A Study Of Freezing Of Gait Recognition In Patients With Parkinson’s Disease Based On Variational Mode Decomposition

Freezing of gait is one of the most common yet also the most dangerous dyskinesia among the patients of Parkinson’s disease.Recognition on freezing of gait,the foundation for medical evaluation,intervention,and non-medicine treatment,has increasing value both of research and practical value.Centered around the objectives of high precision and real-time efficiency,the paper starts by looking at aspects such as signal processing on limb acceleration,feature extraction and classifier design,and algorithm optimization,and then expands on the study of recognizing freezing of gait on patients with Parkinson’s disease.In order to improve the broad applicability of the recognition method in the process of analyzing the acceleration signals of different Parkinson’s patients’ limbs in multi-task scenarios,and thus improve the global recognition accuracy,this paper used Variational Model Decomposition(VMD)to perform adaptive Decomposition.the paper discusses the principle and derivation process of VMD.Concerning the situation that the acceleration signal of the freezing of gait limb is unclear when functioning as the actual signal composition,the Average Instantaneous Frequency Method is used to determine the decomposition level of the body acceleration signal of the freezing of gait,therefore signal decomposition is completed in this way.The paper also proposes to replace the existing methods that focus on the fixed frequency band division of the acceleration signal of the freezing of gait limbs with the VMD adaptive decomposition result.In order to quantitatively describe the properties of the body acceleration signals in Parkinson’s patients for the recognition of freezing of gait,the paper extracts the features of freezing of gait based on the adaptive decomposition of the signals.the paper puts forward feature design on the freezing of gait based on VMD from the perspectives of energy features and time-domain characteristics.Then,preliminary identification experiments are carried out around the selection of sensor placement position and the division of time window length,and then parameters are determined to complete feature extraction.In this paper,the overall design and optimization of the recognition algorithm is carried out with the goal of higher recognition accuracy and faster recognition speed.Firstly,based on Classification and Regression Tree(CART),the above 66 dimensional features were ranked in importance,and 22 features with high importance were selected to form the optimal feature subset.The feature calculation and storage overhead of the recognition algorithm are optimized without significantly reducing the recognition accuracy.In this paper,the combined method of data sampling and ensemble classifier is designed and tested for the unbalanced number of samples of freezing of gait and the limited performance of a single classifier.The experimental results show that the recognition algorithm based on RUSBoost and SMOTE-Adaboost is more effective for the recognition of freezing of gait.For freezing gait recognition problem,the problem of high rate of common mistakes in this paper designs the corresponding optimization goal,from the perspective of the super parameter adjustment algorithm,using Bayesian Optimization of recognition algorithm of CART base classifier model split and largest integrated classifier on the number of base classifiers to parameter optimization,quantitative description of the relations between the two kinds of super parameter and target function.The optimized recognition algorithm combined with the optimal special collection extracted based on the VMD decomposition results described above can better qualify for the task of freezing of gait classification and recognition,and finally obtains the freezing of gait recognition effect with the accuracy of 91.9%,sensitivity of 92.3% and specificity of 91.5%.