EEG Feature Extraction And Classification Of Autism Based On Singular Spectrum Analysis Alogorithm

Autism spectrum disorder is a neurological disease that usually develops in the early stages of development.It often affects patients’ cognition,emotions,sports,and social interactions.The cause of autism is unknown and may include genetic and environmental factors,and other factors.However,the current diagnosis of autistic patients mainly depends on the evaluation of the scale and the observation of behavior,etc.It is subjective and lacks objective indicators..Therefore,it is very important to seek potential objective indicators for more accurate diagnosis of autism.EEG is a convenient and non-invasive technique for recording fluctuations in the scalp of the brain.It has a millisecond resolution and can reflect different states of the brain.It has been widely used in the study of various neurological diseases.This article uses the resting brain Electric technology researches children with autism.Based on the SSA algorithm to extract EEG features,it provides auxiliary diagnostic indicators for the assessment of brain development in children with autism.The study enrolled 80 children,including 40 children with autism and 40 normal children with age and gender matching.First,the obtained EEG signals are pre-processed for artifacts such as myoelectricity and respiration to obtain relatively clean EEG data.Then,the SEG algorithm is used to decompose and reconstruct the EEG signals,remove the noise components,and extract the desired rhythm.The frequency and time domain methods are used to extract EEG features and classify them.Then,based on the SSA algorithm,the individual peak frequency and rhythm are studied.The time-domain relative energy characteristics were extracted to explore individual differences between autistic children and normal children.And carry out correlation analysis and classification algorithm research.Finally,research attempts to combine EEG signals with deep learning.The EEG data processed by the SSA algorithm is divided;the EEG data is converted into a time-frequency map by time-frequency analysis to be suitable for deep learning networks,and then a convolutional neural network framework is constructed.The results show that the SSA algorithm can effectively remove low frequency artifact signals and extract EEG rhythms.And the relative energy of rhythm in the autism group was lower than that in the normal group.The classification accuracy of EEG features extracted by time domain method have higher classification accuracy,and the classification accuracy rate is 87.5%.The individualized peak frequency of children with autism was shifted to a lower frequency than the normal group,and its classification accuracy was 81.5%.The relative energy classification accuracy of individual rhythms is 80%.After the two are fused,the final classification accuracy rate is 91.25%.At the same time,the rhythmic time-domain waveform reconstructed based on the SSA algorithm can be applied to deep learning through timefrequency analysis.The accuracy rate is 83.28%,the recall rate is 99.38%,and the accuracy rate is 75.18%.An idea of transforming EEG data into image data based on SSA algorithm for deep learning model is proposed.Based on the SSA algorithm,this paper starts from the relative energy of the EEG and individualized EEG rhythm,verifies the effectiveness of the SSA algorithm and seeks potential objective physiological indicators,and provides an effective method for the auxiliary diagnosis of autism.And by constructing an EEG data format suitable for deep learning,it provides the possibility for using deep learning autism brain point features.