Research On Classification Method Of Nonlinear Medical Sensor Data Based On Continuous Learnin

The constructive development of the Internet of Things and sensor technology has promoted the intelligence,automation,and informatization of the medical Internet of Things.The future trend will be the deep integration of artificial intelligence technology with the medical Internet of Things.Currently,the limitations of applying artificial intelligence technology to the medical Internet of Things are the weak generalization and robustness of the models.Existing deep learning models cannot adapt to multiple tasks and are prone to catastrophic forgetting.In addition,the data distribution of medical sensor data can be disrupted by environmental,seasonal,and equipment-related factors,leading to the degradation of the models.To address these issues,this paper focuses on researching the classification method of non-linear medical sensor data based on continuous learning.The specific research content of this paper is as follows:(1)A non-linear medical sensor data classification method based on continual learning is proposed and named SCALT.SCALT is a transferable and scalable classification model that can be easily extended or transferred to learn new classes or tasks.In order to achieve the scalability of the class,the scalable micro-classifier mechanism Per-Class is added to SCALT,which can quickly adapt to the emergence of new class data and also reduce the confusion with other class features during classification.In order to realize the continual learning ability of the model,the parameter protection mechanism is introduced into SCALT,which can protect the knowledge of previous tasks,quickly adapt to new tasks,and ensure the stability of the model in all tasks.In addition,this paper verifies the generalization and plasticity of SCALT based on different nonlinear medical sensor datasets.(2)A continual learning-based non-linear medical sensor data classification method is proposed and named Meta CL.Some medical data are scarce and very expensive,and the dynamic distribution of data will lead to changes in the characteristics and statistical attributes of the data,resulting in the previous model no longer being used.To better adapt deep learning models to the dynamic distribution of medical sensor data,Meta CL uses a mask function and a knowledge base based on meta-learning updates.This approach balances new and old knowledge and effectively utilizes a small amount of labeled data,enabling Meta CL to achieve continual learning capabilities and improve model robustness.In addition,this paper verifies the robustness and plasticity of Meta CL based on different nonlinear medical sensor datasets.