Medical behavior fine-grained identification device and computer-readable storage medium

Abstract

The invention relates to a medical behavior fine-grained identification device, comprising a camera, a processor, a memory and the following algorithms: collecting real-time medical behavior video through the camera; preprocessing the video data; obtaining the sampling experience of the convolution kernel at the previous moment Calculate the local coherence constraint between the image at the previous moment and the image at the current moment; calculate the displacement of the receptive field at the current moment; add the corresponding displacement to each sampling point to obtain the deformed convolution kernel; perform convolution at the current moment Calculate the difference of the receptive field over time and extract the motion feature information; combine the above motion features and appearance features to obtain the spatiotemporal features; input the spatiotemporal features into the three-dimensional convolutional neural network, through the fully connected layer Obtain action segmentation and prediction results; perform feature segmentation and action classification prediction on the output results of the three-dimensional convolutional neural network. The invention can improve the speed, efficiency and recognition accuracy of the system.

Description

technical field [0001] The present invention relates to a device for fine-grained identification of medical behavior and a computer-readable storage medium including a program for performing the fine-grained identification of medical behavior. Background technique [0002] Since 2012, deep learning has flourished in just a few years, achieving great success in several subfields of artificial intelligence such as computer vision, natural language processing, and human behavior recognition. Human behavior recognition refers to the type of behavior performed by humans in the data perceived and recognized by the computer according to the image or video data collected by the camera. 3DCNN methods based on three-dimensional convolution (such as C3D, I3D), network methods based on time series models (such as RNN, LSTM). [0003] Existing behavior recognition methods generally use manual observation and design, manually design feature operators that can represent actions, and then ...

AI Technical Summary

Problems solved by technology

[0006] Existing technologies cannot effectively perceive and identify fine-grained medical behaviors, so the compliance judgment of medical behaviors and the intelligent optimization of medical procedures lack the technical basis for behavior recognition

Images