [0002] In intelligent health diagnosis, using pulse to detect health status is an important research field at present. Common application methods include expert interview method, scale evaluation method and automatic detection method. Automatic analysis and detection can be monitored and evaluated in real time through pulse wave diagrams. Compared with expert consultation and diagnosis, automatic analysis and detection is more convenient and faster. [0003] The existing automatic analysis of pulse wave has made some progress, such as extracting the peak and valley of the waveform through wavelet analysis, so as to analyze the pulse wave frequency; and using the Gaussian mixture model to extract pulse wave features for feature analysis. With the development...

AI Technical Summary

Problems solved by technology

[0004] However, the pulse wave feature analysis method based on the Gaussian model and wavelet analysis can only do basic preprocessing on the pulse wave data, and cannot reach the judgment of whether it is healthy or not

However, the two-channel convolutional neural network method can better perform qualitative analysis on features. However, separating the frequency domain from the time domain will cause the correlation between the two to be ignored. On the other hand, the traditional convolutional neural network Each layer of the network will obtain information from its leading layer, and then convert it into useful feature information. Different types of network layers are responsible for extracting different types of feature information, and the output of convolution kernels of different sizes is also different. It is difficult to know which transformation can provide the most useful information

Images